JPH11324782A - Learning method in air-fuel ratio control device of internal combustion engine - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a learning method cupuble of simply and accurate by initializing by acquiring the middle value of plural teaching data as teaching data of its corresponding operating condition points. SOLUTION: The whole area of operating condition parameters forming a model is divided into plural divided operation areas, and the operating condition is changed in transient between two operating conditions a1, a2 dimarcating the respective deviled operation areas. From the amount of shift of an exhaust air fuel ratio from a target air fuel ratio caused at this time, teaching data for correcting output parameters of the model in two operating conditions a1, a2 changed to be transient is acquired. In the case where two teaching data are acquired for one operating condition, the average value is taken as teaching data to perform initial learning of the model. Thus, it is possible to obtain a model such that output parameter is smoothly changed in all areas of operating condition parameters without complicated operation.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an air-fuel ratio control apparatus for an internal combustion engine which models at least the dynamic behavior of fuel injected from a fuel injection device as a fuel adhesion rate and controls the air-fuel ratio using the model. In the model learning method.

[0002]

2. Description of the Related Art In recent years, there has been a demand for high-precision fuel supply to internal combustion engines in order to respond to stricter exhaust gas regulations and to market demands for improved fuel efficiency and responsiveness. In addition, it has become mainstream to supply fuel to an internal combustion engine using a fuel injection device. But,
The fuel injection device has a problem that the amount of fuel adhering to the intake pipe and the intake valve is larger than that of the conventional carburetor, and the controllability of the air-fuel ratio during the transition is deteriorated. In order to solve such a problem, the applicant has disclosed in Japanese Patent Application No. 7-271188 a fuel attachment ratio in which the rate at which fuel injected from a fuel injector adheres to an intake pipe and an intake valve is defined as a fuel adhesion rate. The rate at which fuel attached to the intake valve evaporates is modeled using the feedforward control logic with a learning function as the evaporation time constant, and the amount of fuel entering the combustion chamber is estimated using this fuel system model. A method for calculating the amount of fuel to be injected based on the above has already been proposed.

[0003]

According to the control method described above, the fuel adhesion rate and the evaporation time constant can be estimated using the fuel system model, and the amount of fuel entering the combustion chamber during transition can be estimated in advance. Although the injection amount can be controlled to an appropriate amount, in the case of using the fuel system model, it is necessary to perform an initial learning for setting an optimum value of a fuel adhesion rate and an evaporation time constant for various operation states in advance. there were. Since the dynamic behavior of the fuel affects the air-fuel ratio when the operating state of the engine is in the transient state, the parameters of the dynamic behavior must be set while the operating state is transiently changed. The fuel deposition rate, one of the parameters of the dynamic behavior, affects the operating state before and after the change when the operating state is transiently changed, so the setting is difficult and it takes time. there were. The present invention provides a learning method in an air-fuel ratio control apparatus for an internal combustion engine, which can easily and accurately perform an initial setting in view of the above-mentioned problems relating to an initial learning in an air-fuel ratio control method using a fuel system model. It is intended to be.

[0004]

In order to achieve the above-mentioned object, a learning method in an air-fuel ratio control apparatus for an internal combustion engine according to the present invention comprises a method for learning at least the dynamic behavior of fuel injected from a fuel injection device. In an air-fuel ratio control device for an internal combustion engine, which models the ratio as an air condition and controls the exhaust air-fuel ratio using the model, the operating conditions are assigned to a plurality of regions, and operating conditions are defined between operating state points defining one region. Is transiently changed, and, based on the deviation of the exhaust air-fuel ratio from the target air-fuel ratio that occurs at that time, the teacher data of the fuel deposition rate at the two changed operating state points is obtained,
While learning the model based on the teacher data,
When a plurality of different teacher data for different regions are obtained for the same driving state point, an intermediate value of the plurality of teacher data is obtained as the teacher data of the corresponding driving state point. Things.

[0005]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of a learning method in an air-fuel ratio control apparatus for an internal combustion engine according to the present invention will be described below with reference to some embodiments shown in the accompanying drawings.
FIG. 1 is a schematic diagram showing the relationship between an engine 1 and a control device 10 capable of executing engine control employing a prediction learning method according to the present invention. In the engine 1, an air-fuel mixture is introduced into a combustion chamber of a cylinder 5 via an air cleaner 3 and a fuel injection device 4 provided in an intake pipe 2, and exhaust gas after combustion is exhausted to the atmosphere via an exhaust pipe 6. This is a four-cycle engine, and other components such as an intake valve and an exhaust valve are omitted in the drawing. In FIG. 1, reference numeral 7 denotes a crankcase, and reference numeral 8 denotes a throttle valve. The control device 10 controls the value of the air-fuel ratio in the exhaust gas by operating the amount of fuel injection from the fuel injection device 4. As shown in FIG. 1, the control device 10 includes information r regarding a crank angle obtained from a crank angle detecting means 13 provided in a crankcase 7 and a throttle opening detecting means 12 provided in a throttle valve 8.
, And information Te about the intake pipe wall temperature obtained from the intake pipe wall temperature detecting means 14 provided in the intake pipe 2, and based on the input information, the intake pipe The operation amount Mf (that is, the fuel injection amount) of the fuel injection device 4 provided in the exhaust pipe 6 is determined and output, and the detection signal E relating to the actual air-fuel ratio obtained from the air-fuel ratio sensor 15 provided in the exhaust pipe 6 is obtained. Is input as necessary, the output of the internal model is corrected based on this information, and the learning of the internal model is performed, so that optimal control can always be performed.

