JPH10176578A - Air-fuel ratio control device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To attain proper air-fuel ratio control and suppress the occurrence of a misfire by detecting multiple physical quantities detectable at a low temperature and indicating the state of an engine, inputting multiple physical quantities as parameters, estimating an air-fuel ratio with a neural network, and correcting the fuel injection quantity. SOLUTION: The outputs of a throttle opening sensor (l), an intake air pressure sensor (m), a crank angle sensor (p), and an air-fuel ratio sensor (q) are inputted to an ECU, and the ECU calculates the fuel injection quantity Gf to keep an air-fuel ratio A/F at the prescribed value and controls an injector I. The ECU uses the physical quantities detected by the sensors as parameters and estimates the air-fuel ratio A/F with a neural network. The ECU detects the deviation between the actually collected air-fuel ratio A/F and the estimated airfuel ratio A/F, it changes the structure of the neural network so that this deviation becomes within an allowable value, e.g. average 0.1 or below as air- fuel ratio conversion, and it calculates the correction quantity of the fuel injection quantity based on the estimated air-fuel ratio A/F.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a control device for a fuel injection control system of an internal combustion engine, and more particularly to a control device for assisting control of an air-fuel ratio of an engine using a neural network.

[0002]

2. Description of the Related Art In recent years, with the increase in the number of automobiles, exhaust gas pollution has become a serious problem. In order to cope with such exhaust gas pollution, a method of reducing NOx, CO, and HC, which are toxic gases generally contained in exhaust gas of automobiles, using a catalyst is adopted. Is used. Such a catalyst is
It can act on multiple types of harmful gases such as Ox, CO and HC. However, when the air-fuel ratio is lean, the NOx cannot be purified, and the air-fuel ratio becomes rich. In the (dense) state, CO and HC cannot be purified. That is, in order to most effectively purify these harmful gases with a catalyst, the air-fuel ratio needs to be maintained at a constant value at which the catalyst can work effectively.

For this reason, air-fuel ratio control for maintaining the air-fuel ratio constant irrespective of the driving state of the automobile is required. In the air-fuel ratio control, feed-forward control is performed for performing fuel increase correction, fuel decrease correction, and the like in accordance with a change in the throttle opening and the like. In addition, feedback control for feeding back the correction amount of the fuel injection amount using the value of an air-fuel ratio sensor 02 sensor or a linear air-fuel ratio sensor (hereinafter referred to as a "LAF sensor") is also generally performed. These controls are particularly effective in a steady operation range such as when idling or running at a constant speed.

However, in an actual engine, even when fuel is injected from an injector, for example, not all of the fuel flows into the cylinder, but a part of the fuel has a complicated behavior such that the fuel adheres to the intake pipe wall surface. The amount of adhesion varies in a complicated manner depending on the operating state (rotational speed, load (intake air pressure), etc.) and the external environment (intake air temperature, cooling water temperature, atmospheric pressure, etc.), and evaporates from the attached fuel and flows into the cylinder. The fuel amount also changes due to various factors, such as the operating state and the external environment. For this reason, it is actually very difficult to control the air-fuel ratio in a transient state such as acceleration or deceleration only by simple feedforward control or feedback control.

Therefore, in order to improve the accuracy of these controls, for example, as disclosed in Japanese Patent Application Laid-Open No. 3-235723, a nonlinear element such as the above-mentioned fuel adhesion is learned by a neural network, and this neural network is used. There has been proposed a control device that controls the correction amount of the fuel injection amount so as to improve the response performance in a transient state.

FIG. 50 shows a general configuration of an air-fuel ratio control device using a neural network. The air-fuel ratio control device shown in FIG. 50 will be briefly described. The air-fuel ratio is maintained at a predetermined value by feed-forward control and feedback control of a control device (not shown).
It is assumed that an air-fuel ratio control device shown in FIG. 50 is provided to appropriately maintain the air-fuel ratio at the time of transition.

A conventional air-fuel ratio control device using a neural network provides a plurality of parameters representing the state of the engine, such as the engine speed (Ne), to the engine.
Intake air pressure (Pb), throttle opening (THL), fuel injection amount (Gf), intake air temperature (Ta), cooling water temperature (T
w), a state detector 210 for detecting an air-fuel ratio (A / F), etc.
Is provided. A plurality of parameters detected by the state detection unit 210 are input, and the correction amount (ΔGf) of the fuel injection amount is calculated by the fuel correction amount estimation unit 220 by a neural network learned using the correction amount of the fuel injection amount as an output. presume. This fuel correction amount estimating unit 22
The fuel injection amount is corrected by adding the correction amount (ΔGf) estimated by 0 to the fuel injection amount (Gf) calculated by a control device (not shown), and assists in controlling the air-fuel ratio (A / F). This makes it possible to control the air-fuel ratio appropriately even for complicated engine behavior in a transient state.

[0008]

However, in the air-fuel ratio control system using the conventionally proposed neural network, the input term of the neural network includes the air-fuel ratio (A /
Since the sensor output of F) is included in the input items, the air-fuel ratio sensor becomes inactive at extremely low temperatures or when the engine is started, and cannot be used. There is a problem that can not be.

[0009] Even after the engine has warmed up, the deterioration of the air-fuel ratio sensor increases as the age of use of the air-fuel ratio sensor increases, and the output value of the neural network shifts.
Failure occurs as a whole in the air-fuel ratio control system. Further, the malfunction of the air-fuel ratio control system is also caused by abnormality or deterioration of some elements constituting the air-fuel ratio control system.

Further, if the correction amount of the fuel injection amount is directly calculated by the neural network,
It is difficult to set teacher data and measure the correction amount of fuel injection amount, etc.If accuracy is not obtained at the time of learning, it is difficult to grasp where there is a problem, and it takes a lot of time and effort in the development process Is expected. Further, when starting the engine, even if the fuel injection amount is excessively injected, or if the injected amount is small, a misfire will occur. In both cases, the situation of misfiring is the same, so the output value of the air-fuel ratio sensor is extremely lean (thin). However, in the conventional air-fuel ratio control system using a neural network, learning is performed in consideration of the difference in the cause of the misfire. Since there is no control, inconvenience occurs during control, misfire cannot be prevented, and harmful unburned fuel may be discharged in many cases.

If the misfire state can be known, the fuel injection amount can be adjusted so as to prevent misfire, and the engine can be started smoothly. It is necessary to provide an internal pressure sensor or the like for each cylinder, which increases costs and is less feasible. On the other hand, a user of a car selects and uses one of a plurality of gasolines on the market, but the properties of the selected gasoline, such as the evaporation rate, are different from each other. It is considered that if the air-fuel ratio control taking into account the difference in the air-fuel ratio is performed, the air-fuel ratio control can be appropriately performed especially at a low temperature where the difference in the evaporation rates is large. However, since the car user does not actually know which gasoline to choose, the manufacturer must prevent the engine from misfiring during cold weather.
At the time of starting, the fuel injection amount is uniformly adjusted according to the properties of gasoline having the lowest evaporation rate. For this reason, since the user of an automobile usually selects gasoline with a higher evaporation rate, there is a problem that the fuel injection amount becomes too large and harmful unburned fuel is released.

Further, even after the engine is started, the fuel injection amount is calculated by feedforward control while fixing the evaporation rate of gasoline to a specific type of gasoline, so that the air-fuel ratio cannot be properly controlled if the type of gasoline is different. In particular, since the air-fuel ratio sensor does not operate in a low temperature range, it cannot be compensated for by feedback control or air-fuel ratio control using a conventional neural network.

It is an object of the present invention to provide a control device capable of appropriately assisting the air-fuel ratio control even at an extremely low temperature or when the engine is started, without having to consider the deterioration of the air-fuel ratio sensor. Aim. Further, the present invention is intended to easily detect any abnormality or deterioration of the components of the air-fuel ratio control system and to perform appropriate air-fuel ratio control in consideration of the abnormality or deterioration as necessary. Aim.

Further, the present invention estimates a physical quantity for calculating the correction amount of the fuel injection amount, which is easy to set and measure the teacher data, by using a neural network, thereby facilitating learning and reducing time in the development process. And aims to reduce the effort. Further, according to the present invention, by estimating the type of fuel by using a neural network and appropriately setting the fuel injection amount at the time of low temperature of the engine, particularly at the time of starting the engine, the misfire is eliminated and the emission of unburned fuel is prevented. The purpose is to:

The present invention achieves more appropriate air-fuel ratio control by using a neural network that takes into account a misfire state, and further controls the fuel injection amount by the neural network so that misfire does not occur. Aim.

[0016]

In order to solve the above-mentioned problems, the present invention provides an air-fuel ratio control which assists the control of the air-fuel ratio by correcting the fuel injection amount to a control system for maintaining the air-fuel ratio at a constant value. In the apparatus, state detection means for detecting a plurality of physical quantities that can be detected even at a low temperature indicating the state of the engine, and air-fuel ratio estimation for estimating the air-fuel ratio using a neural network by inputting the detected plurality of physical quantities as parameters Means and a fuel correction amount calculating means for calculating a correction amount of the fuel injection amount from the estimated air-fuel ratio. Here, the low temperature refers to a temperature below the temperature at which the air-fuel ratio sensor does not operate, for example, a temperature of about 50 ° C. or less based on the existing air-fuel ratio sensor. It is sufficient that these sensors operate at 0 ° C. or higher. The calculation of the correction amount of the fuel injection amount includes calculation of a correction coefficient for calculating the fuel injection amount including the correction amount by multiplying the fuel injection amount.

The air-fuel ratio control device further includes fuel time-series data storage means for storing time-series data of the fuel injection amount. It is desirable to include A parameter range determining unit that determines whether a value of at least one parameter input to the air-fuel ratio estimating unit is within a range set in advance for the parameter; It is effective to provide parameter conversion means for replacing the value of the parameter determined not to be within the range with a preset value. Here, it is effective that the preset range for the parameter is determined based on the maximum value and the minimum value of the value of the physical quantity serving as the parameter, which are input in the learning of the neural network. .

Further, in the above-described air-fuel ratio control apparatus, a transient state detecting means for detecting a transient state amount of the engine, and a fuel injection amount obtained by the fuel correction amount calculating means based on the detected transient state amount. Fuel correction amount adjusting means for adjusting the amount of correction may be provided. The transient state detecting means detects the transient state quantity based on the change amount of at least one physical quantity obtained by the state detecting means, or detects the change amount of the air-fuel ratio estimated by the air-fuel ratio estimating means. It is also possible to detect a transient state quantity based on an air-fuel ratio sensor that detects an air-fuel ratio, and a transient state quantity that detects a transient state quantity based on a change amount of an output value of the air-fuel ratio sensor. It may be constituted by state calculation means.

Then, the neural network used in the air-fuel ratio estimating means, when the output of the air-fuel ratio sensor is in a super-lean state in the learning process and is not consistent with the fuel injection amount, the air-fuel ratio is in a super-rich state. It is desirable that learning be performed using teacher data including information that exists. Further, in the air-fuel ratio control device, a fuel type detection unit that detects a type of fuel used in the engine is provided, and a parameter input by the air-fuel ratio estimation unit corresponds to a property of the detected fuel type. It may include a numerical value.

Further, in order to solve the above-mentioned problems, the present invention provides an air-fuel ratio control device for assisting control of the air-fuel ratio by correcting the fuel injection amount to a control system for maintaining the air-fuel ratio at a constant value. State detection means for detecting a plurality of physical quantities representing the state of the engine including the injection amount and the air-fuel ratio; and inputting at least two or more of the detected plurality of physical quantities as parameters, and using a neural network to calculate the weight of the air flowing into the cylinder using a neural network. A fuel for calculating a correction amount of the fuel injection amount from the estimated inflow air weight estimation means, the fuel injection amount and the air-fuel ratio detected by the state detection means, and the cylinder inflow air weight estimated by the inflow air weight estimation means. And a correction amount calculating means.

According to another aspect of the present invention, there is provided an air-fuel ratio control apparatus for assisting control of an air-fuel ratio by correcting a fuel injection amount to a control system for maintaining the air-fuel ratio at a constant value. A state detecting means for detecting a plurality of physical quantities that can be detected even at a low temperature indicating a state, and a change for inputting a plurality of detected physical quantities as parameters and estimating a change amount of a physical quantity related to an air-fuel ratio using a neural network. There is provided an amount estimating means and a fuel correction amount calculating means for calculating a correction amount of the fuel injection amount from a change amount of the physical quantity related to the estimated air-fuel ratio. Note that the physical quantity related to the air-fuel ratio refers to a physical quantity that affects the increase / decrease of the value of the air-fuel ratio. The change amount includes a change rate and a change speed.

The change amount estimating means inputs a plurality of detected physical amounts as parameters, estimates a physical amount related to an air-fuel ratio using a neural network, and calculates a change amount of the physical amount to obtain an air-fuel ratio. It is preferable that the amount of change of the related parameter is estimated, and the physical quantity related to the air-fuel ratio is preferably the air-fuel ratio itself.

Further, the neural network used in the change amount estimating means is also characterized in that the air-fuel ratio is in a super-rich state when the output of the air-fuel ratio sensor is in a super-lean state during the learning process and is not consistent with the fuel injection amount. It is effective to use what is learned using the teacher data including the information that In order to solve the above-mentioned problems, the present invention provides an air-fuel ratio control system comprising a control system for maintaining the air-fuel ratio at a constant value, and a control system for assisting the air-fuel ratio control by correcting the fuel injection amount to the control system. In the abnormality / deterioration detection device for detecting abnormality or deterioration of the fuel ratio control system, state detection means for detecting a plurality of physical quantities that can be detected even at a low temperature indicating the state of the engine, and inputting the detected plurality of physical quantities as parameters. An air-fuel ratio estimating means for estimating an air-fuel ratio using a neural network, an air-fuel ratio sensor detecting an air-fuel ratio, an air-fuel ratio estimated by the air-fuel ratio estimating means, and an air-fuel ratio detected by the air-fuel ratio sensor. And a sensor characteristic judging means for detecting a change in dynamic characteristics of the air-fuel ratio sensor by comparing

The sensor characteristic determining means obtains respective extreme values by differentiating the time change of the air-fuel ratio estimated by the air-fuel ratio estimating means and the time change of the air-fuel ratio detected by the air-fuel ratio sensor. By comparing the time and the value of each of the extreme values, it is possible to obtain a phase lag and a gain change of the air-fuel ratio sensor as a dynamic characteristic change of the air-fuel ratio sensor. The detected at least one physical quantity, the air-fuel ratio estimated by the air-fuel ratio estimating means, and the air-fuel ratio detected by the air-fuel ratio sensor are input as parameters, and the dynamic characteristic change of the air-fuel ratio sensor is determined using a neural network. It can also be configured to estimate.

In order to solve the above-mentioned problem, in the abnormality / deterioration detecting device, the air-fuel ratio estimating means includes:
Causing the air-fuel ratio before the catalyst in the exhaust pipe to be estimated; and causing the air-fuel ratio sensor to detect the air-fuel ratio after the catalyst in the exhaust pipe; Alternatively, a configuration may be adopted in which the air-fuel ratio detected by the air-fuel ratio sensor is compared with a catalyst deterioration detection unit that detects catalyst deterioration by comparing the air-fuel ratio.

Then, in order to solve the above-mentioned problems, an air-fuel ratio control device which assists the control of the air-fuel ratio by correcting the fuel injection amount to a control system for maintaining the air-fuel ratio at a constant value indicates the state of the engine. State detection means for detecting a plurality of physical quantities that can be detected even at a low temperature, an air-fuel ratio sensor for detecting an air-fuel ratio, and a physical quantity representing the state of the engine. Operating state determining means for determining whether or not there is, when at least one physical quantity detected by the state detecting means is determined when the engine is in a predetermined operating state, and changing the physical quantity from the change time of the physical quantity; Response time detecting means for measuring and storing a response time until the air-fuel ratio detected by the air-fuel ratio sensor fluctuates; a plurality of detected physical quantities; An air-fuel ratio estimating unit that inputs time as a parameter and estimates an air-fuel ratio using a neural network, and a fuel correction amount calculating unit that calculates a correction amount of a fuel injection amount from the estimated air-fuel ratio is provided. .

The neural network used in the air-fuel ratio estimating means performs neuro-learning on learning data obtained using the deteriorated sensor using the output value of the air-fuel ratio sensor without deterioration as a teacher signal. It is better to have done. Furthermore, in order to solve the above-mentioned problem, the present invention provides an air-fuel ratio control device that assists the control of the air-fuel ratio by correcting the fuel injection amount to a control system that maintains the air-fuel ratio at a constant value. State detection means for detecting a plurality of physical quantities, a plurality of detected physical quantities are input as parameters, a fuel type determination means for determining the type of fuel using a neural network, and a physical quantity representing a state of the engine, Operating state determining means for determining whether the engine is in a predetermined state based on the physical quantity; and, when the engine is in a predetermined state,
Fuel correction amount calculating means for calculating a correction amount of the fuel injection amount based on the fuel type determined by the fuel type determining means.

In this case, the state detecting means includes:
As one of the physical quantities representing the state of the engine, the number of cranks from energization of the spark plug to the complete explosion of the air-fuel mixture in the engine is detected. It is possible to detect the energization of the plug and determine whether or not the engine is in a starting state based on the physical quantity. Also,
It is preferable that the state detecting means detects a battery voltage of the engine as the physical quantity representing the state of the engine.

An engine temperature detecting section for detecting an engine temperature as the physical quantity representing the engine state; and determining whether or not the engine temperature is equal to or lower than a first set temperature. And a fuel correction amount calculating unit that calculates the correction amount of the fuel injection amount in consideration of the engine temperature detected by the engine temperature detecting unit. .

