JPH1011105A - State control system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To learn a state at all times and enable control, and to improve the control response by controlling the state index of a controlled body to larger than a previously set permissible quantity, and adding the manipulated variable of a 2nd control part having a learning function to the manipulated variable of the 1st control part and obtaining a final manipulated variable. SOLUTION: For air fuel ratio control by EFI, a feedback injection quantity arithmetic part 30 detects the combustion state of an engine from rotating speed information on a crank shaft to calculate the state index of the controlled body and performs control so that the state index is larger than the previously set permissible quantity. Further, an injection quantity by feedback control from a total injection quantity arithmetic part 80 is used as tutor data and a neural network 40 is made to always learn online in the operation of the engine 1 according to the engine rotating speed and a throttle opening extent. Then the air fuel ratio of the engine 1 is controlled according to the addition value of the injection quantity by the feedback control of rotating speed variation information and the injection quantity of the neural network 40.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a state control system, for example, a state control system of an engine mounted on a vehicle or a ship, a robot of a machine tool, a motor of an electric vehicle, and the like.

[0002]

2. Description of the Related Art An engine mounted on a vehicle or a ship, a robot of a machine tool, a motor of an electric vehicle, etc.
State control is performed. For example, an engine mounted on a vehicle or a ship performs air-fuel ratio control by EFI, but feedback control using an O ₂ sensor is generally performed, thereby controlling the air-fuel ratio to a stoichiometric air-fuel ratio. Is possible.

[0003]

However, since this method cannot control the air-fuel ratio other than the stoichiometric air-fuel ratio, in such a case, it is necessary to perform feedback control using a linear air-fuel ratio sensor. Become. However, the linear air-fuel ratio sensor is not only expensive, but also has a problem that it cannot be used in a rich environment because it deteriorates drastically.

Furthermore, feedback control generally has poor response and it is difficult to control the air-fuel ratio in the transient state. Therefore, if the air-fuel ratio in the steady state is to be made lean in consideration of exhaust gas and fuel consumption, it is difficult to control the air-fuel ratio in the transient state. There is a problem in that the air-fuel ratio becomes lean more than necessary, resulting in impaired driving performance such as deterioration of acceleration performance.

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and has as its object to provide a state control system capable of always learning and controlling a state with an inexpensive structure and improving control responsiveness. .

[0006]

In order to solve the above problems and achieve the object, a state control system according to the first aspect of the present invention uses one or a plurality of pieces of information output from an object to be controlled. A first control unit that calculates a state index of the control object and performs control such that the state index becomes a preset allowable amount; and has a learning function for the operation amount of the first control unit. It is characterized in that control is performed using a value obtained by adding the operation amount of the second control unit as a final operation amount. Control is performed so that the state index of the control object becomes a predetermined allowable amount, and a value obtained by adding the operation amount of the second control unit having the learning function to the operation amount of the first control unit is finally obtained. By performing the control as the operation amount, it is possible to control the state by always learning the state, and it is possible to perform the state control in which the control response is improved.

The state control system according to the second aspect of the present invention is characterized in that the first control unit performs feedback control. Simple and reliable control is performed by feedback control so that the state index becomes a preset allowable amount.

The state control method according to the third aspect of the present invention is characterized in that the second control unit realizes the learning function by a neural network. Performs learning using neural networks and always performs appropriate state control.

According to a fourth aspect of the present invention, there is provided the state control method, wherein a part of the second control unit has an inverse model based on a dynamic behavior of the controlled object, and control is performed using the inverse model. It is characterized by performing. In a transient state or the like, appropriate state control is performed using an inverse model.

The state control method according to the invention described in claim 5 is characterized in that the controlled object is an engine and the state of the controlled object is an operating state. The control object is an engine, and controls the operating state of the engine.

In the state control method according to the present invention, the combustion state of the engine is detected from the rotation fluctuation information of the crankshaft, and the feedback control of the air-fuel ratio is performed using the rotation fluctuation information as a feedback signal. The neural network is learned online using the injection amount of the neural network as teacher data, and the air-fuel ratio control is performed based on the added value of the injection amount by the feedback control and the injection amount by the neural network. Without using a linear air-fuel ratio sensor, it is possible to control the air-fuel ratio required by the engine from the crankshaft rotation fluctuation information, and further impair the driving performance even when trying to make the air-fuel ratio lean using feedback control. Control without.

According to the state control method of the present invention, when the engine is in a transient state and feedback control cannot be performed, or when the engine is in a transient state, accurate air-fuel ratio control for a target air-fuel ratio is required as in stoichiometric air-fuel ratio control. In such a case, an injection amount for obtaining the fuel amount required by the engine is obtained by using an inverse model based on the dynamic behavior of the engine, and the air-fuel ratio control is performed using this value. When the engine is in a transient state and feedback control cannot be performed, or when accurate air-fuel ratio control with respect to the target air-fuel ratio is required as in stoichiometric air-fuel ratio control, injection to achieve the fuel amount required by the engine is performed. The amount is obtained, and the air-fuel ratio control is performed based on this value.