FIG. 2 is a schematic block diagram showing the configuration of the control device 10. The control device 10 includes a model-based control unit 20 that determines a basic fuel injection amount Mfn using a feedforward control logic along the target air-fuel ratio Ep, and an engine speed calculation that calculates an engine speed N from the crank angle information r. Unit 30, a target air-fuel ratio calculation unit 4 that calculates the target air-fuel ratio Ep based on a map or an arithmetic expression prepared in advance based on the engine speed N and the throttle opening degree
0, a conversion unit 50 that converts the basic fuel injection amount Mfn obtained by the model base control unit 20 into a fuel injection amount Mf as a pulse signal, and detects a deviation amount of the model base control unit 20 and corrects the deviation. And a model correction unit 60 for acquiring teacher data for performing the operation.

(Model Base Control Unit) The configuration of the model base control unit 20 will be described below with reference to FIGS. FIG. 3 is a schematic block diagram showing the configuration of the model base control unit 20 in FIG. The model-based control unit 20 includes an air system forward model 21 that models the behavior of air in the intake pipe 2, a fuel system forward model 22 that models the behavior of fuel injected from the fuel injection device 4, and each forward model. Estimated air amount Av output from 21 and 22
And an estimated air-fuel ratio calculation unit 23 for calculating an estimated air-fuel ratio Ev based on the estimated fuel amount Fv. Further, the model base control unit 20 includes a feedback loop for feeding back the estimated air-fuel ratio Ev output from the estimated air-fuel ratio calculation unit 23 to the basic operation amount calculation unit 24. The basic manipulated variable calculator 24 receives the estimated air-fuel ratio Ev output from the estimated air-fuel ratio calculator 23 and the target air-fuel ratio Ep output from the target air-fuel ratio calculator 30 and receives the fuel injection device of the engine 1. 4 basic operation amount Mfn (basic fuel injection amount)
Is calculated. The basic operation amount Mfn is output from the model base control unit 20 and also input to the fuel system forward model 22. The fuel system forward model 22 outputs the basic operation amount Mfn.
The estimated fuel amount Fv is obtained based on As mentioned above,
In the model base control unit 20, a forward model of the engine 1 is configured by the air system forward model 21, the fuel system forward model 22, and the estimated air-fuel ratio calculation unit 23, and the fuel system forward model 22, the estimated air-fuel ratio calculation unit 23 and a feedback loop including a basic manipulated variable operation unit 24 to constitute an inverse model of the engine that outputs the basic manipulated variable Mfn by feeding back the estimated air-fuel ratio Ev output from the forward model of the engine 1. .

The air system forward model 21 has a throttle opening α and an engine speed N.
Is input, and the intake air amount is estimated and calculated based on these. Specifically, the air amount is modeled by a hydrodynamic formula using the throttle opening and the intake negative pressure, and the intake negative pressure is modeled by a hydrodynamic formula based on the pressure and the volumetric efficiency calculated from the air amount. The pressure is modeled, the volumetric efficiency is modeled using a fuzzy neural network that takes the engine speed N and the throttle opening α as inputs, and the initial value of the intake negative pressure is set. And the estimated air amount Av can be calculated from the engine speed N. The fuzzy neural network constituting the volumetric efficiency model is configured to be able to learn to feed back the exhaust air-fuel ratio and correct the deviation between the exhaust air-fuel ratio and the target air-fuel ratio when the engine is in a steady state. Can be done. Further, the modeling of the volumetric efficiency is not limited to a fuzzy neural network, and any feedforward control logic that can be learned, such as a neural network, a CMAC, or a simple map, may be used.