Further, it is desirable that the air-fuel ratio control device is provided with a discrimination permitting unit that permits discrimination by the fuel type discriminating means only when the engine temperature is equal to or lower than a predetermined second set temperature. The air-fuel ratio control device is provided with a fuel type storing means for storing the fuel type determined by the fuel type determining means each time the fuel type is updated, and the fuel correction amount calculating means stores the fuel type at the time of starting the engine. It is effective to calculate the correction amount of the fuel injection amount based on the fuel type stored in the storage means.

Then, the state detecting means detects at least the fuel injection amount and the air-fuel ratio as physical quantities, and the fuel type judging means inputs a plurality of detected physical quantities as parameters, and calculates the weight of the air flowing into the cylinder by the neural network. An inflow air weight estimating unit for estimating by a network;
A fuel characteristic calculating unit that calculates a fuel evaporation rate and / or a fuel adhesion rate based on the fuel injection amount and the air-fuel ratio detected by the state detection unit and the cylinder inflow air weight estimated by the inflow air weight estimation unit, or both; And a type discriminating unit for discriminating the type of fuel from the fuel evaporation rate and / or the fuel adhesion rate calculated by the fuel calculating unit.

In this case, the fuel characteristic calculation unit is configured to calculate both the fuel evaporation rate and the fuel adhesion rate,
Further, a calculated value storage unit for sequentially storing the calculated fuel evaporation rate and fuel adhesion rate is provided, and the type discrimination unit includes a plurality of fuel evaporation rates and fuel adhesion stored in the calculated value storage unit. It is desirable to determine the type of fuel from the rate.

Further, the fuel type determining means uses a neural network learned by inputting and learning, as a parameter, a physical quantity detected by the state detecting section when the engine is operated using a reference type of fuel. ,
An air-fuel ratio estimator for estimating an air-fuel ratio by inputting a physical quantity detected by the state detector as a parameter, an air-fuel ratio sensor for detecting an air-fuel ratio, an air-fuel ratio estimated by the air-fuel ratio estimator, A fuel type estimating unit for estimating the fuel type using the air-fuel ratio detected by the fuel ratio sensor.

According to another aspect of the present invention, there is provided a vibration correction control device for correcting a vibration component in a control system in which an input parameter is a factor for adding a vibration component to a control target. A convergence degree estimating means for inputting a physical quantity related to the control object as a parameter and estimating the degree of convergence of the control object by a neural network; and And a correction amount calculating means for calculating the correction amount.

This vibration correction control device is an air-fuel ratio control device which assists the control of the air-fuel ratio by correcting the fuel injection amount to a control system for maintaining the air-fuel ratio at a constant value.
State detecting means for detecting a plurality of physical quantities representing the state of the engine, misfiring degree estimating means for estimating the degree of misfiring of the engine by inputting the detected plurality of physical quantities as parameters and using a neural network, and misfiring degree estimating means And a fuel correction amount calculating means for calculating a correction amount of the fuel injection amount from the degree of misfire obtained in step (1).

In this air-fuel ratio control device,
When at least one of the plurality of physical quantities detected by the state detecting means determines that the engine is in the starting state, and when the starting state determining means determines that the engine is in the starting state, It is preferable to provide the fuel correction amount calculation means only with the fuel correction amount calculation permission means for permitting the calculation of the correction amount of the fuel injection amount.

Further, the neural network used in the misfire degree estimating means determines the misfire degree when the rate of change of the output of the air-fuel ratio sensor is negative during the learning process from the start of the engine to the steady state. It is desirable that the learning is performed using teacher data including information to be set to 0.

[0038]

Embodiments of the present invention will be described below with reference to the drawings. (Embodiment 1) FIG. 1 is a schematic diagram schematically showing the configuration of a control device including an air-fuel ratio control device according to the present invention. FIG.
In the control device shown in FIG. 1, the output of the throttle opening sensor l, the intake air pressure sensor m, the intake air temperature sensor n, the cooling water temperature sensor o, the crank angle sensor p, the air-fuel ratio sensor q, and, if necessary, the output of the battery voltmeter v The input is input to the control unit C. From this input value, the engine control unit C calculates a fuel injection amount (Gf) that keeps the air-fuel ratio (A / F) at a predetermined value and outputs it to the injector I. Then, the injector I injects fuel into the engine by the calculated fuel injection amount (Gf). However, the output of the air-fuel ratio sensor q is not used in a cold state.

FIG. 2 shows a hardware configuration diagram of the engine control unit C. As shown in FIG. 2, the engine control unit C includes a CPU 20 for performing arithmetic processing.
1. ROM 202 storing control programs, various maps, weight values of neural networks, threshold values, transfer functions, etc., RA used for storing numerical values during arithmetic processing, etc.
It comprises M203, A / D converter 3, and D / A converter 204.

FIG. 3 shows a functional block diagram of the engine control unit C realized by using this hardware configuration. As shown in the figure, the engine control unit C mainly includes a basic fuel injection amount calculation unit 1 and an air-fuel ratio auxiliary control calculation unit 2. The basic fuel injection amount calculation unit 1 performs feedforward control using a map from each input value converted into a digital amount from the various sensors via the A / D conversion unit 3 and in a temperature range where the air-fuel ratio sensor q can be used. Is a part for calculating the basic fuel injection amount (Gfb) by a known method in consideration of feedback control. In normal steady state, the value of the air-fuel ratio (A / F) is kept at a predetermined constant value only by the basic fuel injection amount calculation unit 1.

The air-fuel ratio auxiliary control calculation unit 2 calculates the basic fuel injection amount (Gf) calculated by the basic fuel injection amount calculation unit 1.
b) is a part for calculating a correction amount (ΔGf) at the time of transition or the like, and includes an input value from various sensors and a basic fuel injection amount (Gfb) calculated by the basic fuel injection amount calculation unit 1 in the previous control cycle. The correction amount (Gf) of the fuel injection amount (Gf) is calculated using the neural network from the actual fuel injection amount (Gf) in the previous control cycle, which is the sum of the correction amount (ΔGf) calculated by the air-fuel ratio auxiliary control calculation unit 2. ΔGf) is calculated. The air-fuel ratio auxiliary control calculation unit 2 constitutes a main part of the air-fuel ratio control device according to the present invention.

Although the correction amount (ΔGf) of the fuel injection amount is always calculated using the neural network without inputting the air-fuel ratio sensor q, the air-fuel ratio is calculated at a low temperature when the air-fuel ratio sensor does not work. The input of the sensor q is not used, and the air-fuel ratio sensor q
May be used, and the configuration of the neural network may be switched accordingly.

To briefly explain the overall control operation having the above configuration, first, in the driving state of the automobile,
In the region where the temperature is high, the sensors q further detect the state quantities of the engine E, respectively. Is input to the basic injection fuel (Gf
b) is calculated. On the other hand, time-series data of the actual fuel correction amount (Gf) calculated in the previous control cycle is stored in the time-series data storage unit 4 configured in the RAM. The stored fuel correction amount (Gf) and the various engine state quantities converted by the A / D conversion unit 3 are input to the air-fuel ratio auxiliary control calculation unit 2, and the correction amount (ΔGf) of the fuel injection amount is calculated. Then, the calculated basic fuel injection amount (Gfb) and correction amount (ΔG
f) are added, the actual fuel injection amount (Gf) is calculated, and the calculated amount of fuel is injected from the injector I of the engine E. Here, only the time-series data of the fuel injection amount (Gf) is stored in the time-series data storage unit 4, but the time-series data of the detection values of various sensors may also be stored.

FIG. 4 is a functional block diagram showing the configuration of the air-fuel ratio control device according to the first embodiment. This air-fuel ratio control device includes a state detection unit 10, an air-fuel ratio estimation unit 20, and a fuel correction amount calculation unit 30. In the configuration of FIG. 3, the air-fuel ratio assisting control calculation unit 2 forms an air-fuel ratio estimation unit 20 and a fuel correction amount calculation unit 30. Is configured. In addition,
The air-fuel ratio control device according to each of the following embodiments is also a part of the control device shown in FIGS.

The state detecting section 10 is a part for detecting a plurality of physical quantities which indicate the state of the engine E and can be detected even at a low temperature. Here, the engine speed (Ne), intake air pressure (Pb),
The throttle opening (THL), the fuel injection amount (Gf), the intake air temperature (Ta), and the cooling water temperature (Tw) are to be detected. And a RAM 203 that stores the fuel injection amount in the control cycle before the last time, and the like. It should be noted that the plurality of physical quantities are not limited to these, and it is obvious that more physical quantities may be detected.

The air-fuel ratio estimating section 20 is a section for inputting the physical quantity detected by the state detecting section 10 as a parameter and estimating the air-fuel ratio using a neural network described in detail below. The learning process of the neural network used in the air-fuel ratio estimating unit 20 will be described with reference to FIG. FIG. 5 shows the engine speed (Ne) and the intake air pressure (P
b), throttle opening (THL), fuel injection amount (G
FIG. 5F is a schematic diagram illustrating a learning process of a neural network for estimating an air-fuel ratio (A / F) using input air temperature (Ta) and cooling water temperature (Tw) as input terms. An air-fuel ratio sensor 10a that detects an air-fuel ratio (A / F) immediately after an exhaust valve of each cylinder or at an exhaust pipe collecting part in a vehicle for learning data collection
(LAF sensor), and the same state detection unit 10 as in FIG.
Is provided. Then, the user actually drives the car and collects learning data. At this time, the air-fuel ratio sensor 10a is heated first so that the air-fuel ratio can be detected even when the engine itself is in a low temperature state. The collected air-fuel ratio (A / F) and each parameter are input to a neural network, and the estimated air-fuel ratio (A / FNN) is output.
And the configuration of the neural network is changed so that the deviation e falls within a certain allowable value, for example, an average of 0.1 or less in air-fuel ratio conversion. In this case, appropriate feedforward control can be performed by using teacher data that is temporally slower than the input data in consideration of the detection delay of the air-fuel ratio sensor 10a and the like. Also, past data of each parameter may be added to the input term. In this case, the time series data of the added parameter may be stored in the time series data storage unit 4.

To change the configuration of the neural network due to the deviation e, each input parameter is converted by a transfer function using a back propagation method, the weight value multiplied by this is multiplied, and the threshold value in each processing unit. By a predetermined method according to the deviation e. Here, the tangent sigmoid function (f (x) = tanh
(X)) is used. In addition, various types of neural networks can be adopted as long as a value such as an air-fuel ratio can be obtained from each parameter. Further, the same method is employed in learning of a neural network described in the following embodiments. As a result of such learning, the neural network obtains the air-fuel ratio (A / F) in a transient state in which the throttle changes with the above parameters as input.
Can be predicted and estimated.

The fuel correction amount calculation unit 30 calculates a correction amount (ΔGf) of the fuel injection amount from the air-fuel ratio (A / F) estimated by the air-fuel ratio estimation unit 20 having the neural network learned as described above. I do. Specifically, the air-fuel ratio (A
/ F) and the correction amount (ΔGf) of the fuel injection amount are approximated to a P (proportional) relationship, a PI (proportional, differential) relationship, and the like.

[0049]

(Equation 1)

Or

[0051]

(Equation 2)

The calculation is performed according to the equation given by setting the above.
K1 and K2 are constants derived from experiments and the like. The operation of the air-fuel ratio control device having the above configuration will be described below. Now, it is assumed that the car is running.
First, the state detection unit 10 detects an engine speed (Ne), an intake air pressure (Pb), a throttle opening (THL), a fuel injection amount (Gf), an intake air temperature (Ta), and a cooling water temperature (Tw). Next, the air-fuel ratio estimating unit 20 estimates and outputs an air-fuel ratio (A / F) using a neural network using the detected physical quantities as input parameters. Using the estimated air-fuel ratio (A / F), the fuel correction amount calculation unit 30 calculates a correction amount (ΔGf) of the fuel injection amount. The actual fuel injection amount (Gf) is calculated by adding the calculated correction amount (ΔGf) of the fuel injection amount and (Gfb) calculated by the basic fuel injection amount calculation unit 1 described above, and the injector I, and this amount of fuel is injected into the engine E.

FIG. 6A shows a control result by feedback control using an O2 sensor when the throttle opening is rapidly opened and closed so that Ne = 2000 rpm and a change in shaft torque is from 5 kgm to 10 kgm. FIG. 6B shows a control result using the air-fuel ratio control device according to the present embodiment under the same conditions. 6 (a) and 6 (b), the control result of FIG. 6 (b) has a smaller variation, and the air-fuel ratio can be changed even in the transient state without inputting the detected value of the air-fuel ratio estimated value. It can be seen that is kept within a certain range of fluctuation.

According to the air-fuel ratio control device having such a configuration, since the input term of the neural network does not include the output value of the air-fuel ratio sensor that does not operate when the engine is in a low temperature state, such as at the time of starting when the engine is in a low temperature state, etc. In this case, appropriate feedforward control using a neural network can be performed. Further, it is not necessary to consider that the air-fuel ratio sensor deteriorates with the years of use.

Incidentally, in learning the neural network used in the air-fuel ratio estimating unit 20, it is desirable to perform learning in consideration of the engine misfire state. That is, at the time of driving, particularly at the time of engine start, even when the fuel is excessively injected from the injector and the air-fuel ratio is super rich, or even when the fuel injection amount is small and the air-fuel ratio is super lean, I will misfire. In both cases, the situation of misfiring is the same, so the air-fuel ratio sensor provided immediately after the exhaust valve of each cylinder or at the exhaust pipe collection section detects the air flowing into the cylinder as it is. Lean. Even though the air-fuel ratio is super rich, the neural network is trained using the data detected by the air-fuel ratio sensor as being super-lean. Control is performed so as to increase the amount, which causes inconvenience. Therefore, if such a misfire state is learned by a neural network in consideration of the cause of the misfire, such inconvenience can be eliminated.

The learning of the neural network in consideration of engine misfire will be described below. FIG. 7 is a diagram schematically illustrating a learning process of the neural network in consideration of engine misfire. FIG. 7 is different from FIG. 5 in that a misfire determination unit 10b is further provided. Misfire determination unit 10
b is the air-fuel ratio (A) detected by the air-fuel ratio sensor 10a.
/ N) and the fuel injection amount (Gf) are compared, and the air-fuel ratio (A /
If N) is super-lean and the fuel injection amount (Gf) is injected in a considerable amount, consistency is not ensured. In such a case, it is determined that the air-fuel ratio (A / F) is super-rich. To do it.

More specifically, for example, if a misfire occurs at the start of the engine, the air-fuel ratio should decrease uniformly, but the change in the air-fuel ratio (A / F) detected by the air-fuel ratio sensor is determined when the misfire occurs. As shown in FIG. 8, an interval X occurs in which the air-fuel ratio sensor reaches the limit value and the change is constant. If such a section X occurs even though the fuel injection amount (Gf) is present to some extent, there is no consistency between the air-fuel ratio sensor output and the fuel injection amount, and a misfire occurs in this section X. You can see that it was. Therefore, after driving the car and collecting data for learning, the value of the air-fuel ratio (A / F) in the section X is changed to a predetermined super-rich value, for example, about 10, and learning of the neural network is performed. Let Thus, even when the air-fuel ratio sensor detects that the air-fuel ratio is super-lean even though the air-fuel ratio is super-rich, it is possible to correct the error and to perform the learning appropriately.

When the fuel injection amount is monotonically increasing, for example, when a misfire occurs at the start of the vehicle, the change in the air-fuel ratio detected by the air-fuel ratio sensor increases as shown in FIG. Section Y appears. Therefore, instead of detecting the section X in which the output of the air-fuel ratio sensor is constant as described above, the change in the output of the air-fuel ratio sensor is increasing even though the fuel injection is monotonically increasing. It is also possible to detect a misfire state by detecting Y. In this case, after driving the car and collecting data for learning,
The neural network learning is performed by changing the air-fuel ratio in the section Y to a super-rich value. The super-rich value to be changed at this time is the air-fuel ratio (A /
It is desirable to calculate in consideration of the increase rate of F) and the increase rate of the fuel injection amount (Gf).

It should be noted that such a method of causing the neural network to perform learning in consideration of the misfire state can be appropriately applied to each embodiment described below. Although the fuel injection amount (Gf) and its change rate are considered here to know the misfire state, it is sufficient if the amount relates to the actual amount of fuel flowing into the cylinder, such as the throttle opening (THL). .

(Embodiment 2) Normally, a neural network guarantees that a good estimation accuracy is given when parameters within the range of learning data are input.
When a parameter out of the range of the learning data is input, the estimation accuracy is not guaranteed at all, and the estimated value may be largely shifted. Therefore, even in the air-fuel ratio control device described in the first embodiment, when a parameter out of the range of the learning data is input, accurate control may not be performed.

For example, in the air-fuel ratio control device of the first embodiment, the learning data range is given from -30 to 60 degrees with respect to the intake air temperature (Ta) which is an input parameter of the neural network. This is because the degree of change in the air-fuel ratio due to a change in the intake air temperature (Ta) outside this range is small, and the frequency outside this range is also small. It is almost enough to learn in this range. In this case, if the data of Ta = 70 degrees comes in, it is out of the learning range, so that the air-fuel ratio estimated value has an offset deviation as it is.

The air-fuel ratio control device according to the second embodiment is provided with means for coping with such a problem. FIG. 10 is a functional block diagram showing a configuration of the air-fuel ratio control device according to the second embodiment. The difference between this air-fuel ratio control device and the air-fuel ratio device according to Embodiment 1 is that the parameter range determination unit 40
a, a parameter conversion unit 40b is provided. The parameter range determination unit 40a is a unit that determines whether the value of the parameter detected by the state detection unit 10 is within a range set in advance for each parameter.
Here, the range from the minimum value to the maximum value of each parameter used at the time of learning the neural network of the air-fuel ratio estimating unit 20 is used as the range preset for the parameter.