The state control method according to the present invention is characterized in that an injector for injecting fuel to enter a cylinder is disposed in an intake passage. Even if the injector is arranged in the intake passage, fuel can be supplied to the cylinder at a predetermined timing.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of a state control system according to the present invention will be described below. In the embodiment of the present invention, the control object of the state control method is an engine mounted on a vehicle or a ship, and the state of the control object is an operation state. However, the present invention is not limited to this. The state control can be similarly performed using a motor or the like of the vehicle as a control target.

FIG. 1 is a system diagram of an engine mounted on a vehicle or a ship. A piston 3 is provided in the cylinder 2 of the engine 1 so as to be able to reciprocate. The piston 3 is connected to a crankshaft 5 via a connecting rod 4.
The crankshaft 5 is rotated by the reciprocating motion. Also, Engine 1
Is provided with a crank angle sensor 6, and the crank angle sensor 6 sends a crank angle signal and an engine speed to the controller 7. An exhaust passage 9 and an intake passage 10 are provided in communication with a combustion chamber 8 formed in the cylinder 2 of the engine 1, and air is sucked into the intake passage 10 from an air cleaner 11. Further, a throttle 12 is provided in the intake passage 10. The throttle 12 controls the amount of intake air, and the throttle opening is sent to the controller 7 by a throttle opening sensor 13. An injector 14 is provided upstream of the throttle 12 in the intake passage 10, and fuel is injected from the injector 14. The injector 14 performs state control for injecting fuel based on information from the controller 7 such as a crank angle signal, an engine speed, and a throttle opening.

By performing state control by the controller 7, even if the injector 14 is arranged on the upstream side of the throttle 12 in the intake passage 10, fuel can be supplied to the cylinder at a predetermined timing. It may be arranged downstream of the throttle 12. In addition, for example, by installing the injector 14 upstream of the throttle 12, the injector 14 is located at a position away from the engine 1, and the fuel is prevented from being reduced in viscosity due to the effect of heat, and the fuel injection is smoothly performed. Become. Further, by disposing the injector 14 upstream of the throttle 12, the flow velocity becomes faster when passing through the throttle 12, so that the atomization of fuel can be improved. Further, for example, the number of injectors can be reduced by using an SPI in which the injectors are arranged in the intake pipe collecting section.

FIG. 2 is a control block diagram of the controller. In the air-fuel ratio control by EFI, when the engine is in a steady state, the feedback control using the rotation fluctuation of the crankshaft and the learning control by the neural network are combined, and when the engine is in a transient state or the stoichiometric air-fuel ratio control. In this case, a system for performing model-based control using an inverse model will be described.

The controller 7 includes a rotation fluctuation detecting section 20 and detects rotation fluctuation based on a crank angle signal. The feedback injection amount calculation unit 30 detects the combustion state of the engine from the crankshaft rotation fluctuation information, calculates the state index of the controlled object, and performs control such that the state index becomes a preset allowable amount. This constitutes a first control unit K1 for performing the control. Feedback control of the air-fuel ratio is performed using this rotation fluctuation information as a feedback signal, and the injection amount by the feedback control from the total injection amount calculation unit 80 is used as teacher data as neural network 4.
0 is always learned on-line based on the engine speed and the throttle opening during the operation of the engine 1 after shipment, and the engine 1 is controlled via the control method switching unit 50 by the addition value of the injection amount by the feedback control and the injection amount by the neural network. Is performed. As described above, without using the linear air-fuel ratio sensor, it is possible to control the air-fuel ratio required by the engine 1 from the rotation fluctuation information of the crankshaft, and even if the air-fuel ratio is made lean using the feedback control. Control can be performed without impairing driving performance.

Further, the neural network 40 constitutes a second control unit K2 having a learning function, and performs control such that the state index of the control object becomes a predetermined allowable amount. By controlling as a final operation amount a value obtained by adding the operation amount of the second control unit K2 having a learning function to the operation amount of K1, the state can always be learned and controlled, and the control responsiveness can be improved. Improved state control is possible.

The first control unit K1 performs feedback control, and performs simple and reliable control such that the state index becomes a preset allowable amount by the feedback control.

Further, the second control unit K2 realizes the learning function by the neural network 40, so that learning can be performed by the neural network 40 and always appropriate state control can be performed.

The control method switching unit 50 is controlled by the steady / transient determination unit 60, and the control method switching unit 50 is connected to the feedback injection amount calculation unit and the model base control unit based on the engine speed and the throttle opening. Switch.

When the engine 1 is in a transient state and feedback control cannot be performed, or when accurate air-fuel ratio control for a target air-fuel ratio is required as in stoichiometric air-fuel ratio control, the mode is switched to the model base control unit. Based on the engine speed and the throttle opening, the model-based control unit 70 calculates an injection amount for achieving the fuel amount of the target air-fuel ratio required by the engine using an inverse model based on the dynamic behavior of the engine. The air-fuel ratio control is performed based on this value. When the engine is in a transient state and feedback control cannot be performed, or when accurate air-fuel ratio control with respect to the target air-fuel ratio is required as in stoichiometric air-fuel ratio control, injection to achieve the fuel amount required by the engine is performed. The amount is obtained, and the air-fuel ratio control can be performed based on this value.