FIG. 4 is a schematic block diagram showing the structure of the fuel system forward model 22.
FIG. 5 shows an image diagram of the fuel system forward model. As shown in FIG. 5, the fuel system forward model 22 determines the behavior of the fuel injected from the fuel injection device 4 by the ratio of the injected fuel adhering to the intake pipe wall or the intake valve without directly entering the cylinder. x (hereinafter referred to as a fuel adhesion rate x) and a speed τ (hereinafter, an evaporation time constant τ) at which fuel adhering to an intake pipe wall or an intake valve evaporates and is modeled. Based on the basic operation amount Mfn (basic fuel injection amount) input from 24 and the fuel adhesion rate, a fuel amount (1-x) Mf injected from the fuel injection device 4 and directly into the cylinder is estimated and calculated, Amount a · Mw in which the deposited fuel Mw evaporates and enters the cylinder based on the deposited fuel Mw obtained from the basic operation amount Mfn and the fuel deposition rate x and the evaporation time constant τ.
And the estimated fuel amount (1-x) M
It estimates the total amount of fuel entering the cylinder from f and a · Mw, and includes a fuel adhesion rate estimating unit 22a, an evaporation time constant estimating unit 22b, a non-adhering fuel calculating unit 22c, and an adhering fuel calculating unit 22.
d and an evaporative fuel amount calculation unit 22e. As shown in FIG. 6, the fuel adhesion rate estimating unit 22a comprises a map in which the fuel adhesion rate x is modeled with respect to the operating state of the engine (engine speed N and throttle opening α). And engine speed calculation unit 30
Then, the current engine speed N and throttle opening are input, and the fuel adhesion rate x is determined based on these.
The evaporating time constant estimating unit 22b is composed of a map that models the evaporating time constant τ with respect to the operating state of the engine (the engine speed N or the throttle opening α and the intake pipe wall temperature Te or the engine water temperature). The engine speed N or the throttle opening and the intake pipe wall temperature Te at that time are input from the degree detecting means 12 or the engine speed calculating section 30 and the intake pipe wall temperature detecting means 14, and based on these, the fuel adhesion rate x To determine. The non-adhered fuel calculating unit 22c is configured to calculate the basic operation amount calculating unit 2 based on the fuel adhering rate x obtained from the fuel adhering rate estimating unit 22a.
The fuel amount (1-x) Mf that directly enters the combustion chamber of the cylinder 5 from the fuel injection device 4 at the basic operation amount Mfn (that is, the basic fuel injection amount) input from Step 4 is calculated. The adhering fuel calculating unit 22d is configured to calculate the basic operation amount calculating unit 24 based on the fuel adhering rate x obtained from the fuel adhering rate estimating unit 22a.
Amount x · Mf adhering to the intake pipe wall surface and the intake valve in the basic operation amount Mfn (basic fuel injection amount) input from
(Mw) is calculated. The evaporative fuel amount calculation unit 22e includes:
The evaporation time constant τ estimated by the evaporation time constant estimation unit 22b
Is used to calculate the evaporated fuel amount a · Mw by integrating the fuel evaporated from the attached fuel obtained by the attached fuel calculation unit 22b. The fuel system forward model 22 calculates the estimated fuel amount Fv by adding the fuel amount calculated by the non-adhered fuel calculating unit 22c and the evaporated fuel amount calculated by the evaporated fuel amount calculating unit 22e. The maps constituting the above-described fuel adhesion rate estimating section 22a and the evaporation time constant estimating section 22b include, in advance, optimal fuels for various operating states so as to correspond to various operating states of the internal combustion engine. Adhesion rate x or evaporation time constant τ
Is acquired as basic data, and the internal combustion engine is actually driven using a map composed of the basic data after the initial learning. However, even if the map (model) is optimized in advance by the initial learning, the basic data and the dynamics of the actual internal combustion engine may be changed after the internal combustion engine is driven due to a change over time of the internal combustion engine or a change in the environment in which the internal combustion engine is used. Deviation from the dynamic behavior occurs. This deviation is corrected by online learning that corrects the deviation between the model and the actual dynamic behavior of the fuel in the internal combustion engine based on the teacher data acquired by a method described later. The acquisition of the teacher data for the initial learning and the teacher data for the online learning is performed by the model correction unit 60 described later in detail.

The estimated air-fuel ratio calculation unit 23 calculates an estimated air-fuel ratio Ev based on the estimated air amount Av and estimated fuel amount Fv obtained from the air system forward model 21 and the fuel system forward model 22. Then, the basic operation amount calculation unit 24 calculates a basic operation amount (that is, a basic fuel injection amount Mfn) based on the estimated air-fuel ratio Ev and the target air-fuel ratio Ep, and converts the basic operation amount into a conversion unit 50.
Output to

(Regarding the Model Correction Unit) Next, the fuel system forward model 22 in the control device configured as described above
The following describes the initial learning and online learning.
Fuel adhesion rate estimation unit 22 in the fuel system forward model 22
The teaching data for the initial learning or online learning of each model of the e and the evaporation time constant estimating unit 22b is obtained by the model correcting unit 60, and the estimating units 22a and 22b determine the operating state at that time based on the teaching data. Initial learning or online learning of the corresponding fuel adhesion rate x or evaporation time constant τ is performed.