The parameter conversion section 40b replaces the value of the parameter determined by the parameter range determination section 40a not within the preset range with a preset value. Specifically, if the parameter value exceeds the maximum value of the learning data, the parameter value is replaced with the maximum value of the learning data, and if the parameter value is lower than the learning data, Replace the value of the parameter.

The operation of the air-fuel ratio control device having such a configuration will be described below. FIG. 11 is a flowchart showing the operation of the air-fuel ratio control device. First, the state detection unit 10
Detects each physical quantity (S101). Next, the parameter range determination unit 40a determines whether the value of each parameter given by the detected physical quantity is within the set range (S102). Here, if there is a parameter that is not within the set range, the parameter conversion unit 40b converts the value of this parameter to a predetermined value (S103) and inputs the value to the air-fuel ratio estimation unit 20. The parameters within the set range are directly input to the air-fuel ratio estimation unit 20 without conversion. The air-fuel ratio estimator 20 estimates the air-fuel ratio based on these inputs, and the fuel correction amount calculator 30 calculates the fuel correction amount (ΔG
f) is calculated.

FIGS. 12 (a) and 12 (b) show experimental examples of control when such parameter value conversion is not performed and control is performed when such conversion is performed. Here, the intake air temperature (Ta) is used as a parameter to be converted. FIG.
The graph above (a) shows the actual air-fuel ratio A / F (dotted line).
FIG. 10 is a graph of an estimated air-fuel ratio A / Fnn (solid line) when the conversion is not performed even when the actual data of the intake air temperature (Ta) becomes a value outside the learning range. The lower graph of FIG. 12A is a graph of actual data of the intake air temperature (Ta). The upper graph in FIG. 12B shows the actual air-fuel ratio A /
It is a graph of F (dotted line) and the estimated air-fuel ratio A / Fnn (solid line) when the conversion is performed when the actual data of the intake air temperature (Ta) is outside the learning range. The lower graph of FIG. 12B shows actual data of the intake air temperature (Ta) (dotted line).
And a graph of the converted value (solid line). When observing each of these graphs, the upper graph in FIG. 12A shows a deviation between the actual air-fuel ratio A / F and the estimated air-fuel ratio A / Fnn. On the other hand, the upper graph in FIG. 12B shows the actual air-fuel ratio A / F by replacing the value of the intake air temperature (Ta) to the input term of the neuro with learning range data (solid line in the figure).
And the estimated air-fuel ratio A / Fnn has almost no bias. As described above, it was confirmed that the air-fuel ratio control device according to the present embodiment can prevent the estimation accuracy from significantly deteriorating even if the input term data is out of the learning range.

As described above, in the air-fuel ratio control apparatus according to the present embodiment, the air-fuel ratio estimating unit 20 is not limited to the case where the parameters to be interpolated are input within the range of the learning data of the neural network. Good estimation accuracy can be obtained even for parameters outside the range of the learning data, ie, extrapolated parameters. (Embodiment 3) In the above embodiment, the air-fuel ratio (A / F) is estimated from the input term of the neural network by omitting the output of the air-fuel ratio sensor that does not work at low temperatures, and using this Although the correction amount of the fuel injection amount is controlled, if the output of the air-fuel ratio sensor is omitted, an appropriate air-fuel ratio cannot be estimated in some cases.

FIGS. 13 (a) and 13 (b) show experimental examples showing this. FIG. 13A shows an air-fuel ratio (A / F) having an air-fuel ratio (A / F) as an input term as shown in FIG.
It is a figure which shows the teacher data for learning about a neural network which estimates F), and an actual output value. The input air-fuel ratio (A / F) is detected or calculated in the previous control cycle. FIG.
FIG. 14B shows training data and actual output values for a neural network for estimating an air-fuel ratio (A / F) having no air-fuel ratio (A / F) as an input term as shown in FIG. Is shown. When the air-fuel ratio (A / F) is an input term as shown in FIG. 13 (a), the teacher data and the actual output value substantially match, while the air-fuel ratio (A / F) as shown in FIG. 13 (b) When (A / F) is not included in the input term, the teacher data does not match the actual output value, and the actual output value of the neural network with respect to the teacher data is generally shifted downward including the steady state. It can be seen that a low value is output (this deviation is hereinafter referred to as “stationary bias”). Although the direct reason for the occurrence of this steady-state bias is unknown, it is believed that this steady-state bias is not always generated, but occurs when some conditions are met. However, since the fluctuation direction during the transition is correctly estimated, it is considered possible to use this as it is during the transition.

The air-fuel ratio control device according to the third embodiment has means for coping with such a bias in a steady state.
FIG. 15 is a functional block diagram illustrating a configuration of the air-fuel ratio control device according to the third embodiment. The air-fuel ratio control device according to the third embodiment includes a state detection unit 10, an air-fuel ratio estimation unit 20, a fuel correction amount calculation unit 30, a transient state detection unit 50, and a fuel correction amount adjustment unit 60. This air-fuel ratio control device is different from the air-fuel ratio control device of the first embodiment in that the transient state detection unit 5
0 and a fuel correction amount adjusting unit 60 are provided.

The transient state detecting section 50 is a section for detecting a transient state quantity of the engine E. Specifically, the transient state detection unit 50 receives the throttle opening (THL) and the engine speed (Ne), which are physical quantities obtained by the state detection unit 10, and receives the absolute value of the difference value of each value in each control cycle. The value of | ΔTHL | or | ΔNe | is compared with a predetermined set value. If the value is smaller than the set value, it is considered that the rate of change is small. If it is larger than this, it is determined to be in a steady state, and 1 is output as a transient state quantity.

As another method, set values are A, B (0 <
A <B), for example, when the throttle opening is input, if | ΔTHL | ≦ A, the rate of change is considered to be small, so 0 is output as the transient state quantity, and B ≦ | Δ
If THL |, the rate of change is considered to be large, so 1 is output as the transient state quantity. And A ≦ | ΔTHL | ≦
In the case of B, since the rate of change is considered to be between the transient and the steady state, for example,

[0071]

(Equation 3)

The value between 0 and 1 given by is output as a transient state quantity. Further, as another method, the values of the throttle opening (THL) and the engine speed (Ne) obtained by the state detection unit 10 may be input and passed through a high-pass filter. When the throttle opening (THL) and the engine speed (Ne) are small in the steady state and the rate of change is small, that is, when the frequency is low, the value of the high-pass filter is zero. On the other hand, when the rate of change is large in the transient state, that is, when the frequency is high, the high-pass filter outputs a value corresponding to the magnitude of the rate of change. Therefore, a value obtained by normalizing the absolute value of the gain from 0 to 1 may be output as the transient state quantity.

The fuel correction amount adjusting section 60 is provided with the transient state detecting section 5.
0, the correction amount (ΔG) of the fuel injection amount obtained by the fuel correction amount calculation unit 30 based on the transient state amount detected by
This is a part for adjusting f). Specifically, the correction amount (ΔGf) of the obtained fuel injection amount is multiplied by the detected transient state amount. That is, the transient state quantity takes a value from 0 to 1 as described above, and in the case of a definite transient state, it is multiplied by one. In the steady state, zero is applied, so the correction amount (ΔGfo) of the fuel injection amount is zero.

The operation of the air-fuel ratio control device having the above configuration will be described below. Now, it is assumed that the car is in a driving state. First, the engine speed (Ne), the intake air pressure (Pb), the throttle opening (THL), the fuel injection amount (Gf), the fuel injection amount (Gf), and the intake air temperature (the physical quantities that can be detected by the state detection unit 10 even at low temperatures in each control cycle). Ta), a cooling water temperature (Tw) is detected. The detected physical quantity is input as a parameter to the neural network of the air-fuel ratio estimating unit 20, and the air-fuel ratio is estimated. Then, the fuel correction amount calculation unit 30 calculates the correction amount (ΔGf) of the fuel injection amount from the estimated air-fuel ratio, and outputs it to the fuel correction amount adjustment unit 60. On the other hand, the throttle opening (THL) of the physical quantities detected by the state detection unit 10 is input to the transient state detection unit 50. The transient state detection unit 50 determines the throttle opening (THL) and the throttle opening (TH) input in the immediately preceding control cycle.
L), the transient state quantity which is a value between 0 and 1 is calculated and output to the fuel correction amount adjustment unit 60 as described above. The fuel correction amount adjusting unit 60 outputs a correction amount (ΔGfo) of the fuel injection amount adjusted by multiplying the input correction amount (ΔGf) of the fuel injection amount and the transient state amount. The correction amount (ΔGfo) of the output fuel injection amount is added to (Gfb) calculated by the basic fuel injection amount calculation unit 1 to calculate an actual fuel injection amount (Gf). The fuel is output from the injector I to the engine E.

If the vehicle is started, or the accelerator is rapidly closed or the accelerator is rapidly opened, 1 is output as the transient state quantity.
The fuel correction amount adjustment unit 60 outputs the value output by the fuel correction amount calculation unit 31 as it is. In addition, when the vehicle is operating in a steady state, 0 is output as a transient state amount, and the correction amount of the fuel injection amount is set to 0 by the fuel correction amount adjusting unit 60 and output. Will not be adjusted.

That is, since the fuel injection amount is not corrected when the vehicle is in a steady state, the air-fuel ratio (A / F) is not input to the neural network and the air-fuel ratio (A / F) by the neural network is input as described above. Even if the steady-state bias is applied to the output value of ()), the correction amount of the fuel injection amount is 0 in the steady state, and no extra correction is performed.
The correction amount of the fuel injection amount is output so as to suppress the fluctuation only at the time of a transition requiring correction. Note that the above-described transient state detecting section detects the transient state quantity by using the physical quantity obtained by the state detecting section 10 as an input. The air-fuel ratio estimated by the estimating unit 20 may be input, and the transient state amount may be detected from the amount of change in the air-fuel ratio. In this case, the transient state detection unit 51 outputs the estimated air-fuel ratio (A /
FNN), the absolute value | ΔA / FNN | of the difference value for each control cycle is used as the amount of change in the air-fuel ratio.
The amount of transient state can be detected by the same method as that for detecting the amount of transient state from the amount of change in physical quantity obtained at zero.

Further, as shown in FIG. 17, an air-fuel ratio sensor 52a for detecting the air-fuel ratio of the engine E and an amount of the transient state are calculated based on the amount of change in the output value of the air-fuel ratio sensor. It is also conceivable to use a configuration including the transient state calculation unit 52b. In this case, the method of calculating the amount of transient state by the transient state calculating unit 52b is a method of inputting the air-fuel ratio (A / FNN) estimated by the air-fuel ratio estimating unit 20 and detecting the amount of transient state from the amount of change. Is the same as In this case, when the engine temperature is low at the time of starting, the transient state detection unit 52 using the air-fuel ratio sensor does not operate, so the correction amount (ΔGf) of the fuel injection amount calculated by the fuel correction amount calculation unit 30 is adjusted by the fuel correction amount adjustment. The data is output as it is without being adjusted by the unit 60. However, since the start-up is usually a transition, the fuel injection amount correction amount (Δ
Gf) is output as it is because the transient state detector 52
Is the same as the output of the transient state quantity 1 from, so that there is no problem as a result.

As described above, in the air-fuel ratio control apparatus according to the present embodiment, even if the above-described steady-state bias is applied,
The fuel injection amount is not corrected in a steady state where no correction is required, and the fuel injection amount is corrected so that the air-fuel ratio approaches an appropriate value only in a transient period in which the correction is required. As a result, in the steady state, control without the influence of the bias in the steady state can be performed.
Appropriate air-fuel ratio control can be performed even during transition, especially at low temperatures.

(Embodiment 4) Embodiment 4 shows another example of the air-fuel ratio control device having means for coping with the above-described steady-state bias. FIG. 18 shows a functional block diagram illustrating a configuration of the air-fuel ratio control device according to the fourth embodiment. This air-fuel ratio control device includes a state detection unit 10, a change amount estimation unit 21, and a fuel correction amount calculation unit 31. This air-fuel ratio controller differs from the air-fuel ratio controller of the first embodiment in that
The change amount estimating unit 21 is provided instead of the air-fuel ratio estimating unit, and the fuel correction amount calculating unit 31 calculates the fuel correction amount based on the output value of the change amount estimating unit 21.

The change amount estimating section 21 is a section for estimating, using a neural network, a change amount of the physical quantity related to the air-fuel ratio using the physical quantity detected by the state detecting section 10 as an input. Here, the air-fuel ratio (A / F) is used as the physical quantity related to the air-fuel ratio, and the change amount (ΔA /
As F), a differential value of the air-fuel ratio or a value passed through a high-pass filter that passes only a high-frequency signal component to the air-fuel ratio is used. That is, the change amount (ΔA / F) of the air-fuel ratio is given by the following equation.

[0081]

(Equation 4)

Or

[0083]

(Equation 5)

Here, s is a Laplace operator, and f (s) is a high-pass filter that cuts low frequency signal components and passes high frequency signal components. As the parameter related to the air-fuel ratio, in addition to the air-fuel ratio, the weight of the air flowing into the cylinder or the amount of fuel flowing into the cylinder may be used. Since the change amount (ΔA / F) of the air-fuel ratio, which is the result of these calculations, is a differential value or a value that has passed through a high-pass filter, it is not affected by a bias in a steady state.

Here, the learning process of the neural network used in the variation estimating unit 21 will be described with reference to FIG. FIG. 19 is a schematic diagram showing a learning process of this neural network. FIG. 19 differs from FIG. 5 in that the collected air-fuel ratio (A / F) is converted into an air-fuel ratio change amount (ΔA / F) using the above equation [1] or equation [2], and a neural network. Is the change in the air-fuel ratio (ΔA / F). Therefore, the learning process is almost the same as the neural network for estimating the air-fuel ratio shown in FIG. That is, the amount of change in air-fuel ratio (ΔA / F) obtained from the detected value and each parameter are input to the neural network using the learning data obtained by actually driving the car, and are estimated as outputs. The configuration of the neural network is changed so that the deviation e from the change amount (ΔA / FNN) of the air-fuel ratio is reduced by the back propagation method.

Here, the change amount estimating section 21 uses the neural network to change the air-fuel ratio change amount (ΔA / F).
Is directly estimated. First, the first embodiment
The air-fuel ratio (A / F) is estimated by the same neural network as the air-fuel ratio estimating unit 20 of FIG. 1, and the estimated air-fuel ratio (A / F) is changed using the above equation [1] or equation [2]. The amount (ΔA / F) may be calculated.

The fuel correction amount calculation unit 31 calculates the correction amount (ΔGf) of the fuel injection amount from the change amount (ΔA / F) of the air-fuel ratio estimated by the change amount estimation unit 21 having the neural network learned as described above. ) Is calculated. In particular,
The change amount (ΔA / F) of the air-fuel ratio and the correction amount (Δ
Gf) is approximated to a P (proportional) relationship, a PI (proportional, derivative) relationship, etc.

[0088]

(Equation 6)

Or

[0090]

(Equation 7)

The calculation is performed according to an equation given by setting the above.
K3 and K4 are constants derived from experiments and the like. The operation of the air-fuel ratio control device having the above configuration will be described below. Now, it is assumed that the engine temperature is low when the vehicle starts running. First, the state detection unit 1
0 is a physical quantity that can be detected even at low temperatures, such as the engine speed (Ne), intake air pressure (Pb), and throttle opening (TH).
L), the fuel injection amount (Gf), the intake air temperature (Ta), and the cooling water temperature (Tw) are detected. Next, the change amount estimating unit 21 estimates and outputs the change amount (ΔA / F) of the air-fuel ratio using a neural network using the plurality of detected physical amounts as input parameters. Estimated air-fuel ratio change (ΔA /
Using F), the fuel correction amount calculation unit 31 calculates the correction amount (ΔGf) of the fuel injection amount. The actual fuel injection amount (Gf) is calculated by adding the calculated correction amount (ΔGf) of the fuel injection amount and (Gfb) calculated by the basic fuel injection amount calculation unit 1 described above, and the injector I, and this amount of fuel is injected into the engine E.

As described above, in the air-fuel ratio control device of this embodiment, the amount of change in the physical quantity related to the air-fuel ratio is estimated using the neural network, and the correction amount of the fuel injection amount is calculated from the estimated amount of change. That is,
Since the amount of change is not affected by the steady-state bias, the influence of the steady-state bias can be excluded from the correction amount of the fuel injection amount.

(Embodiment 5) If the correction amount of the fuel injection amount is directly calculated by a neural network as in an air-fuel ratio control device using a conventional neural network, setting of teacher data and fuel injection It is difficult to measure the amount of correction of the amount, and when accuracy is not obtained at the time of learning, it is difficult to grasp where there is a problem, and it is expected that a very long time and effort will be required in the development process. This can be mitigated by estimating the air-fuel ratio or the like using a neural network as in the above-described embodiment, and calculating the fuel correction amount from this. In the present embodiment, an air-fuel ratio control device that is easier to develop than estimating the air-fuel ratio and the amount of change by a neural network will be described.

FIG. 20 shows the configuration of the air-fuel ratio control device according to the fifth embodiment. The air-fuel ratio control device includes a state detection unit 10
And the inflowing air weight estimating unit 22 and the fuel correction amount calculating unit 32
It is composed of The air-fuel ratio control device according to the first embodiment
The difference from the air-fuel ratio control device is that the state detection unit 10 has an air-fuel ratio sensor unit 10a for detecting the air-fuel ratio, and the detection of the fuel injection amount (Gf) is indispensable. The second embodiment is different from the first embodiment in that an inflow air weight estimating unit 22 is provided instead of the unit and the input parameters of the fuel correction amount calculating unit 32 are different.

The inflowing air weight estimating unit 22 includes the state detecting unit 10.
This is a portion for estimating the in-cylinder air weight (Qa) by a neural network using at least two or more of the physical quantities detected by the above as input parameters. The learning of the neural network used here is performed by the learning process shown in FIG. The learning process shown in FIG. 21 will be briefly described. Each parameter is input to a neural network using learning data collected by actually driving a car, and the estimated in-cylinder air weight (QaNN) is output as an output. ) Detected fuel injection amount (Gf)
To calculate an estimated air-fuel ratio (A / FNN). Here, the air-fuel ratio (A / F) may be used as an input term as a parameter. At that time, it is necessary to add the value of the air-fuel ratio (A / F) to the input term of the inflow air weight estimation unit 22. Then, the configuration of the neural network is changed so that the deviation e between the estimated air-fuel ratio (A / FNN) and the detected air-fuel ratio (A / F) is reduced by the back propagation method. As a result, the neural network is configured so that the in-cylinder air weight (QaNN) can be estimated using various parameters as input.

Here, the in-cylinder air weight (QaNN) is calculated by the neural network using the air-fuel ratio sensor value that can be detected as a teacher signal. In comparison, trial and error for converging the learning can be reduced, and even when accuracy is not obtained in the learning process, it is possible to easily find out where the problem is.

The fuel correction amount calculation unit 32 calculates the fuel injection amount (Gf) and the air-fuel ratio (A /
F), and the correction amount (ΔGf) of the fuel injection amount is calculated from the cylinder inflow air weight (QaNN) estimated by the inflow air weight estimation unit 22. This fuel injection amount correction amount (ΔG
The calculation formula of f) will be described below. Even when fuel is injected from the injector, not all of the fuel flows into the cylinder, but a part of the fuel adheres to the wall of the intake pipe, and a part of the fuel that has adhered evaporates and flows into the cylinder. Therefore, the fuel injection amount (Gf) is different from the cylinder inflow fuel amount (Gfc) actually flowing into the cylinder, and this difference is calculated as the fuel injection amount correction amount (ΔGf). Therefore, first, in order to calculate the correction amount (ΔGf) of the fuel injection amount, the cylinder inflow fuel amount (Gfc) actually flowing into the cylinder is calculated.

The relationship among the air-fuel ratio A / F, the cylinder inflow air weight Qa, and the cylinder inflow fuel amount Gfc is given by the following equation.

[0099]

(Equation 8)

Now, if this equation is modified and the detection delay of the A / F is assumed to be one sampling, the fuel amount (Gfc) flowing into the cylinder becomes

[0101]

(Equation 9)

Are represented as follows. From this equation, Gfc (z) is obtained by substituting the air-fuel ratio (A / F) detected by the state detection unit 10 and the cylinder inflow air weight (QaNN) estimated by the inflow air weight estimation unit 22. Then, the state detection unit 1
The difference between the fuel injection amount (Gf) detected by 0 and the in-cylinder fuel amount (Gfc) is calculated as a correction amount (ΔGf) of the fuel injection amount. That is, the following equation:

[0103]

(Equation 10)

Thus, the correction amount of the fuel injection amount is calculated. As described later, the fuel injection amount (Gf),
Since the fuel adhesion rate (a) and the fuel evaporation rate (b) can be calculated from the air-fuel ratio (A / F) and the weight of the air flowing into the cylinder (QaNN), the fuel correction amount calculation unit 32 calculates the fuel adhesion rate ( After calculating a) and the fuel evaporation rate (b),
Using a map showing the relationship between the fuel adhesion rate (a), the fuel evaporation rate (b), and the correction amount (ΔGf) of the fuel injection amount,
The correction amount (ΔGf) of the fuel injection amount may be calculated. Here, the detection delay of the air-fuel ratio (A / F) is set to 1
Although sampling is assumed, a configuration in which the detection delay is variably given according to the operation state may be adopted.

The operation of the air-fuel ratio control device having the above configuration will be described below. Now, it is assumed that the car is in a driving state. First, in each control cycle, the state detection unit 10 determines the engine speed (Ne), the intake air pressure (Pb), the throttle opening (THL), and the fuel injection amount (Gf), which are physical quantities indicating the state of the engine E to the engine E. , Intake air temperature (T
a), the cooling water temperature (Tw) and the air-fuel ratio (A / F) are detected. Then, among the detected physical quantities, the air-fuel ratio (A / F)
Are input as parameters to the neural network of the inflowing air weight estimation unit 22, and the neural network estimates the inflowing air weight (Qa) in the cylinder. Then, the fuel injection amount (Gf) and the air-fuel ratio (A / F) detected by the state detection unit 10 and the in-cylinder inflow air weight (Qa) estimated by the inflow air weight estimation unit 22 are calculated by the fuel correction amount calculation unit 32. The fuel correction amount calculation unit 32 calculates the correction amount (ΔGf) of the fuel injection amount from these values. An actual fuel injection amount (Gf) is calculated from the calculated correction amount (ΔGf) in the same manner as in the above-described embodiment.
This amount of fuel is injected from injector I to engine E.

The air-fuel ratio control device having such a configuration is configured to estimate the inflowing air weight by a neural network, so that the air-fuel ratio control device is trial and error compared with a configuration in which the correction amount of the fuel injection amount is directly calculated by the neural network. This reduces the number of steps to be performed, making it easier to determine where the problem is, even when accuracy is not obtained during learning, thus reducing the development process and contributing to a reduction in development cost and time. it can.

(Embodiment 6) FIG. 22 is a functional block diagram of an air-fuel ratio control device including an abnormality / deterioration detection device according to the present invention. This is obtained by adding a sensor characteristic determination unit 70 and an air-fuel ratio sensor 80 to the air-fuel ratio control device shown in FIG. Sensor characteristic determination unit 7 serving as a means
0, constituted by an air-fuel ratio sensor 80.

This device detects a change in dynamic characteristics of an air-fuel ratio sensor using an air-fuel ratio estimated by a neural network. That is, in the basic fuel injection amount calculation unit 1 of the control device shown in FIG. 3, the feedback control is performed based on the value of the air-fuel ratio sensor. It may deteriorate and oscillate at worst. Therefore, as a general measure, the feedback gain is reduced, and the control performance is reduced. Therefore, in the present embodiment, even if the air-fuel ratio sensor deteriorates without excessively lowering the feedback gain, the degree of deterioration can be determined, and oscillation of the control system can be prevented.

In FIG. 22, the state detector 10, the air-fuel ratio estimator 20, and the fuel correction amount calculator 30 are the same as those shown in FIG. The air-fuel ratio sensor 80 measures the air-fuel ratio of the exhaust pipe, and uses the air-fuel ratio sensor q in FIG. The sensor characteristic determination unit 70 detects a change in dynamic characteristics of the air-fuel ratio sensor by comparing the estimated value of the air-fuel ratio obtained by the air-fuel ratio estimation unit 20 with the air-fuel ratio detected by the air-fuel ratio sensor 80.

More specifically, the sensor characteristic determination unit 70
Has an extreme value detecting unit 70a and an extreme value comparing unit 70b as a specific configuration as shown in FIG. Extreme value detector 7
0a is the air-fuel ratio (A / F) estimated by the air-fuel ratio estimation unit 20.
NN) and the air-fuel ratio detected by the air-fuel ratio sensor 80 (A /
An extreme value is obtained by differentiating each of F). And
The extreme value comparing unit 70b compares the time and value at which the extreme value is obtained with each of the obtained extreme values, and calculates the phase lag (Δδ) and the gain shown in FIG. 23B as characteristics of the air-fuel ratio sensor. Find the change (ΔK). Note that the extreme value comparison unit 70
A time constant or a dead time may be adopted as the output value of b.

The amount of change in the dynamic characteristics of the air-fuel ratio sensor calculated by the sensor characteristic determination unit 70 is sent to the basic fuel injection amount calculation unit 1 shown in FIG. Then, the basic fuel injection amount calculation unit 1 can maintain the control performance by changing the feedback gain based on this.
The operation of the abnormality / deterioration detecting device will be briefly described. First, the state detecting section 10 detects physical quantities such as the engine speed (Ne) and the intake air pressure (Pb), and the air-fuel ratio estimating section 20 detects the physical quantities. The air-fuel ratio (A / FNN) is estimated and output by a neural network using a plurality of physical quantities as input parameters. At the same time, the air-fuel ratio sensor 80 detects the air-fuel ratio (A / F) after the catalyst. Then, the sensor characteristic determination unit 70 compares the estimated air-fuel ratio (A / FNN) with the detected air-fuel ratio (A / F), calculates the phase change amount and the gain change amount, and calculates the basic fuel injection amount. Send to unit 1. Receiving this, the basic fuel injection amount calculation unit 1 adjusts the feedback gain to an appropriate value according to the phase change amount and the gain change amount.

FIGS. 24 (a) and 24 (b) show that the feedback gain is changed in consideration of the dynamic characteristics of the air-fuel ratio sensor as in this embodiment when the throttle opening is suddenly opened / closed at random. Comparative examples of control results when the control is performed and when the control is not performed are shown. FIG. 24A shows a control result when the feedback gain is not changed using the deteriorated sensor, and FIG. 24B shows a control result when the feedback gain is changed based on the dynamic characteristic change of the air-fuel ratio sensor. It is. Comparing these, it can be seen that the vibration is suppressed by changing the feedback gain based on the dynamic characteristic change.

In the abnormality / deterioration detecting device described above, the sensor characteristic judging section 70 compares the air-fuel ratio output from the air-fuel ratio sensor 80 with the extreme value of the air-fuel ratio output from the air-fuel ratio estimating section 20 over time. By doing so, the change in the dynamic characteristic of the air-fuel ratio sensor 80 is detected, but it is also possible to estimate this by a neural network. FIG. 25 shows a functional block diagram of an abnormality / deterioration detection device having a sensor characteristic determination unit 71 for estimating the dynamic characteristics of the air-fuel ratio sensor using a neural network.

The sensor characteristic determining section 71 in FIG.
The air-fuel ratio (A / FNN) estimated by the air-fuel ratio estimator 20, the air-fuel ratio (A / F) detected by the air-fuel ratio sensor 80, and the physical quantity output by the state detector 10 are input items, and the output item is A neuro calculation is performed using the dynamic characteristic change amount of the air-fuel ratio sensor. Here, the dynamic characteristic of the air-fuel ratio sensor is defined as a first-order lag, and the amount of change in the dynamic characteristic is defined as the amount of change in the time constant and gain of the first-order lag model. The control gain in the basic fuel injection amount calculation unit 1 is also changed according to the change amount of the first-order lag model. When the dynamic characteristic is set to be equal to or more than the second order delay, 2
It outputs the change amount of two time constants. Further, the dead time may be output in addition to the time constant and the gain.

According to such a configuration, a neuro operation is performed by using the estimated value of the air-fuel ratio when the sensor is not deteriorated and the actual output value of the air-fuel ratio sensor as inputs.
The deterioration degree can be determined with a small number of development man-hours. By setting the control gain in the basic fuel injection amount calculation unit 1 to be smaller in advance according to the phase lag when the phase lag increases, the oscillation can be suppressed as well. The neuro output term may be configured to directly output the control gain used in the basic fuel injection amount calculation section 1.

As described above, according to the abnormality / deterioration detecting device of the present embodiment, even if the air-fuel ratio sensor deteriorates, the feedback gain is adjusted in accordance therewith, so that the control performance is not reduced and oscillation is prevented. Can be. (Embodiment 7) Next, an abnormality / deterioration detecting device for detecting deterioration of a catalyst in an exhaust pipe will be described. FIG. 26 shows a functional block diagram of an air-fuel ratio control device including the abnormality / deterioration detection device according to the seventh embodiment. This abnormality / deterioration detection device includes a state detection unit 10, an air-fuel ratio estimation unit 20a, a fuel correction amount calculation unit 30, a catalyst deterioration detection unit 72, and an air-fuel ratio sensor 80. This abnormality / deterioration detection device differs from the abnormality / deterioration detection device shown in Embodiment 6 in that the air-fuel ratio sensor 80 detects the air-fuel ratio behind the catalyst in the exhaust pipe.
The point is that a catalyst deterioration detecting section 72 is provided as abnormality / deterioration detecting means, and the fuel correction amount calculating section 30 calculates a fuel correction amount from an air-fuel ratio before the catalyst.

In order to estimate the air-fuel ratio before the catalyst in the exhaust pipe by the air-fuel ratio estimating unit 20a, the air-fuel ratio estimating unit 20a
27, an air-fuel ratio sensor for a teacher signal is provided in front of the catalyst as shown in FIG. 27, and this air-fuel ratio (A / Fcp) is used as a teacher signal as in the neural network shown in FIG. What is necessary is just to learn so as to estimate the air-fuel ratio before the catalyst by the back propagation method.

The catalyst deterioration detector 72 compares the air-fuel ratio (A / FcpNN) estimated by the air-fuel ratio estimator 20a with the air-fuel ratio (A / F) detected by the air-fuel ratio sensor 80. Detect catalyst degradation. Specifically, when the catalyst is operating normally, even if the oxygen concentration before the catalyst changes, the change in the oxygen concentration after the catalyst is small due to the oxygen storage capacity of the catalyst, so that the change after the catalyst is small. Is smaller than the change in air-fuel ratio before the catalyst. However, when the catalyst deteriorates, the amount of change in the air-fuel ratio behind the catalyst decreases. From this, the catalyst deterioration detection unit 72 detects that the catalyst has deteriorated when the difference between the change amounts of the air-fuel ratio before and after the catalyst has become smaller than a predetermined value. Further, the catalyst deterioration detection unit 7
2 emits a predetermined signal when detecting that the catalyst has deteriorated. This signal is used, for example, as a trigger to alert the driver of the vehicle that the catalyst has deteriorated.

The operation of the abnormality / deterioration detecting device will be briefly described.
e), the physical quantity such as the intake air pressure (Pb) is detected, and the air-fuel ratio estimating unit 20 estimates and outputs the air-fuel ratio (A / FcpNN) before the catalyst by a neural network using the plurality of detected physical quantities as input parameters. I do. At the same time, the air-fuel ratio sensor 80 detects the air-fuel ratio (A / F) after the catalyst. The catalyst deterioration detector 72 compares the estimated air-fuel ratio (A / FcpNN) before the catalyst with the estimated air-fuel ratio (A / F) after the detected catalyst, and determines that the difference between the two is a predetermined value. It is determined whether or not: Here, when the difference is equal to or smaller than a predetermined value, the catalyst deterioration detecting section 72 issues a predetermined signal.
This signal activates an alarm so that the vehicle user can know the catalyst has deteriorated. On the other hand, if the difference between the estimated air-fuel ratio (A / FcpNN) before the catalyst and the detected air-fuel ratio (A / F) exceeds the predetermined value, the catalyst deteriorates. Therefore, the catalyst deterioration detecting section 72 does not emit a signal.

As described above, the abnormality / deterioration detecting device according to the embodiment can notify the user of the vehicle of the deterioration of the catalyst and prompt the exchange of the catalyst. (Embodiment 8) In this embodiment, an air-fuel ratio control device capable of performing appropriate air-fuel ratio control even if a response time delay occurs due to deterioration of the air-fuel ratio sensor is presented. FIG. 28 shows a functional block diagram of the air-fuel ratio control device according to the present embodiment. The air-fuel ratio control device includes a state detection unit 10, an air-fuel ratio estimation unit 20b, a fuel correction amount calculation unit 30, an air-fuel ratio sensor 8,
0, an operating state determining unit 90a and a response time detecting unit 90b. This air-fuel ratio control device is different from the air-fuel ratio control device of the first embodiment in that an air-fuel ratio sensor 80, an operating state determination unit 90a, and a response time detection unit 90b are provided, and an air-fuel ratio estimation unit 20b The point is that the response time (Tr) detected by the time detector 90b and the air-fuel ratio (A / F) detected by the air-fuel ratio sensor 80 are used as input parameters.

The air-fuel ratio sensor 80 controls the air-fuel ratio in the exhaust pipe. The operating state determining unit 90a determines whether the engine is in a predetermined operating state based on at least one of the physical quantities indicating the state of the engine detected by the state detecting unit 10. Here, the throttle opening (THL) and the engine speed (Ne) are used as physical quantities representing the state of the engine, and it is determined whether or not the engine is idling based on the throttle opening (THL) and the engine speed (Ne). Here, the physical quantity detected by the state detection unit 10 is used as it is, but the physical quantity for determining whether or not the engine is in a predetermined state may be separately detected.

The response time detecting section 90b detects at least one physical quantity detected by the state detecting section 10 when the operating state determining section 90a determines that the engine is in a predetermined operating state, here an idling state. The response time from the change time of the physical quantity to the change of the air-fuel ratio detected by the air-fuel ratio sensor is measured and stored. Here, the fuel injection amount is adopted as the physical quantity to be changed, and the fuel injection amount is increased by a predetermined minute amount (Gfs). Then, the time from when the fuel injection amount is increased to when the value detected by the air-fuel ratio sensor 80 changes, that is, the response time (Tr) of the air-fuel ratio sensor is measured.