(Initial Learning) Initial learning will be described with reference to the fuel adhesion rate estimating unit 22a as an example. FIG. 7 shows an initial state of a map constituting the fuel adhesion rate estimating unit 22a. The map constituting the fuel adhesion rate estimating unit 22a (or the evaporation time constant estimating unit 22b) includes an entire region (in FIG. 7, the engine speed and the throttle opening in FIG. 7) representing the operating state of the engine. The engine speed N: 0 rpm to 6000 rpm; the throttle opening α: 0 to 90 degrees) is divided into a plurality of parts in an appropriate operation state (in FIG. 7, the engine speed N is 10 in units of 600 rpm).
And the throttle opening is divided into nine in 10-degree units. ), A set of data configured to set an optimal output parameter, that is, a fuel adhesion rate x for each operation state that defines each divided operation region (hereinafter, referred to as a divided operation region). In the present specification, each operation state that defines the divided operation region is referred to as an operation state point. Therefore, for example, the divided operation region Z in FIG.
Throttle opening is 50 at 3600 rpm
Operating state point z1 in degrees and engine speed 3600r
Operating state point z2 with 60 degree throttle opening at pm
An operating state point z3 at which the engine rotation speed is 4200 rpm and the throttle opening is 50 degrees;
Operating state point z where throttle opening is 60 degrees at 0 rpm
4 will be defined. In the initial state, the fuel adhesion rate x in the above-described map is set to an appropriate constant value (20% in FIG. 7) at all operating state points. In the initial learning, from this state, one parameter (engine speed) of the operating state is fixed, and the other parameter (throttle opening) is transiently changed between two adjacent operating state points. The waveform of the exhaust air-fuel ratio is detected by the air-fuel ratio sensor 15, and based on the detected waveform of the exhaust air-fuel ratio, teacher data of output parameters at two transiently changed operating state points is obtained. Learning of output parameters is performed based on. Although not shown, the evaporating time constant estimating unit 22b also has a map of the evaporating time constant with respect to the operating state of the engine, similarly to the fuel adhering rate estimating unit 22a. It is divided into a plurality to correspond.