The air-fuel ratio estimating section 20b includes a physical quantity (Ne, Pb...) Detected by the state detecting section 10, an air-fuel ratio (A / F) detected by the air-fuel ratio sensor, and a response detected by the response time detecting section 90b. Time (Tr) is input as a parameter, and the air-fuel ratio is estimated using a neural network. The learning process of the air-fuel ratio estimator 20b will be described below. FIG.
FIG. 9 shows a schematic diagram illustrating this learning process. As shown in the figure, this neural network receives as inputs the engine speed (Ne), intake air pressure (Pb), throttle opening (THL), fuel injection amount (G
f), intake air temperature (Ta), cooling water temperature (Tw),
An air-fuel ratio (A / F) and a response time (Tr) are input. The input air-fuel ratio uses the air-fuel ratio detected by the deteriorated air-fuel ratio sensor, and the response time (Tr) is the response time of the deteriorated air-fuel ratio sensor. On the other hand, the air-fuel ratio detected by the air-fuel ratio sensor without deterioration is used as the air-fuel ratio serving as the teacher signal. The air-fuel ratio sensor that detects the input air-fuel ratio is replaced with a plurality of air-fuel ratio sensors having different degrees of deterioration. / F) and the air-fuel ratio (A / FN) estimated by the neural network
The configuration of the neural network is changed by the back propagation method so as to reduce the deviation e from N). Thus, in this neural network,
Even if there is an input from the deteriorated air-fuel ratio sensor, an appropriate air-fuel ratio value without response delay or gain change can be estimated.

The operation of the air-fuel ratio control device having such a configuration will be described below. FIG. 30 is a flowchart showing this operation. First, the state detection unit 10 detects each physical quantity (S401). Next, the operating state determination unit 90a determines whether the engine is idling based on the throttle opening (THL) and the engine speed (Ne) among the detected physical quantities (S402). Here, if the engine is idling, the response time detecting section 90b increases the fuel correction amount by a predetermined amount (Gfs) (S40).
3). Then, the time from the time when the fuel correction amount is increased to the time when the detection value of the air-fuel ratio sensor 80 changes, that is, the response time is measured and recorded (S404).

Then, the air-fuel ratio estimating section 20b calculates the recorded response time together with the physical quantity detected by the state detecting section 10 and the air-fuel ratio (A / F) one control cycle before detected by the air-fuel ratio sensor 80. The air-fuel ratio is estimated by using a neural network by including the parameters (S405). If it is determined in S402 that the engine is not in the idling state, the response time recorded in the previous control cycle is input to the neural network. Finally, the fuel correction amount calculating section 30 calculates a fuel correction amount (ΔGf) from the estimated air-fuel ratio (A / FNN) (S406).

With such an operation, the response time of the air-fuel ratio sensor is recorded every time the engine is idling, and this is used for the air-fuel ratio estimation by the air-fuel ratio estimator 20b. Is output, so that appropriate air-fuel ratio control can be performed even if the air-fuel ratio sensor deteriorates. FIG. 31 (a), FIG.
(B) shows a comparison experiment result between the conventional air-fuel ratio control device and the air-fuel ratio control device of the present embodiment. These are respectively
FIG. 31A shows a control result when the throttle opening is suddenly opened and closed so that Ne = 2000 rpm and the shaft torque change is from 5 kgm to 10 kgm. FIG. b) is based on the air-fuel ratio control device of the present embodiment. All of the air-fuel ratio sensors used for these sensors are deteriorated. When these are compared, FIG.
It can be seen that the use of the predicted estimated value A / Fnn (solid line) shown in FIG.

Although the response time is used as the input term in the air-fuel ratio prediction estimating section 20b, a phase delay, a time constant, a gain, or the like may be used. In addition, the estimation accuracy can be improved by combining the response time and the gain as an input term. (Embodiment 9) Here, an air-fuel ratio control device that appropriately controls the fuel injection amount especially at the time of starting regardless of the type of fuel used by the user of the vehicle is shown.

FIG. 32 shows a functional block diagram of the air-fuel ratio control device according to the present embodiment. This air-fuel ratio control device includes a state detection unit 10, a fuel type determination unit 23, a fuel correction amount calculation unit 33, a fuel type storage unit 100, and an engine temperature detection unit 110. The state detection unit 10 detects a plurality of physical quantities representing the state of the engine E by a configuration substantially similar to that of the first embodiment. However, here, the output from the air-fuel ratio auxiliary control calculation unit 2 is a fuel correction coefficient (Cg
f), and this value is different in that the basic fuel injection amount (Gfb) output from the basic fuel injection amount calculation unit 1 is multiplied.

The fuel type discriminating section 23 is a section for judging the type of fuel by a neural network by using a plurality of detected parameters as inputs. The type of fuel mentioned here means the type of fuel currently on the market. In order to estimate the fuel type by the neural network, for example, the following learning may be performed. FIG.
FIG. 3 is a schematic diagram showing a learning process of the neural network for determining the fuel type. The neural network is configured so that a physical quantity (Gf, Ne,...) Detected by the state detection unit 10 that has been collected in advance is input, and a parameter representing gasoline properties is output. As the output, for example, it is conceivable to use a value of T50 which is a temperature at which gasoline is vaporized by 50%. This T50
Is known in advance for each type of fuel, the physical quantities (Gf, Ne,...) Collected by operating the engine using a certain type of fuel are input, and the T50 of the type of fuel used is input. Is the teacher data. The configuration of the neural network is changed so that the deviation e between the value of T50 estimated by such a configuration and T50, which is actual teacher data, is within an allowable value. Such learning is performed for all types of fuel to be estimated. Accordingly, it is possible to obtain a neural network that estimates and outputs T50s of all types of fuels to be estimated from the physical quantities detected by the state detection unit 10. If this neural network is used, the fuel type can be estimated from the estimated value of T50.

The configuration for determining the fuel type is as follows.
The following can also be adopted. That is, as shown in FIG. 34, the state detecting unit 10
Input the detected physical quantities (Gf, Ne,...)
There is provided a neural network 23a that has learned to output the air-fuel ratio (A / FNN). The air-fuel ratio (A / FNN) estimated by the neural network 23a and the air-fuel ratio (A / F) detected by the air-fuel ratio sensor are provided. ) And
A judging unit 23b for judging gasoline properties based on the difference between the two behaviors may be provided. The determination method in the determination unit 23b includes the air-fuel ratio (A / FNN) estimated by the neural network 23a, the air-fuel ratio (A / F) detected by the air-fuel ratio sensor, and the physical quantity (Gf) detected by the state detection unit 10. , Ne,...) As inputs and an output as T50.

The engine temperature detecting section 110 is a sensor which detects an engine cooling water temperature (Tw) as an engine temperature and converts it into an electric signal. Note that the temperature of the engine is not limited to the cooling water temperature, and the intake air temperature Ta or other temperature sensor output may be used. Also, the function using one or two or more of these temperature sensor output values is used. The determination may be made based on the output value of the function. Although the engine temperature detecting section 110 is provided separately from the state detecting section 10 here, the engine temperature detecting section 110 may share the cooling water temperature sensor in the state detecting section 10.

The fuel type storage unit 100 stores the fuel type determining unit 2
The fuel type estimated in step 3 is stored. Each time the fuel type is output from the fuel type determination unit 23, the fuel type is rewritten to the output type. When the engine temperature detected by the engine temperature detecting unit 110 is equal to or lower than the predetermined first set temperature, the fuel correction amount calculating unit 33 determines the fuel by the fuel type determining unit 23 and stores it in the fuel type storage unit 100. This is a part for calculating a correction coefficient (Cgf) as a correction amount of the fuel injection amount based on the detected fuel type, a map provided for each fuel type, and the detected engine temperature. Here, the first set temperature is set at a boundary point of a temperature at which the air-fuel ratio sensor does not operate, for example, about 50 ° C., and when it is lower than this temperature, that is, at a low temperature at which the air-fuel ratio sensor does not operate and the feedback control is not performed. The correction coefficient of the injected fuel at that time is calculated. Further, as shown in FIG. 35, the map provided for each fuel type stores a correction coefficient (Cgf) corresponding to each of the fuel types A to E. When the vehicle is stopped and the vehicle is started next time, the fuel correction amount calculating unit 33 stores the fuel type storage unit 100.
The correction coefficient (Cgf) for the fuel injection amount at the time of starting is calculated using the fuel type stored in the.

The operation of the air-fuel ratio control device having the above configuration will be described below. FIG. 36 is a flowchart showing the operation in one control cycle of the air-fuel ratio control device. Assuming that a certain type of fuel has been charged for the first time, when the vehicle is driven, the state detection unit 10 and the engine temperature detection unit 1
10 detects a plurality of physical quantities indicating the state of the engine including the cooling water temperature (Tw) (s501). Next, the fuel type determination unit 23 inputs the detected physical quantities into the neural network as parameters, and determines the type of fuel (s502). Then, the fuel type storage unit 100 stores the determined fuel type (s503). Then, the fuel correction amount calculating unit 33 determines whether or not the cooling water temperature (Tw) detected by the engine temperature detecting unit 110 is 50 ° C. or less (s504). If the cooling water temperature (Tw) is 50 ° C. or less. For example, a fuel injection amount correction coefficient (Cgf) is calculated from the fuel type determined as the cooling water temperature (Tw) using the above-described map (s505). Then, the calculated fuel injection amount correction coefficient (Cgf) is multiplied by the basic fuel injection amount (Gfb) to derive the actual fuel injection amount (Gf), and this amount of fuel is injected into the engine E. You. In addition, the cooling water temperature (Tw) is 5
When the temperature exceeds 0 ° C., the correction coefficient (Cgf) of the fuel injection amount is not calculated, and the correction of the fuel injection amount is left to the control of the basic fuel injection amount calculation unit 1 including normal feedback control.

Further, when the operation of the vehicle is once stopped and the vehicle is started again, that is, at the time of the next start, the engine temperature detecting unit 110 detects the cooling water temperature (Tw) in the first control operation. S501 to s50
The operation of step 3 is not performed. First, the fuel correction amount calculation unit 33 in s504 determines whether the cooling water temperature (Tw) is equal to or lower than 50 ° C. Here, it is at the time of startup, and the cooling water temperature (Tw) must
° C or less, the process proceeds to s505, and the fuel correction amount calculation unit 3
3 calculates a correction coefficient (Cgf) of the fuel injection amount at the time of starting from the fuel type stored in the fuel type storage unit 100 in the previous operation and the detected coolant temperature (Tw). Usually, it is considered that the fuel put in the vehicle is almost the same fuel at the next start, and thus the fuel injection amount at the start can be appropriately adjusted by such an operation.

The air-fuel ratio control apparatus further includes a fuel type discriminating unit 23 only when the temperature of the engine detected by the engine temperature detecting unit 110 is lower than a predetermined second set temperature as shown in FIG. May be provided. Here, as the second set temperature, for example, a boundary point at which a difference in properties (fuel evaporation rate, fuel adhesion rate, etc.) depending on the fuel type clearly appears 8
It is set to about 0 ° C.

The operation up to the storage of the fuel type in the fuel type storage unit 100 when the determination permission unit 111 is provided will be described with reference to FIG. FIG. 38 is a flowchart showing such an operation. First, in the operating state of the automobile, the state detection unit 10 and the engine temperature detection unit 110 detect a plurality of physical quantities indicating the state of the engine E including the cooling water temperature (Tw) (s601). Next, the determination permission unit 111 determines whether the cooling water temperature (Tw) is equal to or lower than 80 ° C. (s60).
2). If the cooling water temperature (Tw) is equal to or lower than 80 ° C., the determination permission unit 111 permits the fuel type determination unit 23 to determine the fuel type. The fuel type determination unit 23 permitted to determine the fuel type inputs a plurality of physical quantities detected in the above operation to the neural network as parameters,
The fuel type is determined (s604). Then, the fuel type storage unit 100 stores the determined fuel type (s60).
5). On the other hand, when the cooling water temperature (Tw) exceeds 80 ° C., the discrimination permission unit 111 does not permit the fuel type discrimination unit 23 to discriminate the fuel type (s606).
No. 3 does not determine the fuel type.

As described above, the accuracy of the determination can be improved by performing the fuel type determination by the neural network only in the temperature region where the difference in the fuel properties depending on the fuel type is remarkable. In addition, since the determination region can be limited (here, only the temperature region of 50 ° C. to 80 ° C.), the amount of learning data in the learning process of the neural network can be reduced, and the development time can be shortened.

In each of the above air-fuel ratio control devices, it is desirable to use the number of cranks and the battery voltage from when the spark plug is energized to when the air-fuel mixture in the engine completely explodes as parameters for estimating the fuel type. That is,
Since the number of cranks is a physical quantity that causes a significant difference depending on the type of fuel, and the battery voltage is a parameter that affects the number of cranks, it is possible to improve the accuracy of estimating the type of fuel by inputting them together. This is because FIG. 39 shows a functional block diagram of an air-fuel ratio control device having such a configuration. This air-fuel ratio control device is different from the air-fuel ratio control device shown in FIG.
Is further provided, and instead of the engine temperature detection unit 110,
A plug current detector 114.

The number-of-cranks detector 112 detects the number of cranks from energization of the spark plug to complete explosion of the engine as a physical quantity representing the state of the engine. Voltage detector 113
Detects the battery voltage as a physical quantity representing the state of the engine. The plug current detection unit 114 detects that the energization of the ignition plug has been turned on by a change in the position of the ignition key of the engine.

The operation of this air-fuel ratio control device is substantially the same as that of the air-fuel ratio control device shown in FIG. 32. Only when it is determined that the engine is in the starting state by energizing the ignition plug, the number of cranks at startup and the battery The fuel type is estimated in consideration of the voltage, the fuel correction amount is calculated based on the voltage, and the air-fuel ratio at the time of starting is appropriately controlled according to the fuel type to prevent a misfire or the like.

The fuel type storage unit 100 in the air-fuel ratio control apparatus is configured not to update the discrimination result when the number of cranks at the start detected by the crank number detection unit 112 is smaller than a certain set value. That is, only when the number of cranks at the time of starting is longer than a certain set value, the determination result is reflected to improve the startability, and when it is within the set value, the fuel injection at the time of starting is performed using the currently used judgment value. I do. This is because if the fuel injection amount is adjusted in accordance with the fuel whose starting crank number becomes shorter, that is, the fuel that is likely to evaporate, the user of the car replaces the fuel and the fuel becomes more difficult to evaporate. In such a case, the possibility of misfiring increases. That is, if a misfire occurs, the exhaust gas will be polluted much more than the air-fuel ratio slightly deviates from the target value. Therefore, priority is given to preventing this.

Further, in the above-described embodiment, the fuel type discriminating section 23 directly discriminates the fuel type by a neural network. However, as described later, this is performed by a neural network.
aNN is estimated, and the fuel adhesion rate (a) and the fuel evaporation rate (b) are calculated using the estimated aNN. A table describing the correspondence between the fuel adhesion rate (a), the fuel evaporation rate (b), and the fuel type. May be used to calculate the fuel type. A method for calculating the fuel adhesion rate (a) and the fuel evaporation rate (b) will be described below.

As described above, even when the fuel is injected from the injector, not all of the fuel flows into the cylinder but a part of the fuel adheres to the wall of the intake pipe, and a part of the adhered fuel evaporates. Flow into the cylinder. The amount of fuel adhering to the wall of the intake pipe varies in a complicated manner depending on the operating conditions (rotation speed, load (intake air pressure), etc.) and the external environment (intake air temperature, cooling water temperature, atmospheric pressure, etc.). The amount of fuel flowing into the cylinder also varies depending on the operating state and the external environment.
Therefore, first, the fuel adhesion amount (a) and the fuel evaporation amount (b) are calculated. FIG. 40 shows a simple model representing the behavior of fuel adhesion. Briefly explaining FIG. 40, when the sampling time is k, of the fuel injection amount (Gf), which is the amount of fuel injected from the injector, the wall surface fuel adhesion amount (Mf), which is the amount of fuel adhering to the wall surface, is a-Gf, and the remaining (1-a) -Gf flows into the cylinder. The amount of fuel evaporated from the attached fuel and flowing into the cylinder is represented by b · Mf. Therefore, the amount of fuel flowing into the cylinder (Gf
c) is the sum of (1-a) · Gf and b · Mf.

Here, assuming that the wall surface fuel adhesion amount at the sampling time k is Mf (k), the fuel injection amount is Gf (k), and the cylinder inflow fuel amount is Gfc (k), the following relational expression is established.

[0145]

[Equation 11]

[0146]

(Equation 12)

The following relational expression is obtained from these two expressions.

[0148]

(Equation 13)

As described above, the relationship among the air-fuel ratio A / F, the cylinder inflow air weight Qa, and the cylinder inflow fuel amount Gfc is as follows.
Assuming that after the detection of A / F is one sampling, it is given by the following equation.

[0150]

[Equation 14]

The following equations are obtained from the equations [3] and [4].

[0152]

(Equation 15)

Here, if (A / F) · Gf = W,

[0154]

(Equation 16)

From this equation, the fuel injection amount (Gf) and the air-fuel ratio (A / F) detected by the state detector 10 are used to calculate W
Is calculated, and two or more equations are obtained by substituting the cylinder inflow air weight (QaNN) estimated by the inflow air weight estimation unit by four or more samplings, and the fuel adhesion rate (a ), The fuel evaporation rate (b) can be calculated.

FIG. 41 shows a configuration of an air-fuel ratio control apparatus having a fuel type discriminating unit 24 using the method for obtaining the fuel adhesion amount (a) and the fuel evaporation amount (a). The fuel type determination unit 24 includes an inflow air weight estimation unit 24a, a fuel characteristic calculation unit 24b, and a type determination unit 24c, and the inflow air weight estimation unit 24a includes the inflow air weight estimation unit 22 shown in FIG.
Is the same as The fuel characteristic calculation unit 24b calculates the fuel injection amount (Gf), the air-fuel ratio (A / F), and the inflow air weight estimation unit 2 detected by the state detection unit 10 using the above-described calculation formula.
The fuel adhesion rate (a) and the fuel evaporation rate (b) are calculated from the in-cylinder air weight (QaNN) estimated by 4a.