Hereinafter, the initial learning including acquisition and learning of teacher data will be described with a specific example.
First, two operation state points a1 that define the divided operation area a
In the following description, the acquisition of teacher data in steps a and a2 will be described as an example. A crank angle signal for initial learning is input to the control device 10 so that the engine speed calculated by the engine speed calculation unit 30 becomes 1000 rpm. The engine speed N at 10 is fixed to 1000 rpm including the operating state points a1 and a2. Then, a throttle opening signal (that is, a 0-degree signal) for initial learning relating to the operating state point a1 is input, and the throttle opening signal is transiently changed to the operating state point a2 (that is, a 10-degree signal). FIG.
In the initial state shown in (1), the output parameter value of the map is set to an appropriate initial value (20% in this embodiment) and is constant, so that the fuel adhesion rate estimating unit 22a in the fuel system forward model 22 outputs the initial value. Finally, the basic operation amount Mfn of the fuel injection device 4 is calculated by the model base control unit 20 based on this initial value, and the fuel is actually injected to operate the engine. In this state, if one of the operation state parameters (throttle opening) is transiently changed as described above, if the initial value (20%) of the output parameter matches the operation state points a1 and a2, Although the deviation of the exhaust air-fuel ratio from the target air-fuel ratio does not exceed the allowable value,
If not, an air-fuel ratio spike occurs in which the exhaust air-fuel ratio deviates from the target air-fuel ratio by more than an allowable value, as indicated by reference numeral S1 in FIG. When an air-fuel ratio spike occurs, the model correction unit 60 divides the air-fuel ratio spike into a characteristic strongly related to the fuel adhesion rate x and a characteristic strongly related to the evaporation time constant τ. It is configured to acquire teacher data of a model (that is, a map) in the fuel adhesion rate estimation unit 22a and the evaporation time constant estimation unit 22b. here,
Explaining the dynamic behavior of the modeled fuel, that is, the difference between the characteristics of the fuel adhesion rate x and the evaporation time constant τ, the fuel adhesion rate x is such that the fuel injected from the fuel injection device 4 does not directly enter the cylinder. Since it shows the rate of adhering to the intake pipe wall and the intake valve, the change appears with a good response as a change in the estimated fuel amount, but the evaporation time constant τ is the speed at which the adhering fuel evaporates gradually and enters the cylinder. Since the change appears in the estimated fuel amount, it takes a longer time than the fuel adhesion rate x, and the amount is small. Model correction unit 6
0 uses the above-mentioned difference in the characteristics of the fuel adhesion rate x and the evaporation time constant τ, when the operating state of the engine changes transiently and an exhaust air-fuel ratio spike occurs as shown in FIG. The exhaust air-fuel ratio spike is divided into a peak value P (that is, a value representing the model deviation with good responsiveness) and a deviation area S to the target air-fuel ratio (that is, a value representing the model deviation over time). The peak value P is used as the fuel adhesion rate estimation unit 22a.
Are obtained as parameters for acquiring the correction learning teacher data of the model constituting, and the deviation area S is acquired as the parameter for acquiring the correction learning teacher data of the model constituting the evaporation time constant estimating unit 22b. In particular,
As shown in the flowchart of FIG.
A value of 0 reads the waveform of the exhaust air-fuel ratio E obtained from the air-fuel ratio sensor 15 (step 1), and sets the difference between the target air-fuel ratio Ep and the exhaust air-fuel ratio E as a deviation ΔE (step 2). The deviation amount ΔE is set to a predetermined allowable value L.
Are monitored (step 3) and whether an air-fuel ratio spike is occurring (step 4), and the processing of steps 1 and 2 is repeated until the slip amount ΔE becomes larger than the allowable value L. If the deviation ΔE exceeds the allowable value L, it is determined that an air-fuel ratio spike has occurred, a spike flag is set (step 5), and the deviation ΔE is set as a peak value P (step 7). After setting the peak value P in step 6, the shift amount ΔE at that time is set as the shift area S (step 8), and the process returns to step 1 again. The above-described processing of setting the peak value P in step 7 is repeatedly performed until the deviation amount ΔE becomes smaller than the previously set peak value P (steps 6 and 7), whereby the true peak value P is finally obtained. Can be In addition, the displacement area S of the step 7
Is repeatedly performed until the deviation ΔE becomes smaller than a preset stability determination value (steps 9 and 8),
At this time, a value obtained by adding the shift amount ΔE at that time to the previous shift area S is newly set as the shift area S, so that the true shift area S is finally obtained. Note that the stability determination value is set smaller than an allowable value. When the deviation ΔE is smaller than the stability determination value, it is determined that the air-fuel ratio spike has disappeared (step 9), the spike flag is cleared (step 10), and the set peak value P and deviation area S
Is obtained (step 11). FIG.
Is a flowchart showing a process of acquiring teacher data;
As shown in this flowchart, the model correction unit 60
Reads the throttle opening α, the engine speed N, and the intake pipe wall temperature Te after clearing the spike flag (step a), and from the information, the peak value P, and the deviation area S, the fuel adhesion rate estimation unit 22a The teacher data xt for correcting the shift and the teacher data τt for correcting the shift of the evaporation time constant estimating unit 22b are calculated using a predetermined arithmetic expression (step b). Specifically, when the transient state of the engine occurs due to an increase in the throttle opening, the teacher data xt and τt are calculated using the following equations (1) and (2), and xt = x + Gx × (P−E) (1) τt = τ + Gτ × S (2) When the transient state occurs due to a decrease in throttle opening, the following equation (3) and Using (4), each teacher data xt
And τt are calculated. xt = x−Gx × (PE) (3) τt = τ−Gτ × S (4) In the above equations (1) to (4), Gx and Gτ are teachers, respectively. This is the gain when acquiring data, and Gx>
0, Gτ> 0. According to the above equation, specifically, for example, when the peak value is larger than the target value when the throttle opening is increasing, the fuel adhesion rate estimating unit 22
a, the fuel deposition rate is estimated to be smaller than the actual fuel deposition rate, so that the acquired teacher data xt
Is larger than the output value x of the fuel adhesion rate estimating unit 22a before learning. Conversely, if the peak value is smaller than the target value, the estimated fuel adhesion rate is larger than the actual fuel adhesion rate. Therefore, the acquired teacher data xT is
The value becomes smaller than the output value x of the fuel adhesion rate estimation unit 22a before learning. If the peak value is smaller than the target value when the throttle opening is decreasing, the fuel adhesion rate estimating unit 22a estimates the fuel adhesion rate to a value smaller than the actual fuel adhesion rate. , Acquired teacher data x
t is a value larger than the output value x of the fuel adhesion rate estimating unit 22a before learning, and conversely, if the peak value is larger than the target value, the estimated fuel adhesion rate is larger than the actual fuel adhesion rate. , The acquired teacher data xT
Is smaller than the output value x of the fuel adhesion rate estimation unit 22a before learning. Further, if a spike occurs in a direction larger than the target value while the throttle opening is increasing, the evaporation time constant estimating unit 22b estimates the evaporation time constant to a value smaller than the actual evaporation time constant. Therefore, the acquired teacher data τt becomes a value larger than the output value τ of the evaporation time constant estimating unit 22b before learning, and conversely, if a spike occurs below the target value, the estimated evaporation data Since the time constant is larger than the actual evaporation time constant, the acquired teacher data τT is smaller than the output value τ of the evaporation time constant estimator 22b before learning. Furthermore, if a spike occurs below the target value while the throttle opening is decreasing, the evaporation time constant estimating unit 22b estimates the evaporation time constant to a value smaller than the actual evaporation time constant. And the acquired teacher data τ
t is a value larger than the output value τ of the evaporation time constant estimating unit 22b before learning. Conversely, if the spike occurs in a direction larger than the target value, the estimated evaporation time constant is smaller than the actual evaporation time constant. Therefore, the acquired teacher data τT is smaller than the output value τ of the evaporation time constant estimator 22b before learning. As described above, the model correction unit 60 acquires the teacher data xt and τt when the throttle opening changes from the operating state point a1 to the operating state point a2. Referring back to FIGS. 7 and 8, the teacher data xt obtained by the model correction unit 60 as described above is calculated from the operating state point a1 (the engine speed 1000 rpm and the throttle opening 0 degree) to the operating state point a2. (Engine speed 1000 rpm and throttle opening 10 degrees), that is, teacher data x for the operating state a1 and the operating state a2 when the throttle is transiently opened.
acquired and stored as t (a1; a2) and τt (a1; a2). In the embodiment of FIG. 8, the mutual change between the operating state points, that is, the change from the operating state point a1 to the operating state point a2 (the change in the opening of the throttle opening) and the operating state point a2
From the operating state point a1 to the operating state point a1 (change of the throttle opening closing) and xt (a1; a2) and x
t (a2; a1), τt (a1; a2) and τt (a
2; a1) is obtained. Here, the influence of the fuel adhesion rate x and the evaporation time constant τ on the two transiently changed operating state points a1 and a2 will be briefly described. For example, the fuel adhesion rate x of the fuel adhesion rate estimating unit 22a is equal to the operating state point a2.
Has the same value as the actual fuel deposition rate, but when the operating state point a1 is smaller than the actual fuel deposition rate, the fuel deposition rate x at the operating state point a1 is Since the value is smaller than the fuel adhesion rate, the amount of evaporated fuel a · Mw is also smaller than the actual amount of evaporated fuel. In this state, the throttle is moved to the operating state point a1.
Is transiently changed from the operating state point a2 to the operating state point a2, even if the fuel adhering rate x at the operating state point a2 matches, the change in the amount of evaporative fuel is slow with a first-order lag. The model-based control unit increases the fuel injection amount (basic operation amount) so that the shortage of a · Mw is compensated for by the fuel amount directly entering the cylinder, and the fuel in the cylinder becomes larger than the actually required fuel amount. Then, the air-fuel ratio waveform becomes a rich spike at the operating state point 2. Thus, the operating state point a
The deviation of the estimated amount of the fuel adhesion rate x at 1 affects the air-fuel ratio waveform at the operating state point a2 when the throttle is transiently changed from the operating state point a1 to the operating state point a2. On the other hand, regarding the evaporation time constant, for example, when the throttle is transiently changed from the operation state point a1 to the operation state point a2, the estimated evaporation time constant τ at the operation state point a1 deviates from the actual evaporation time constant. If the adhesion amount and the evaporation amount are equal, the air-fuel ratio waveform is not affected, but if the estimated evaporation time constant τ at the operating state point a2 is shifted, the estimated evaporation time Since the response of the fuel amount changes and the fuel injection amount is determined in accordance therewith, the air-fuel ratio waveform is affected. Therefore, for the fuel adhesion rate estimating unit 22a, the teacher data xt (a1; a2) and xt (a2; a1) obtained by the mutual change between the two operating state points.
Is used as teacher data for each of the operating state points a1 and a2, and the evaporating time constant estimating unit 22b obtains teacher data τt (a1; a2) obtained when the operating state point a1 is changed to the operating state point a2. ) Is used as the teacher data of the operating state point a1, and the teacher data τt obtained when it is changed conversely.
(A2; a1) are acquired as teacher data of the operating state point a2. In the present embodiment, since the model is configured by a map, in learning each output parameter based on teacher data, the value of each teacher data is stored as it is as a new output parameter of the corresponding operating state point. The above-described processing is repeatedly performed until the deviation of the exhaust air-fuel ratio from the target air-fuel ratio becomes equal to or less than the allowable value when the operation state is changed between the operation state points a1 and a2. Obtain and update output parameters. When performing the initial learning, one operating state parameter is fixed, and the other operating state parameter is transiently changed between operating state points. It is preferable to change one after another between successive operating states and to set output parameters at each operating state point. Accordingly, after the setting of the output parameters of the operating state points a1 and a2 is completed, the throttle opening is transiently changed between the operating state points a2 and a3 that define the divided operating area b in FIG. Make settings. At this time, the output parameter value at the operating state point a2 has already been set.
If the processing is performed by setting the output parameter value at the operation state point a2, the air-fuel ratio can be stabilized early. In the above-described state, the same processing as the processing for setting the output parameter values of the operating state points a1 and a2 is performed, and the model correction unit 60 finally obtains the teacher data xt (a2; a3) and xt (a3;
a2), τt (a2; a3) and τt (a3; a2) are obtained. The evaporation time constant estimating unit 22b updates the acquired teaching data as it is as the teaching data of each of the operating state points a2 and a3.
In order to affect both the operating state points a2 and a3 before and after the change, two teaching data xt (a1; a2) + xt (a2; a1) for one operating state point a2,
xt (a2; a3) + xt (a3; a2) will be obtained. As described above, when the model correction unit 60 acquires two pieces of teacher data regarding different operation areas (for example, A and B) for one operation state point (for example, a2), the average value of the two pieces of teacher data is obtained. Is output as teacher data of the operating state point. Therefore, in this case, the throttle opening is transiently changed between the operating state points a2 and a3 until the deviation amount of the exhaust air-fuel ratio from the target air-fuel ratio becomes equal to or less than the allowable value. xt (a2; a3) + xt (a3: a2) is output as it is for the operating state point a3, but for the operating state point a2, the teacher data xt (a1; a2) + xt (a2; a1) and the newly obtained teacher data xt (a2; a3) + xt (a3; a
2) is output as teacher data XT. The process described above is performed for all the operating state points, and when the optimum output parameter value (fuel adhesion rate x) for each operating state point is obtained, the initial learning processing ends and FIG. A map as shown is obtained.