The type discriminating section 24c is shown in FIGS. 42 (I) and (II).
The fuel adhesion rate (a) and the fuel evaporation estimated by the fuel characteristic calculation unit 24b are shown in FIG. 3 using a table showing the relationship between the fuel adhesion rate (a), the fuel evaporation rate (b), and the fuel type. The type of fuel is determined from the rate (b). FIG.
To explain the tables (I) and (II), first,
Based on the cooling water temperature (Tw), one of the tables shown in FIG. 42 (II) determined for each cooling water temperature by the table shown in FIG. 42 (I) is selected. For example, a table in which the cooling water temperature is 5 ° C. is selected. Then, the fuel type is determined from the selected table in FIG. 42 (II) based on the fuel adhesion rate (a) and the fuel evaporation rate (b).

Here, the type of fuel is determined from the fuel adhesion rate (a) and the fuel evaporation rate (b). This is because the fuel adhesion rate (a) or the fuel evaporation rate (b) is determined. The fuel type may be determined from any one of the above. In this case, it suffices that the fuel characteristic calculation unit 24b calculates only the necessary one of the fuel adhesion rate (a) and the fuel evaporation rate (b). It should be noted that the fuel adhesion rate (a) and the fuel evaporation rate (b) obtained by the above equation are learned as a teacher signal, and the a,
b may be estimated, and the correction coefficient (Cgf) of the fuel injection amount may be calculated from a and b.

Further, in order to ensure the determination of the fuel type in the fuel type determining section 24, the calculated fuel adhesion rate (a) and fuel evaporation rate (b) are sequentially sent to the fuel characteristic calculating section 24b. A storage unit that stores the information may be provided, and the type determination unit 24c may determine the type of the fuel using the plurality of sequentially stored fuel adhesion rates (a) and fuel evaporation rates (b). .

Also, in the air-fuel ratio control device having the fuel type discriminating section 24, the discriminating permission section 111 as shown in FIG. 37 can be provided. In this case, since the air-fuel ratio sensor does not operate below the first set temperature, the fuel type discriminating unit 23 which requires the input of the air-fuel ratio sensor does not perform the fuel type judgment below the first set temperature. Therefore, the fuel correction amount calculation unit 33 calculates the correction coefficient (Cgf) of the fuel injection amount exclusively using the fuel type stored in the fuel type storage unit 100 during the previous operation.

(Embodiment 10) In the ninth embodiment, the fuel injection amount at the time of starting is controlled in particular according to the fuel type. However, here, the fuel type is controlled regardless of whether or not the engine is started. 1 shows an air-fuel ratio control device that can be reflected in air-fuel ratio control. FIG. 43 shows a functional block diagram of the air-fuel ratio control device according to the present embodiment. This air-fuel ratio control device includes a state detection unit 10, an air-fuel ratio estimation unit 20c, and a fuel correction amount calculation unit 3.
0, a fuel type detection unit 120. This air-fuel ratio control device differs from the air-fuel ratio control device of the first embodiment in that a fuel type detection unit 120 is provided, that a state detection unit also detects an air-fuel ratio, that an air-fuel ratio estimation unit is used. 20c
Is the output value of the fuel type detection unit 120 and the air-fuel ratio (A /
F) is input.

The fuel type detector 120 detects the type of fuel used for the engine. The detection of the fuel type is performed by the fuel type determination unit 2 of the air-fuel ratio control device shown in FIG.
As in the case of 3, the estimation is performed by using a neuro, or by directly applying an ultrasonic wave to gasoline and detecting the density by measuring the density from the transmission speed. Further, the fuel type detection unit 12
0 outputs a value (Gs) corresponding to the fuel type. Specifically, the fuel type is divided into three groups according to the degree of volatility, and the higher the volatility group, the larger the value (Gs) is given and output.

The air-fuel ratio estimating unit 20c uses the physical quantities (Ne, Pb, A / F...) Detected by the state detecting unit 10 and the numerical values corresponding to the fuel type detected by the fuel type detecting unit 120 as input parameters. , By neural network,
Estimate the air-fuel ratio. FIG. 44 is a schematic diagram showing a learning process of this neural network. As shown in the figure, the physical quantity (Ne, Pb, A / F...) Detected by the state detection unit 10 and the value Gs according to the type of fuel actually used are used as input parameters, and the air-fuel ratio sensor is used. Configures a neural network using the detected air-fuel ratio (A / F) as teacher data, and changes the configuration of the neural network by a back propagation method so as to reduce the deviation between the estimated value and the teacher data. Here, the above-described values are used as values according to the type of fuel, but T50 or the like may be used. Thus, a neural network for estimating the air-fuel ratio in consideration of the fuel type can be obtained.

The operation of the air-fuel ratio control device having such a configuration will be briefly described. First, while the vehicle is running, the state detection unit 10 detects the engine speed (Ne), the intake air pressure (Pb), and the air-fuel ratio in the previous control cycle. (A / F)... And the like. At the same time, the fuel type detection unit 120 detects the type of fuel and outputs a value (Gs) corresponding to the type of fuel. Next, the air-fuel ratio estimating unit 20 uses a plurality of detected physical quantities (Ne, Pb, A / F,...) And a value (Gs) corresponding to the fuel type as input parameters to input an air-fuel ratio (A / F) is estimated and output. Using the estimated air-fuel ratio (A / F), the fuel correction amount calculation unit 30 calculates a correction amount (ΔGf) of the fuel injection amount. The actual fuel injection amount (Gf) is calculated by adding the calculated correction amount (ΔGf) of the fuel injection amount and (Gfb) calculated by the basic fuel injection amount calculation unit 1 described above, and the injector I, and this amount of fuel is injected into the engine E. Such an operation enables air-fuel ratio control according to the fuel type.

(Embodiment 11) If the control of the air-fuel ratio is performed by correcting the fuel injection amount, a misfire will occur if the fuel injection amount is excessively injected. Then, the output of the air-fuel ratio sensor outputs a high air-fuel ratio despite the low air-fuel ratio in the engine. As a result, for example, the output of the air-fuel ratio sensor at the time of starting the engine is determined as shown in FIG.
As shown in FIG. 45A, the output value of the air-fuel ratio sensor is mixed with a vibration component as shown in FIG. In a control system in which such an input parameter causes a vibration component to be added to a control target, it is possible to predict the degree of convergence of the vibration and correct the input parameter accordingly to quickly converge the vibration. Conceivable.

FIG. 46 shows a general configuration of a vibration correction control device for converging the vibration component at an early stage in the control system in which the input parameter causes the vibration component to be added to the control target. The vibration correction controller uses the convergence degree estimator 310 for estimating the degree of convergence of the control target by using a neural network with parameters related to the control target Z as input, and the control target converges quickly using the estimated convergence degree. Thus, the correction amount calculating unit 320 calculates the correction amount of the input parameter.

As a specific example, consider an air-fuel ratio control device. FIG. 47 shows the configuration. This air-fuel ratio control device
It comprises a state detection unit 10, a misfire degree estimation unit 25, a fuel correction amount calculation unit 34, a start state determination unit 130, and a fuel correction amount calculation permission unit 140. Misfire degree estimating unit 25,
The fuel correction amount calculation unit 34 corresponds to the convergence degree estimation unit 310 and the correction amount calculation unit 320, respectively.

The state detecting section 10 is a section for detecting parameters related to the control object input to the neural network. Here, a plurality of parameters representing the state of the engine E are detected by the same configuration as in the first embodiment. The misfire degree estimating unit 25 is a part that estimates the misfire degree of the engine E using a neural network with the plurality of parameters detected by the state detecting unit 10 as inputs. Here, the misfire degree of the engine E refers to the convergence degree of the output of the air-fuel ratio sensor provided immediately after the exhaust valve of the engine E or in the exhaust pipe assembly.

This misfire degree (R) can be expressed by various methods. For example, in FIG. 45 (b), a change amount Δm of a negative change rate portion and a change amount Δp of a positive change rate portion in a predetermined time width d at a point where the change rate of the output of the air-fuel ratio sensor changes from negative to positive. Can be determined as the misfire degree (R) by the sum of the absolute values of. It is also possible to use the inclination of the tangent at the sampling time in the portion where the change rate is positive as the misfire degree (R).

FIG. 48 shows the learning process of the neural network for estimating the degree of misfire (R). This diagram is basically similar to the learning process of the neural network for estimating A / F shown in FIG. The difference is that the output obtained by converting the output of the air-fuel ratio sensor into the misfire degree (R) by the misfire degree calculation unit 1c is the teacher data. While the output from the air-fuel ratio sensor is monotonically decreasing, the misfire degree calculation unit 1c sets 0 as the misfire degree.
Is output, and when the output from the air-fuel ratio sensor increases, an increase rate ΔA / F of the output of the air-fuel ratio sensor at the sampling time is calculated, and this is output as the value of the degree of misfire. ΔA / F is obtained by differentiating the collected A / F. In addition, each of the change amount Δm of the negative change rate portion and the change amount Δp of the positive change rate portion in the predetermined time width d from the point where the change rate of the output of the air-fuel ratio sensor changes from negative to positive as the degree of misfire. When using the sum of absolute values, ΔA
Instead of / F, the sum of Δm and Δp is calculated and output as the misfire degree in the positive rate of change portion.

When learning the neural network by the configuration of FIG. 48, the vehicle is driven from a stopped state to a steady state, learning data at the time of starting is collected, and the degree of misfire ( R) is calculated. This is input to the neural network with a plurality of parameters, and is compared with an estimator (RNN) of the degree of misfire estimated as an output to detect a deviation e, and a backpropagation method is used to reduce the deviation e. Learn by changing the configuration of the neural network.

The fuel correction amount calculating section 34 includes a misfire degree estimating section 2
The correction amount (ΔG) of the fuel injection amount from the misfire degree estimated in
This is a part for calculating f), and when the inclination of the tangent at the sampling time is the misfire degree (R), the correction amount (ΔGf) of the fuel injection amount is calculated by the following equation.

[0173]

[Equation 17]

This K5 is a constant obtained by experiments and the like. The absolute value of the change amount Δm of the negative change rate portion and the change amount Δp of the positive change rate portion in the predetermined time width d at the point where the change rate of the output of the air-fuel ratio sensor changes from negative to positive When the sum is the misfire degree (R), the correction amount (ΔGf) of the fuel injection amount is calculated by the following equation.

[0175]

(Equation 18)

This K6 is also a constant obtained by experiments and the like. The starting state determining unit 130 is a unit that determines whether the engine E is in a starting state based on at least one of the parameters detected by the state detecting unit 10. Here, the temperature of the cooling water of the engine E is used as a parameter. (T
Using w), when the temperature of the cooling water is equal to or lower than a predetermined value, here, equal to or lower than 50 ° C., the misfire degree estimating unit 25 determines that the engine E is in the starting state.

The fuel correction amount calculation permission unit 140 permits the fuel correction amount calculation unit 34 to calculate the correction amount of the fuel injection amount only when the starting state determination unit 130 determines that the engine E is in the starting state. Part. The operation of the air-fuel ratio control device having the above configuration will be described below. FIG.
FIG. 9 is a flowchart showing the operation of this air-fuel ratio control device in one control cycle. First, when the automobile is driven, the state detection unit 10 detects a physical quantity representing the state of the engine E (s701). Next, the misfire degree estimating unit 25 estimates the misfire degree by inputting the detected physical quantities into the neural network as parameters (s702). On the other hand, the starting state determining unit 130 determines whether the engine E is in the starting state from at least one of the parameters detected by the state detecting unit 10. Here, the engine E depends on whether the cooling water temperature (Tw) is 50 ° C. or less as a parameter.
It is detected whether or not is in the starting state (s703).

Here, if the cooling water temperature (Tw) is 50 ° C. or less, the starting state judging section 130 sets the fuel correction amount calculation permission section 1
A signal to that effect is output to 40, and upon receipt of this signal, the fuel correction amount calculation permission unit 140 permits the fuel correction amount calculation unit 34 to perform the calculation (s704). The fuel correction amount calculation unit 34 permitted to perform the calculation calculates the correction amount (ΔGf) of the fuel injection amount using the misfire degree estimated by the misfire degree estimation unit 25 (s).
705). Correction amount of calculated fuel injection amount (ΔGf)
Is added to the basic fuel injection amount (Gfb) to calculate the actual fuel injection amount (Gf), and this amount of fuel is injected from the injector I.

If the cooling water temperature (Tw) exceeds 50 ° C., the starting state judging section 130 judges that the engine E is not starting, and outputs a signal to that effect to the fuel correction amount calculation permitting section 140. In response to this, the fuel correction amount calculation permission unit 14
0 does not allow the fuel correction amount calculation unit 34 to calculate the fuel correction amount (s706), and as a result, the correction amount (ΔG
f) is not calculated.

Although the air-fuel ratio control device described here is provided with the starting condition judging unit 130 and the fuel correction amount calculating unit 90 and operates only at the time of starting, the starting condition judging unit 130 and the fuel correcting amount It is also possible to adopt a configuration in which the quantity calculation unit 90 is omitted. In that case, misfire degree estimating unit 2
The neural network of No. 5 may learn the misfire degree not only at the time of starting but also under various driving patterns in the learning process.

[0181]

As described above, the present invention has the following effects. First, a state detecting means according to the present invention,
In an air-fuel ratio control device having an air-fuel ratio detection sensor and a fuel correction amount calculating means, a plurality of physical quantities which can be detected even at a low temperature and which represent the state of the engine detected by the state detecting means are used as parameters. And the air-fuel ratio estimating means estimates the air-fuel ratio based on these parameters. Then, the fuel correction amount calculating means calculates a correction amount of the fuel injection amount from the estimated air-fuel ratio.

With this operation, the air-fuel ratio can be estimated by the neural network even at a low temperature at which the air-fuel ratio sensor does not work, and an appropriate fuel correction amount can be calculated from this. Therefore, the air-fuel ratio control by the neural network is performed even at the low temperature. It is not necessary to consider the deterioration of the air-fuel ratio sensor. Further, a time-series data storage unit for storing the time-series data of the fuel injection amount is provided, and when this is added to the parameter input to the air-fuel ratio estimation unit, the time-series data of the fuel injection amount is large in the change of the air-fuel ratio. Because of the influence, the accuracy of the estimation of the air-fuel ratio is improved.

When the parameter range determining means and the parameter converting means are provided in the air-fuel ratio control device, the parameter range determining means determines that the value of at least one parameter inputted to the air-fuel ratio estimating means is a value of the parameter. On the other hand, it is determined whether or not the value is within a preset range, and the parameter conversion unit replaces the value of the parameter determined not to be within the preset range with a preset value. With such an operation, when the parameter input to the neural network is an unexpected value, the parameter is converted to a predetermined value appropriately determined in advance and input to the neural network, so that the estimated air-fuel ratio is reduced. Good accuracy is maintained.

If the predetermined value is determined based on the maximum value and the minimum value of the physical quantity serving as the parameter, which are input in the learning of the neural network, the neural network Parameters outside the learning range, whose operation cannot be guaranteed, can be converted to appropriate values, and the accuracy of the estimated air-fuel ratio can be kept constant.

Further, if the air-fuel ratio control device is provided with a transient state amount detecting means and a fuel correction amount adjusting means, the transient state detecting means detects the transient state amount of the engine, and the fuel correction amount adjusting means is detected. The correction amount of the fuel injection amount obtained by the fuel correction amount calculation means is adjusted based on the transient state amount. As a result, if the correction of the fuel injection amount is suppressed in the steady state, the bias in the steady state caused by not adding the air-fuel ratio to the input parameters of the neural network can be dealt with, and the appropriate air-fuel ratio control can be performed in the transient state Becomes

Also, the transient state detecting means may detect the transient state quantity based on the change amount of at least one parameter obtained by the state detecting means or the change amount of the air-fuel ratio estimated by the air-fuel ratio calculating means. For example, the amount of transient state can be detected without providing a new sensor or the like.
Further, if the transient detecting means is constituted by an air-fuel ratio sensor and a transient state calculating means for detecting a transient state amount based on a change amount of an output value of the air-fuel ratio sensor,
Since the air-fuel ratio sensor measures the air-fuel ratio and the transient state calculating means calculates the change amount of the air-fuel ratio, the transient state amount of the engine can be calculated from the change amount of the air-fuel ratio.

When the output of the air-fuel ratio sensor is in a super-lean state during the learning process and is not consistent with the fuel injection amount, the neural network used in the air-fuel ratio estimating means is set to a super-rich state. When learning using the teacher data containing certain information is used, even if the output of the air-fuel ratio sensor at the time of learning differs from the actual air-fuel ratio due to misfire, etc., learning of the neural network that follows the actual air-fuel ratio Therefore, the obtained neural network can appropriately estimate the air-fuel ratio.

Then, the air-fuel ratio estimating means is provided with fuel type detecting means for detecting the type of fuel used in the engine, and a numerical value corresponding to the property of the detected fuel type is input to the air-fuel ratio estimating means. If the parameters are included in the parameters, the air-fuel ratio can be controlled according to the type of fuel selected by the vehicle user. Further, a state detecting means
In the air-fuel ratio auxiliary control device according to the present invention including the inflow air weight estimating means and the fuel correction amount calculating means, a plurality of parameters representing the state of the engine including the fuel injection amount and the air-fuel ratio are detected by the state detecting means, The inflow air weight estimating means estimates at least two or more of the plurality of detected parameters as inputs and estimates the inflow air weight of the cylinder by a neural network, and the fuel correction amount calculating means detects the fuel injection amount and the air-fuel ratio detected by the state detecting means. The correction amount of the fuel injection amount is calculated from the cylinder inflow dead weight estimated by the inflow air weight estimating means.