(On-line learning) When all initial learning is completed, the control device 10 is mounted on an engine and used. However, even if all the output parameters are set to the optimum values in the initial learning as described above, the basic data and the actual internal combustion engine may be changed after the internal combustion engine is driven due to a change over time of the internal combustion engine or a change in the environment in which the internal combustion engine is used. Deviation from the dynamic behavior in the engine occurs. Therefore, the model correction unit 60
While the control device 10 is mounted on the engine and the control device 10 is operating, the actual operating state parameters (engine speed and throttle opening degree) are determined based on the information from the detecting means 12, 13 and 14 and the air-fuel ratio sensor 15. ) And the exhaust air-fuel ratio are monitored, and when the operating state changes transiently and an air-fuel ratio spike occurs, the teaching data for the operating state point closest to each operating state before and after the transient change is described above. In the same manner as in the initial learning, and online correction learning is performed for each operating state point.

As described above, the entire operating condition parameter region constituting the model is divided into a plurality of divided operating regions, and the operating state is transiently changed between two operating state points defining each divided operating region. From the deviation amount of the exhaust air-fuel ratio with respect to the target air-fuel ratio generated at that time, the teacher data for correcting the output parameters of the model at the two operating state points that have been transiently changed is obtained. When two sets of teacher data are acquired, the initial value of the model is obtained by acquiring the average value of them as teacher data, so that the output parameters change smoothly in the entire range of the operating state parameters without complicated calculations. This has the effect of obtaining a model that performs Further, the acquisition of the teacher data at each operating state point is based on the characteristic of the deviation amount between the deviation amount of the exhaust air-fuel ratio with respect to the target air-fuel ratio from the allowable value to the stability determination value and the peak value and the characteristic. By dividing the integral model and using them to acquire teacher data for correcting the model deviation in the fuel adhesion rate estimating section 22a and the evaporation time constant estimating section 22b, the model can be used during use. It is possible to adapt to individual differences of the engine to be applied, changes in the use environment, changes over time, and the like, and to always perform air-fuel ratio control with an optimal model without complicated work.

The above-described division of the deviation amount characteristic is not limited to the present embodiment. The characteristic of the fuel adhesion rate, in which the change appears as a change in the estimated fuel amount with good responsiveness, and the change in the characteristic of the estimated fuel amount, Any method can be used as long as it can be divided into characteristics that take into account the characteristics of the evaporation time constant, which takes time to appear, as shown in FIGS. 12 (a) to 12 (c). The characteristic data (that is, the integral value of the first half) and the characteristic that appears late (that is, the integral value of the second half) are divided, and the teacher data of the fuel adhesion rate is converted into the integral value of the second half based on the integral value of the first half. Based on this, teacher data of the evaporation time constant may be obtained. FIG. 12A is a diagram showing another example of division of the characteristic of the shift amount. In this example, FIG.
The shift amount is divided into a first half and a second half with a peak value P as a boundary, and each teacher data is obtained based on the respective integral values S1 and S2. FIG. 12B is a diagram showing still another example of the division of the characteristic of the shift amount. In this example, the shift (spike) occurrence time is set to a predetermined ratio (for example,
It is divided into the first half and the second half by (1: 3), and each teacher data is obtained based on the respective integral values S1 and S2. FIG. 12C is a diagram showing still another example of the division of the characteristic of the shift amount. In this example, the shift amount is set in the first half after a predetermined time (for example, 100 ms) from the occurrence of the shift (spike). Part and the latter half part, and each integral value S1
And each teacher data is acquired based on and S2.

In the embodiment described above, the teacher data xt and τt are obtained by two types of arithmetic expressions corresponding to the increase and decrease of the throttle opening. However, this is not limited to this embodiment. Instead, for example, the differential value dα of the throttle opening may be calculated, and using the differential value dα, two types of equations for each teacher data may be unified as in the following equations (5) and (6). . xt = x + Kx × dα × (PE) (5) τt = τ + Kτ × dα × S (6)

In the above-described embodiment, the process of acquiring teacher data is described using an air-fuel ratio spike having only one peak value as an example. However, this is not limited to this embodiment. When an air-fuel ratio spike having a plurality of peak values occurs, an appropriate air-fuel ratio spike (for example, the first peak value or the maximum peak value) can be selected.

Further, in the above-described embodiment, the explanation has been made such that the correction value for the output parameter at each operation state point is acquired as it is as the teacher data. However, this is not limited to the present embodiment. As shown in FIG. 13, the correction range for the output parameter may be acquired as teacher data, and each output parameter may be updated. Further, in the above-described embodiment, for one operating state point,
It has been described that when two sets of teacher data relating to different divided driving regions are acquired, an average value thereof is acquired as teacher data. However, this is not limited to the present embodiment. The teacher data can be obtained from any value as long as it is an intermediate value of the data. Furthermore, in the present embodiment, the operating state of the engine is divided at equal intervals, but this may be arbitrarily divided without being limited to the present embodiment. Further, in the present embodiment, the dynamic behavior of the fuel is modeled using the map. However, the modeling method is not limited to the present embodiment, and the method using the feedforward control logic that can be learned is used. If so, any modeling method may be used. For example, a fuzzy neural network or a neural network may be used. When a fuzzy neural network is used, an output value of the optimal fuzzy neural network for each operating state point is acquired as teacher data, and based on the teacher data, each layer used in the fuzzy neural network is obtained. Initial learning or online learning of the appropriate coupling coefficient of.