By adopting a configuration in which the inflow air weight is estimated by the neural network, the rate of trial and error is reduced as compared with a configuration in which the correction amount of the fuel injection amount is directly calculated by the neural network. Even if the accuracy is not obtained, it is possible to more easily determine where the problem is, so that the number of development steps can be reduced, which contributes to reduction of development cost and development time.

Further, in the air-fuel control device according to the present invention comprising the state detecting means, the change amount estimating means, and the fuel correction amount calculating means, a plurality of detecting means which can detect even at a low temperature detected by the state detecting means. Are input to the neural network of the change amount calculating means, a parameter related to the air-fuel ratio is estimated using the neural network, and the fuel injection amount is calculated from the change amount of the parameter related to the air-fuel ratio estimated by the fuel correction amount calculating means. Is calculated.

As described above, the amount of change in the parameter related to the air-fuel ratio is estimated using the neural network, and the correction amount of the fuel injection amount is calculated from the estimated amount of change, that is, the amount of change is constant. Since there is no influence even if there is a constant bias, the influence of the bias at the steady state can be excluded from the correction amount of the fuel injection amount, and only the parameters that can be detected even at low temperatures are used as the input terms of the neural network. In addition, since the air-fuel ratio is appropriately controlled even when the engine is at low temperature, particularly at the time of starting, it is possible to reduce the harmful gas in the exhaust gas when the engine is at low temperature.

If the change amount estimating means first estimates the state amount of the parameter related to the air-fuel ratio using a neural network and calculates the change amount from the estimated state amount, the estimated air-fuel ratio It is possible to use the state quantity of the parameter related to as an input parameter of another control device. Further, by setting the parameter relating to the air-fuel ratio to the air-fuel ratio itself, the correction amount of the fuel injection amount can be calculated with the highest accuracy.

Then, when the output of the air-fuel ratio sensor is super-lean in the learning process and is not consistent with the fuel injection amount, the neural network used in the above-mentioned change amount estimating means is in the super-rich state. Assuming that what is learned using the teacher data including the information that exists is used, even if the output of the air-fuel ratio sensor at the time of learning is different from the actual air-fuel ratio due to misfiring, etc. Since the network can be learned, it is possible to appropriately estimate the amount of change in the parameter related to the air-fuel ratio.

Further, in the abnormality / deterioration detecting means according to the present invention having the state detecting means, the air-fuel ratio estimating means, the air-fuel ratio sensor, and the sensor characteristic judging means, the state detecting means has a low temperature indicating the state of the engine. However, the plurality of detectable physical quantities are detected, and the air-fuel ratio estimating means estimates the air-fuel ratio by inputting the detected plurality of physical quantities as parameters. on the other hand,
An air-fuel ratio sensor detects the actual air-fuel ratio. Then, the sensor characteristic determining unit compares the air-fuel ratio estimated by the air-fuel ratio sensor estimating unit with the air-fuel ratio detected by the air-fuel ratio sensor, and detects a change in dynamic characteristics of the air-fuel ratio sensor.

By such an operation, a change in the dynamic characteristic of the air-fuel ratio sensor can be detected. From this change in the dynamic characteristic, for example, it is possible to appropriately adjust the feedback gain of a system that performs feedback control using the air-fuel ratio sensor. it can. The above-mentioned sensor characteristic determining means obtains respective extreme values by differentiating the time change of the air-fuel ratio estimated by the air-fuel ratio estimating means and the time change of the air-fuel ratio detected by the air-fuel ratio sensor. By comparing the time and the value of the extreme value of the above, the phase lag and the gain change of the air-fuel ratio sensor are obtained as the dynamic characteristic change of the air-fuel ratio sensor, and the feedback gain is adjusted using the changes as they are. be able to.

Further, the sensor characteristic determining means may use at least one physical quantity detected by the state detecting means, an air-fuel ratio estimated by the air-fuel ratio estimating means, and an air-fuel ratio detected by the air-fuel ratio sensor as parameters. By inputting and estimating the dynamic characteristic change of the air-fuel ratio sensor using a neural network, it is possible to estimate various dynamic characteristic changes as necessary in addition to the above-described phase delay and gain change. This can be used to make a wide range of system adjustments.

Further, the air-fuel ratio estimating means estimates the air-fuel ratio before the catalyst in the exhaust pipe, and the air-fuel ratio sensor detects the air-fuel ratio after the catalyst in the exhaust pipe. Is replaced by catalyst deterioration detecting means for detecting catalyst deterioration by comparing the air-fuel ratio estimated by the air-fuel ratio estimating means with the air-fuel ratio detected by the air-fuel ratio sensor. For example, the user of the car can be prompted to change the catalyst.

Further, the state detecting means, the air-fuel ratio sensor,
In the air-fuel ratio control device including the operating state determining unit, the response time detecting unit, the air-fuel ratio estimating unit, and the fuel correction amount calculating unit, the state detecting unit can detect a plurality of physical quantities that can be detected even at a low temperature indicating the state of the engine. And the air-fuel ratio sensor detects the air-fuel ratio. Then, the operating state determination means detects a physical quantity representing the state of the engine, and based on this physical quantity,
Determining whether the engine is in a predetermined operating state, and when the response time detecting means determines that the engine is in the predetermined operating state, changing at least one physical quantity detected by the state detecting means; The response time from when the physical quantity is changed to when the air-fuel ratio detected by the air-fuel ratio sensor changes is measured and stored. Further, the plurality of physical quantities detected by the air-fuel ratio estimating means and the stored response time are input as parameters, the air-fuel ratio is estimated using a neural network, and the fuel correction amount calculating means calculates the fuel based on the estimated air-fuel ratio. The correction amount of the injection amount is calculated.

With such an operation, the response time of the air-fuel ratio sensor is detected each time the engine enters a predetermined operating state, and the air-fuel ratio is estimated by a neural network taking this response time into account. Accordingly, even if the response time is delayed due to deterioration of the air-fuel ratio sensor, an appropriate air-fuel ratio is output, and a correct fuel correction amount can be calculated based on the air-fuel ratio.

The neural network used for the air-fuel ratio estimating means is subjected to neuro-learning with respect to the learning data obtained using the deteriorated sensor using the output value of the air-fuel ratio sensor without deterioration as a teacher signal. By using the result, the air-fuel ratio is estimated by adjusting the delay of the response time due to the deterioration of the air-fuel ratio sensor, and appropriate air-fuel ratio control can be performed.

Then, in the air-fuel ratio control device according to the present invention having the state detecting means, the fuel type determining means, the operating state determining means, and the fuel correction amount calculating means, the state detecting means indicates the state of the engine. A plurality of physical quantities are detected, and the fuel type determining means inputs the detected plurality of physical quantities as parameters, and determines the type of fuel using a neural network. Then, the operating state determination means detects a physical quantity representing the state of the engine, determines whether or not the engine is in a predetermined state based on the physical quantity, and determines whether the fuel correction amount calculation means has the engine in the predetermined state. The correction amount of the fuel injection amount is calculated based on the fuel type determined by the fuel type determination unit.

By such an operation, the type of fuel used for the engine is determined by the neural network, and the correction amount of the fuel injection amount when the engine is in a predetermined state is calculated according to the fuel type. . Therefore,
Appropriate air-fuel ratio control can be performed according to the nature of the fuel selected by the vehicle user. Then, the state detecting means detects the number of cranks from energization of the spark plug to a complete explosion of the air-fuel mixture in the engine as a physical quantity representing the state of the engine. If the energization of the spark plug is detected as a physical quantity to be expressed, and whether or not the engine is in a starting state is determined based on this physical quantity, the energization of the ignition plug, which has a strong correlation with the type of fuel, can be performed within the engine. Estimation accuracy can be improved by using the number of cranks until the air-fuel mixture completely explodes as a parameter of the neural network. The fuel injection amount can be adjusted.

Further, if the state detecting means detects the battery voltage of the engine as the physical quantity representing the state of the engine, the neural network can control the fuel voltage in consideration of the battery voltage affecting the number of cranks. Since the type is detected, more accurate estimation of the fuel type can be performed. Then, the operating state determining means detects the engine temperature as the physical quantity representing the state of the engine by the engine temperature detecting unit, and the engine temperature determining unit determines that the engine temperature is equal to or lower than the first set temperature. If the fuel correction amount calculating means calculates the correction amount of the fuel injection amount in consideration of the temperature of the engine detected by the engine temperature detecting section, the first By appropriately setting the set temperature, in the temperature range where the air-fuel ratio sensor does not work and feedback control cannot be performed, appropriate air-fuel ratio control can be performed based on the fuel type and engine temperature. The harmful gas in can be reduced.

Further, if a determination permission section is provided which permits the determination by the fuel type determination means only when the temperature of the engine is equal to or lower than the predetermined second set temperature, the determination permission section allows the determination permission section to determine whether the engine temperature is higher than the second predetermined temperature. Since the determination by the fuel type determination means is permitted only when the temperature is equal to or lower than the set temperature, if the second set temperature is appropriately determined, the fuel type determination by the neural network is performed only in a temperature range in which the difference in fuel properties depending on the fuel type is remarkable. Is performed, the determination accuracy can be improved, and since the temperature region is limited, the amount of learning data in the learning process of the neural network can be reduced, and the development time can be reduced. Become.

The air-fuel ratio control device further includes a fuel type storage means for storing the fuel type determined by the fuel type determination means each time the fuel type is updated, and the fuel correction amount calculation means includes a function for starting the engine. At the time, based on the fuel type stored in the fuel type storage means,
By calculating the correction amount of the fuel injection amount, the first fuel injection amount for starting the engine from the stopped state can also be adjusted according to the type of fuel, and the air-fuel ratio at the time of starting can be appropriately controlled. it can.

Then, the state detecting means is made to detect at least the fuel injection amount and the air-fuel ratio as physical quantities.
If constituted by the fuel characteristic calculation section and the type determination section, the inflow air weight estimation section inputs a plurality of physical quantities detected by the state detection means as parameters, and estimates the cylinder inflow air weight by a neural network. ,
The fuel characteristic calculation unit calculates the fuel evaporation rate and / or the fuel adhesion rate based on the fuel injection amount and the air-fuel ratio detected by the state detection unit and the cylinder inflow air weight estimated by the inflow air weight estimation unit. The type determination unit determines the type of the fuel from the evaporation rate of the fuel and / or the adhesion rate of the fuel calculated by the fuel calculation unit.

As described above, when the inflow air weight is estimated by the neural network, it is possible to more easily determine where the problem exists even if the accuracy cannot be obtained at the time of learning, so that the number of development steps is reduced. Can reduce development costs and development time. Further, the fuel characteristic calculation unit calculates both the evaporation rate of the fuel and the adhesion rate of the fuel, and further,
If a calculation value storage unit is provided, and the type determination unit determines the type of fuel from the evaporation rate and the fuel adhesion rate of the plurality of fuels stored in the calculation value storage unit, based on the plurality of calculation results, Since the fuel type is determined, the probability of making an erroneous determination can be reduced, and more accurate air-fuel ratio control can be performed.

In this air-fuel ratio control device, when the fuel type determining means is composed of an air-fuel ratio estimating unit, an air-fuel ratio sensor, and a fuel type estimating unit, the air-fuel ratio estimating unit determines the reference type of fuel. Using a neural network learned by inputting the physical quantity detected by the state detection unit as a parameter when the engine is operated by using the parameter and inputting the physical quantity detected by the state detection unit as a parameter, and estimating the air-fuel ratio On the other hand, the air-fuel ratio sensor detects the air-fuel ratio. Then, the fuel type estimating unit estimates the fuel type using the air-fuel ratio estimated by the air-fuel ratio estimating unit and the air-fuel ratio detected by the air-fuel ratio sensor. The fuel type can also be estimated by such an operation.

Then, in the vibration correction control device having the convergence degree estimating means and the vibration correcting means according to the present invention, the convergence degree estimating unit estimates the parameters related to the control target by using a neural network and obtains the correction amount. The calculating means calculates the correction amount of the input parameter using the estimated degree of convergence so that the control target converges quickly.

In a control system in which an input parameter causes a vibration component to be added to a control target by such an operation,
When a vibration component occurs, it can be quickly converged. Further, in the air-fuel ratio auxiliary control device according to the present invention including the state detecting means, the misfire degree estimating means, and the fuel correction amount calculating means, the state detecting means detects a plurality of parameters representing the state of the engine and detects the misfire degree. The estimating means estimates the degree of engine misfire using a neural network with the plurality of parameters detected by the state detecting means as input,
The correction amount calculating means calculates a correction amount of the fuel injection amount from the misfire degree obtained by the misfire degree estimating means.

By correcting the fuel injection amount in accordance with the estimated misfire degree, it is possible to prevent engine misfire and reduce harmful exhaust gas due to misfire. Further, since knocking of the engine can be prevented, the effect of improving the riding comfort of the car is also achieved. In such a configuration, when the starting state determining means and the fuel correction amount calculation permitting means are further provided, the starting state determining means determines the starting state of the engine from at least one of the plurality of parameters detected by the state detecting means. Is determined, and the fuel correction amount calculation permitting means causes the fuel correction amount calculation means to calculate the correction amount of the fuel injection amount only when the starting state determination means determines that the engine is in the starting state. To give permission.

That is, it is particularly easy to cause a misfire, and it is not necessary to control the misfire in an operating state where the misfire is not so much a problem. It is sufficient if the learning of the misfire degree is performed only at the time of starting, and the development process can be reduced. Then, the neural network used in the misfire degree estimating means sets the misfire degree to 0 when the rate of change of the air-fuel ratio sensor output is negative during the learning process until the engine changes from the starting state to the steady state.
When learning is performed using teacher data including the following information, it is possible to obtain a neural network that appropriately estimates a state in which no misfire has occurred.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration of an entire control device according to an embodiment.

FIG. 2 is a diagram illustrating a hardware configuration of an engine control unit.

FIG. 3 is a functional block diagram illustrating a configuration of an engine control unit.

FIG. 4 is a functional block diagram of the air-fuel ratio control device according to the first embodiment.

FIG. 5 is a schematic diagram showing a learning process of a neural network of an air-fuel ratio estimating unit.

FIG. 6A is a diagram illustrating a result of air-fuel ratio control by feedback control during a transition, and FIG. 6B is a diagram illustrating a result of air-fuel ratio control by the air-fuel ratio control device according to the first embodiment during a transition; FIG.

FIG. 7 is a schematic diagram showing a learning process of a neural network of an air-fuel ratio estimating unit in consideration of a misfire.

FIG. 8 is a diagram showing an example of a change in an air-fuel ratio at the time of a misfire.

FIG. 9 is a diagram showing another example of a change in the air-fuel ratio at the time of misfire.

FIG. 10 is a functional block diagram illustrating an air-fuel ratio control device according to a second embodiment.

FIG. 11 is a flowchart showing an operation of the air-fuel ratio control device according to the second embodiment.

12A is a diagram comparing an air-fuel ratio estimated based on input data outside a learning range with an actual air-fuel ratio, FIG.
(B) is a diagram comparing the air-fuel ratio estimated after converting the input data outside the learning range with the actual air-fuel ratio.

13A is a diagram comparing the estimated air-fuel ratio when estimating the air-fuel ratio with a neural network having an air-fuel ratio sensor input and teacher data, and FIG. 13B is a diagram illustrating a neural network without an air-fuel ratio sensor input; It is the figure which compared the estimated air-fuel ratio in the case of estimating an air-fuel ratio, and teacher data.

14A is a diagram illustrating a configuration of a neural network having an air-fuel ratio sensor input, and FIG. 14B is a diagram illustrating a configuration of a neural network having no air-fuel ratio sensor input.

FIG. 15 is a functional block diagram showing an air-fuel ratio control device according to a third embodiment.

FIG. 16 is a functional block diagram showing another example of the air-fuel ratio control device according to the third embodiment.

FIG. 17 is a functional block diagram showing another example of the air-fuel ratio control device according to the third embodiment.

FIG. 18 is a functional block diagram illustrating an air-fuel ratio control device according to a fourth embodiment.

FIG. 19 is a schematic diagram showing a learning process of a neural network of a change amount estimating unit.

FIG. 20 is a functional block diagram illustrating an air-fuel ratio control device according to a fifth embodiment.

FIG. 21 is a schematic diagram showing a learning process of a neural network of an inflow air weight estimation unit.

FIG. 22 is a functional block diagram showing an air-fuel ratio control device including an abnormality / deterioration detection device according to Embodiment 6.

FIG. 23A is a diagram illustrating a configuration of a sensor characteristic determination unit, and FIG. 23B is a diagram illustrating a state where characteristics of an estimated air-fuel ratio and a detected air-fuel ratio are different.

24A is a diagram showing a result of performing air-fuel ratio control by feedback control in a state where the dynamic characteristic of the air-fuel ratio sensor changes, and FIG. 24B is a diagram showing feedback according to a change in the dynamic characteristic of the air-fuel ratio sensor. FIG. 9 is a diagram illustrating a result of performing air-fuel ratio control by feedback control while changing a gain.

FIG. 25 is a functional block diagram showing another example of the air-fuel ratio control device including the abnormality / deterioration detection device according to the sixth embodiment.

FIG. 26 is a functional block diagram showing an air-fuel ratio control device including the abnormality / deterioration detection device according to the seventh embodiment.

FIG. 27 is a diagram showing detection positions of an air-fuel ratio before a catalyst and an air-fuel ratio after a catalyst.

FIG. 28 is a functional block diagram showing an air-fuel ratio control device according to an eighth embodiment.

FIG. 29 is a schematic diagram showing a learning process of the neural network of the air-fuel ratio estimation unit.