[0020]

As described above, the initial learning method in the air-fuel ratio control apparatus for an internal combustion engine according to the present invention uses at least a feedforward control logic with a learning function to control a fuel injection device for a parameter relating to an engine operating state. Of an internal combustion engine that controls the exhaust air-fuel ratio of the engine to the target air-fuel ratio using a forward model of the internal combustion engine that has a fuel system forward model that models the dynamic behavior of the fuel injected from the engine as the fuel adhesion rate In the air-fuel ratio control device, the entire area of the parameters relating to the operating state is divided into a plurality of divided operating state areas, and the parameters relating to the operating state are transiently changed between operating state points that define one divided operating state area. Based on the deviation of the exhaust air-fuel ratio from the target air-fuel ratio, Obtain teacher data of the sticking rate, learn the fuel sticking rate and the evaporation time constant for each of the operating state points of the model based on the teacher data, and perform two learnings between different divided regions for one operating state point. When two different teacher data are acquired, the intermediate value of the two teacher data is acquired as the teacher data of the corresponding operating state point, so when the operating state is transiently changed, the driving before and after the change is obtained. The optimum value of the fuel adhesion rate affecting the state point can be obtained without complicated calculation, and the fuel adhesion rate that smoothly changes over the entire operating state can be obtained after the completion of the initial learning. Therefore, there is an effect that the initial setting of complicated parameters can be performed more accurately and promptly.

[Brief description of the drawings]

FIG. 1 is a schematic diagram showing a relationship between an engine 1 and a control device 10 that controls an air-fuel ratio of the engine 1 using a prediction learning method according to the present invention.

FIG. 2 is a schematic block diagram illustrating a configuration of a control device 10.

FIG. 3 is a schematic block diagram showing a configuration of a model base control unit 20 in FIG.

FIG. 4 is a schematic block diagram showing a configuration of a fuel system forward model 22.

FIG. 5 is a model image diagram of a fuel system.

FIG. 6 shows a map constituting a fuel adhesion rate estimation unit.

FIG. 7 shows an initial state of a map constituting a fuel adhesion rate estimation unit before initial learning.

FIG. 8 is a graph showing a relationship among a throttle opening, an air-fuel ratio waveform, and a fuel adhesion rate (teacher data) during execution of initial learning.

FIG. 9 is a diagram showing an example of details of a change in an exhaust air-fuel ratio during a transition.

FIG. 10 is a flowchart illustrating a flow of processing of a model correction unit.

FIG. 11 is a flowchart illustrating a flow of teacher data acquisition processing in a model correction unit.

FIGS. 12A to 12C show other examples of component division of the amount of deviation of the exhaust air-fuel ratio during a transition.

FIG. 13 is a diagram corresponding to FIG. 8, showing another embodiment of the initial learning.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Engine 2 Intake pipe 3 Air cleaner 4 Fuel injection device 5 Cylinder 6 Exhaust pipe 7 Crankcase 8 Throttle valve 10 Control device 12 Throttle opening detection means 13 Crank angle detection means 14 Intake pipe wall temperature detection means 15 Air-fuel ratio sensor 20 Model base Control unit 21 Air system forward model 22 Fuel system forward model 22a Fuel adhesion rate estimation unit 22b Evaporation time constant estimation unit 22c Non-adhesion fuel calculation unit 22d Adhesion fuel calculation 22e Evaporation fuel amount calculation unit 23 Estimated air-fuel ratio calculation unit 24 Basic operation amount Calculation unit 30 Engine rotation speed calculation unit 40 Target air-fuel ratio calculation unit 50 Conversion unit 60 Model correction unit N Engine rotation speed Throttle opening degree Te Intake pipe wall temperature E Control amount Ep Target control amount Ev Estimated control amount Mf Operation amount Av Estimation Air volume Fv Estimated fuel volume x Fuel adhesion rate τ Estimated evaporation time constant xt teacher data (fuel adhesion rate estimation unit) τt teacher data (evaporation time constant estimation unit)

Claims (2)

[Claims] 1. An air-fuel ratio control device for an internal combustion engine that models a dynamic behavior of fuel injected from a fuel injection device as at least a fuel adhesion rate and controls an exhaust air-fuel ratio using the model. The operating conditions are transiently changed between operating state points that define one area by allocating parameters to multiple areas, and the two operating modes are changed based on the deviation of the exhaust air-fuel ratio from the target air-fuel ratio that occurs at that time. When acquiring the teacher data of the fuel adhesion rate at the state point, learning the model based on the teacher data, and acquiring a plurality of different teacher data regarding different regions for the same driving state point, An air-fuel ratio control device for an internal combustion engine, wherein an intermediate value of a plurality of teacher data is obtained as teacher data of a corresponding operating state point.習方 method. 2. The learning method according to claim 1, wherein the intermediate value is an average value of a plurality of teacher data.