FIG. 30 is a flowchart showing an operation of the air-fuel ratio control device according to the eighth embodiment.

FIG. 31A is a diagram showing a result of air-fuel ratio control by a conventional air-fuel ratio control device, and FIG. 31B is a diagram showing air-fuel ratio control by an air-fuel ratio control device by an air-fuel ratio control device according to Embodiment 8; It is a figure showing a result.

FIG. 32 is a functional block diagram showing an air-fuel ratio control device according to a ninth embodiment.

FIG. 33 is a schematic diagram showing a learning process of the neural network of the fuel type determination unit.

FIG. 34 is a diagram showing another configuration example of the fuel type determination unit in the ninth embodiment.

FIG. 35 is a diagram showing a map used in a fuel correction amount calculation unit according to the ninth embodiment.

FIG. 36 is a flowchart showing an operation of the air-fuel ratio control device according to the ninth embodiment.

FIG. 37 is a functional block diagram illustrating an air-fuel ratio control device according to a ninth embodiment provided with a determination permission unit.

FIG. 38 is a flowchart illustrating an operation up to storage of a fuel type of the air-fuel ratio control device according to the ninth embodiment having a determination permission unit;

FIG. 39 is a functional block diagram showing an air-fuel ratio control device according to a ninth embodiment provided with a crank number detection unit and the like.

FIG. 40 is a schematic view showing an adhesion behavior model of injected fuel.

FIG. 41 is a diagram showing another example of the air-fuel ratio control device according to the ninth embodiment.

FIG. 42 (I) is a diagram showing a part of a table used for a fuel type determining unit, and (II) is a diagram showing another part of a table used for a fuel type determining unit.

FIG. 43 is a functional block diagram showing an air-fuel ratio control device according to Embodiment 10.

FIG. 44 is a schematic diagram showing a learning process of the neural network of the air-fuel ratio estimation unit.

FIG. 45 (a) is a diagram showing a change in the air-fuel ratio when no misfire occurs, and FIG. 45 (b) is a diagram showing a change in the air-fuel ratio when a misfire occurs.

FIG. 46 is a block diagram showing a vibration correction control device according to an eleventh embodiment.

FIG. 47 is a block diagram showing an air-fuel ratio control device according to Embodiment 11.

FIG. 48 is a diagram illustrating a learning process of the neural network of the misfire degree estimation unit.

FIG. 49 is a flowchart showing an operation of the air-fuel ratio control device according to the embodiment.

FIG. 50 is a block diagram showing a configuration of a conventional air-fuel ratio control device.

[Explanation of symbols]

 Reference Signs List 1 basic fuel injection amount calculation unit 2 air-fuel ratio auxiliary control calculation unit 3 A / D conversion unit 4 time series data storage unit 10 state detection unit 10a air-fuel ratio detection unit 20, 20a, 20b, 20c air-fuel ratio estimation unit 21 change amount estimation Unit 22 Inflow air weight estimation unit 23, 24 Fuel type determination unit 24a Inflow air weight estimation unit 24b Fuel characteristic calculation unit 24c Type determination unit 25 Misfire degree estimation unit 30, 31, 32, 33, 34 Fuel correction amount calculation unit 40a Parameter Range determining unit 40b Parameter converting unit 50, 51, 52 Transient state detecting unit 52a Air-fuel ratio sensor 52b Transient state calculating unit 60 Fuel correction amount adjusting unit 70, 71 Sensor characteristic determining unit 70a Extreme value detecting unit 70b Extreme value comparing unit 72 Catalyst Deterioration detection unit 80 Air-fuel ratio sensor 90a Operating state determination unit 90b Response time detection unit 100 Fuel type storage unit 110 Engine temperature detection unit 111 Discrimination permission unit 112 Crank number detection unit 113 Voltage detection unit 114 Plug voltage detection unit 120 Fuel type detection unit 130 Start state determination unit 140 Fuel correction amount calculation permission unit 310 Convergence degree estimation unit 320 Correction amount calculation unit 201 CPU 202 ROM 203 RAM 204 D / A converter C Engine control unit E Engine Z Control target

────────────────────────────────────────────────── ─── Continued on the front page (51) Int.Cl. ⁶ Identification code FI F02D 41/34 F02D 41/34 W 45/00 368 45/00 368Z (72) Inventor Norihiro Fujioka Amishima East Four in Kohoku-ku, Yokohama-shi Matsushita Communication Industrial Co., Ltd.

Claims (34)

[Claims] 1. An air-fuel ratio control device that assists control of an air-fuel ratio by correcting a fuel injection amount to a control system that maintains an air-fuel ratio at a constant value. State detecting means for detecting a plurality of physical quantities, air-fuel ratio estimating means for estimating an air-fuel ratio using a neural network, and detecting a fuel injection amount from the estimated air-fuel ratio. An air-fuel ratio control device having fuel correction amount calculating means for calculating a correction amount. 2. The air-fuel ratio control device according to claim 1, further comprising: fuel time-series data storage means for storing time-series data of a fuel injection amount; An air-fuel ratio controller including the time series data of the quantity. 3. The air-fuel ratio control device according to claim 1, wherein a value of at least one parameter input to the air-fuel ratio estimating means is within a range set in advance for the parameter. An air-fuel ratio control apparatus comprising: a parameter range determining unit that determines whether the parameter value is not within a predetermined range; and a parameter converting unit that replaces a parameter value determined not to be within a preset range with a preset value. 4. A range preset for the parameter is determined based on a maximum value and a minimum value of a value of a physical quantity serving as the parameter input in learning of the neural network. 3. The air-fuel ratio control device according to 3. 5. The air-fuel ratio control device according to claim 1, further comprising: a transient state detecting means for detecting a transient state amount of the engine; and An air-fuel ratio control device provided with fuel correction amount adjusting means for adjusting the correction amount of the fuel injection amount obtained by the correction amount calculating means. 6. An air-fuel ratio control apparatus according to claim 5, wherein said transient state detecting means detects a transient state quantity based on at least one physical quantity change obtained by said state detecting means. 7. The air-fuel ratio control device according to claim 5, wherein said transient state detecting means detects a transient state quantity based on an air-fuel ratio change amount estimated by said air-fuel ratio estimating means. 8. The transient state detecting means includes: an air-fuel ratio sensor for detecting an air-fuel ratio; and a transient state calculating means for detecting a transient state amount based on a change amount of an output value of the air-fuel ratio sensor. 6. The air-fuel ratio control device according to claim 5. 9. The neural network used in the air-fuel ratio estimating means may be configured such that when the output of the air-fuel ratio sensor is in a super-lean state during the learning process and is inconsistent with the fuel injection amount, the air-fuel ratio is in a super-rich state. 9. Learning is performed using teacher data including information indicating that
An air-fuel ratio control device according to any one of the preceding claims. 10. The air-fuel ratio control device according to claim 1, further comprising a fuel type detection unit that detects a fuel type used in the engine, wherein the air-fuel ratio estimation unit includes: An air-fuel ratio control device including a parameter corresponding to a property of a detected fuel type as a parameter to be input. 11. An air-fuel ratio control device for assisting control of an air-fuel ratio by correcting a fuel injection amount to a control system for maintaining an air-fuel ratio at a constant value, comprising: State detection means for detecting a plurality of physical quantities representing a state, at least two or more of the detected plurality of physical quantities are input as parameters, and inflow air weight estimation means for estimating cylinder inflow air weight using a neural network; An air-fuel ratio having a fuel correction amount calculating means for calculating a correction amount of the fuel injection amount from the fuel injection amount and the air-fuel ratio detected by the state detecting means and the cylinder inflow air weight estimated by the inflow air weight estimating means; Control device. 12. An air-fuel ratio control device for assisting control of an air-fuel ratio by correcting a fuel injection amount to a control system for maintaining an air-fuel ratio at a constant value, which can be detected even at a low temperature indicating an engine state. State detecting means for detecting a plurality of physical quantities, and change quantity estimating means for inputting the detected plurality of physical quantities as parameters and estimating a change quantity of a physical quantity related to an air-fuel ratio using a neural network. An air-fuel ratio control device comprising: a fuel correction amount calculating unit that calculates a correction amount of a fuel injection amount from a change amount of a physical amount related to an air-fuel ratio. 13. The change amount estimating means inputs a plurality of detected physical amounts as parameters, estimates a physical amount related to the air-fuel ratio using a neural network, and calculates a change amount of the physical amount to calculate the amount of change in the physical amount. 13. The air-fuel ratio control device according to claim 12, wherein a change amount of a parameter related to a fuel ratio is estimated. 14. The air-fuel ratio control device according to claim 12, wherein the physical quantity related to the air-fuel ratio is the air-fuel ratio itself. 15. The neural network used in the change amount estimating means is configured such that the air-fuel ratio is in a super-rich state when the output of the air-fuel ratio sensor is in a super-lean state during the learning process and is not consistent with the fuel injection amount. The learning is performed by using teacher data including information that exists.
15. The air-fuel ratio control device according to any one of 14. 16. An abnormality or deterioration of an air-fuel ratio control system comprising a control system for maintaining the air-fuel ratio at a constant value and a control system for assisting the control of the air-fuel ratio by correcting the fuel injection amount for the control system. An abnormality / deterioration detection device that detects a plurality of physical quantities that can be detected even at a low temperature indicating the state of the engine, and inputs the detected plurality of physical quantities as parameters and uses a neural network. An air-fuel ratio estimating means for estimating the air-fuel ratio, an air-fuel ratio sensor detecting the air-fuel ratio, and comparing the air-fuel ratio estimated by the air-fuel ratio estimating means with the air-fuel ratio detected by the air-fuel ratio sensor. An abnormality / deterioration detecting device comprising: a sensor characteristic determining unit configured to detect a change in dynamic characteristics of the air-fuel ratio sensor. 17. The method according to claim 17, wherein the sensor characteristic determining means differentiates the time change of the air-fuel ratio estimated by the air-fuel ratio estimating means and the time change of the air-fuel ratio detected by the air-fuel ratio sensor. 17. The abnormality / deterioration detection device according to claim 16, wherein the phase delay and the gain change of the air-fuel ratio sensor are obtained as the dynamic characteristic change of the air-fuel ratio sensor by comparing the time and the value of the respective extreme values. 18. The sensor characteristic determining means uses at least one physical quantity detected by the state detecting means, an air-fuel ratio estimated by the air-fuel ratio estimating means, and an air-fuel ratio detected by the air-fuel ratio sensor as parameters. type in,
17. The abnormality / deterioration detection device according to claim 16, wherein a change in dynamic characteristics of the air-fuel ratio sensor is estimated using a neural network. 19. The air-fuel ratio estimating means estimates an air-fuel ratio before a catalyst in an exhaust pipe, the air-fuel ratio sensor detects an air-fuel ratio after a catalyst in an exhaust pipe, and the sensor characteristic determining means 17. The abnormality / deterioration according to claim 16, wherein said air / fuel ratio estimated by said air / fuel ratio estimating means is compared with an air / fuel ratio detected by said air / fuel ratio sensor to detect catalyst deterioration. Detection device. 20. An air-fuel ratio control device for assisting control of an air-fuel ratio by correcting a fuel injection amount to a control system for maintaining the air-fuel ratio at a constant value, which can be detected even at a low temperature indicating an engine state. State detecting means for detecting a plurality of physical quantities, an air-fuel ratio sensor for detecting an air-fuel ratio, and a physical quantity representing the state of the engine, and determining whether the engine is in a predetermined operating state based on the physical quantity. Operating state determining means for performing, when it is determined that the engine is in a predetermined operating state,
Response time detecting means for changing at least one physical quantity detected by the state detecting means, measuring and storing a response time from a change time of the physical quantity to a change in an air-fuel ratio detected by the air-fuel ratio sensor, An air-fuel ratio estimator for estimating an air-fuel ratio using a neural network by inputting a plurality of detected physical quantities and a stored response time as parameters, and calculating a correction amount of the fuel injection amount from the estimated air-fuel ratio An air-fuel ratio control device having a fuel correction amount calculating means. 21. A neural network used in the air-fuel ratio estimating means performs a neural learning on learning data obtained by using a deteriorated sensor using an output value of an air-fuel ratio sensor without deterioration as a teacher signal. The air-fuel ratio control device according to claim 20, which has been performed. 22. An air-fuel ratio control device for assisting control of an air-fuel ratio by correcting a fuel injection amount to a control system for maintaining an air-fuel ratio at a constant value, wherein a plurality of physical quantities representing an engine state are detected. State detecting means, a plurality of detected physical quantities are inputted as parameters, a fuel type determining means for determining a fuel type using a neural network, and a physical quantity representing an engine state are detected, and based on the physical quantities, An operating state determining means for determining whether the engine is in a predetermined state; and, when the engine is in the predetermined state, calculating a correction amount of the fuel injection amount based on the fuel type determined by the fuel type determining means. An air-fuel ratio control device having a fuel correction amount calculating means. 23. The state detection means detects the number of cranks from energization of a spark plug to a complete explosion of the air-fuel mixture in the engine as a physical quantity representing the state of the engine. 23. The air-fuel ratio control device according to claim 22, wherein the energization of the spark plug is detected as a physical quantity representing the state of (i), and whether the engine is in a starting state is determined based on the physical quantity. 24. The air-fuel ratio control device according to claim 22, wherein the state detection means detects a battery voltage of the engine as the physical quantity representing the state of the engine. 25. An engine temperature detecting section for detecting an engine temperature as the physical quantity representing an engine state, wherein the operating state determining means determines whether or not the engine temperature is equal to or lower than a first set temperature. An engine temperature determining unit that determines the fuel injection amount, wherein the fuel correction amount calculating unit calculates a correction amount of the fuel injection amount in consideration of an engine temperature detected by the engine temperature detecting unit. An air-fuel ratio control device according to any one of the preceding claims. 26. The air-fuel ratio control device according to claim 22, further comprising: permitting the determination by the fuel type determination unit only when the engine temperature is equal to or lower than a predetermined second set temperature. An air-fuel ratio control device provided with a determination permitting unit. 27. The air-fuel ratio control device according to claim 22, further comprising a fuel type storage unit that stores a fuel type determined by the fuel type determination unit each time the fuel type is updated. An air-fuel ratio control device, wherein the fuel correction amount calculating means calculates a correction amount of the fuel injection amount based on the fuel type stored in the fuel type storage means at the time of starting the engine. 28. The state detecting means detects at least a fuel injection amount and an air-fuel ratio as physical quantities, and the fuel type determining means inputs a plurality of detected physical quantities as parameters, and calculates a weight of air flowing into a cylinder by a neural network. And the fuel injection amount and air-fuel ratio detected by the state detection means, and the cylinder inflow air weight estimated by the inflow air weight estimation unit, and the fuel evaporation rate or the fuel adhesion rate or 28. The fuel cell system according to claim 22, further comprising: a fuel characteristic calculating unit that calculates both of them; and a type determining unit that determines the type of fuel based on the fuel evaporation rate or the fuel adhesion rate calculated by the fuel calculating unit or both. The air-fuel ratio control device according to claim 1. 29. The air-fuel ratio control device according to claim 28, wherein the fuel characteristic calculation unit calculates both a fuel evaporation rate and a fuel adhesion rate, and further calculates the calculated fuel evaporation rate and fuel adhesion rate. An air-fuel ratio control device comprising: a calculated value storage unit that sequentially stores the fuel type, wherein the type determination unit determines the type of fuel from the evaporation rates and the adhesion rates of the plurality of fuels stored in the calculated value storage unit. 30. The fuel type determination means, using a neural network learned by inputting and learning, as a parameter, a physical quantity detected by the state detection unit when an engine is operated using a reference type fuel. An air-fuel ratio estimator that estimates an air-fuel ratio by inputting a physical quantity detected by the state detector as a parameter, an air-fuel ratio sensor that detects an air-fuel ratio, an air-fuel ratio estimated by the air-fuel ratio estimator, The air-fuel ratio control device according to any one of claims 22 to 27, further comprising a fuel type estimation unit that estimates a fuel type using an air-fuel ratio detected by the air-fuel ratio sensor. 31. A control device for correcting a vibration component in a control system in which an input parameter causes a vibration component to be added to the control target, wherein a degree of convergence of the control target is determined by using a physical quantity related to the control target as a parameter. A vibration correction control device comprising: a convergence degree estimating means for inputting and estimating by a neural network; and a correction amount calculating means for calculating a correction amount of a parameter to be input so that a controlled object converges quickly using the estimated convergence degree. . 32. An air-fuel ratio control device for assisting control of an air-fuel ratio by correcting a fuel injection amount to a control system for maintaining the air-fuel ratio at a constant value, wherein a plurality of physical quantities representing an engine state are detected. State detection means, a plurality of detected physical quantities are inputted as parameters, and a misfire degree estimation means for estimating the degree of engine misfire using a neural network; and a fuel injection amount based on the misfire degree obtained by the misfire degree estimation means. An air-fuel ratio control device having a fuel correction amount calculating means for calculating a correction amount of the fuel. 33. The air-fuel ratio control device according to claim 32, further comprising: a starting state determining unit that determines whether the engine is in a starting state based on at least one of the plurality of physical quantities detected by the state detecting unit. Air-fuel ratio control comprising: fuel correction amount calculation permitting means for permitting the fuel correction amount calculating means to calculate the correction amount of the fuel injection amount only when the engine is determined to be in the starting state by the starting state determining means. apparatus. 34. The neural network used in the misfire degree estimating means determines the misfire degree when the rate of change of the output of the air-fuel ratio sensor is negative during the learning process from the start of the engine to the steady state. The air-fuel ratio control device according to claim 32 or 33, wherein the air-fuel ratio control device is learned using teacher data including information to be set to 0.