As described above, in the steady state as a whole control flow, the torque fluctuation of the engine 1 is estimated by detecting the rotation fluctuation of the crankshaft, and the magnitude of the torque fluctuation becomes a preset allowable value. The feedback control is performed as follows. As a result, it is possible to control the air-fuel ratio so that exhaust gas / fuel efficiency and driving performance are compatible as much as possible. In order to improve the responsiveness of the feedback control, the injection amount output by the feedback control is learned by the neural network, and the feedback control is gradually replaced with the feedforward control. This allows
Even when the operating state is switched, the target air-fuel ratio can be quickly controlled. In the transient state, the engine 1
By performing model-based control using the inverse model of, the air-fuel ratio can be quickly controlled without sacrificing acceleration performance. Thereby, even when the air-fuel ratio lean control is performed in the steady state, the operation performance in the transient state can be ensured. Also, by applying this model-based control to stoichiometric air-fuel ratio control, lean (rich) spikes during transition can be prevented, and accurate air-fuel ratio control becomes possible.

Next, each component of the controller shown in FIG. 2 will be described. First, the configuration of the rotation fluctuation detecting unit 20 will be described in detail. FIG. 3 is a block diagram of the rotation fluctuation detecting unit. The rotation fluctuation detecting unit 20 uses the crank angle signal of the crank angle sensor 6 (for example, generates a pulse every 30 °) of the engine 1 shown in FIG. Calculate the angular velocity at the point. Further, the angular acceleration between the two points is calculated by the angular acceleration detection unit 22 from the angular velocities at the two points. A deviation between the obtained angular acceleration and a value obtained by smoothing the angular acceleration by the primary filter 23 of the following equation is obtained. A value obtained by cumulatively adding, for example, 100 deviations by the deviation accumulating unit 24 is defined as rotation fluctuation.

Ave (n) = α · Ave (n−1) +
(1-α) · Acc (n) Ave: average angular acceleration Acc: angular acceleration Next, the configuration of the feedback injection amount calculation unit 30 will be described in detail. FIG. 4 is a block diagram of the feedback injection amount calculation unit. A comparison unit 32 compares the permissible value of the rotation fluctuation previously recorded in the memory by the fluctuation permissible value calculation unit 31 with the engine speed and the throttle opening as parameters, and the permissible value of the rotation fluctuation obtained from the rotation fluctuation detection unit 20. I do. Note that the allowable value of the rotation fluctuation may be obtained from a map using the engine speed and the throttle opening as parameters.

Based on this comparison, the injection amount calculation section 33 outputs the injection amount by feedback. If the rotation fluctuation is larger than the permissible value, it is determined that the engine is lean too much, and the injection amount of the feedback stored in the memory is increased by a small amount. Conversely, if the rotation fluctuation is smaller than the allowable value, it is determined that the engine can still be made lean, and the injection amount is reduced by a small amount.

Next, the configuration of the neural network 40 will be described in detail. FIG. 5 is a configuration diagram of the neural network. Neural network (neural network) 4
As is well known, 0 performs highly parallel distributed processing type information processing, and is applied to understanding of an external environment. Further, as a typical neural network 40, a Perceptron type network, a Hopfield, etc.
Network and Boltzman machine
n Machine). Among them, the one using a three-layer type perceptron guarantees the convergence to the optimal solution, and the input layer 41, the intermediate layer 42, and the output layer 4
Consists of three.

The units 41a, 4a in the input layer 41
1b and 41c are not connected, and similarly, the units 42a, 42b, 42c and 42d in the intermediate layer 42 are not connected, and the units are connected with the weight w of the connection only between the layers. A unit 41c in the input layer 41 indicates a bias neuron, and a unit 42d in the intermediate layer 42
Also indicate a bias neuron, and the units in the intermediate layer 42 are the units 41a, 41b, 41c of the input layer 41.
, The sum (internal state value) is calculated by y = f (Σxi · wi + θ), and an appropriate output function f (x) = 1 / (1 + e ^−x ) is multiplied. Output.

Here, xi represents an input, wi represents a weight, θ represents a bias amount, and an output function f (x) represents a sigmoid function.

As described above, the engine speed and the throttle opening are input to the input layer 41 of the neural network 40, the input is weighted by the intermediate layer 42, and the input is similarly weighted by the output layer 43. To output the injection amount. The learning data of the neural network 40 is taken in immediately after the transition from the steady state to the transient state. The input data is the data input last time (at the end of the steady state), and the teacher data corresponding to this input data is the injection amount currently stored in the feedback injection amount calculation unit 30 and the previous neural network 40 outputs. It is the value obtained by adding the injection amount. At this time (immediately after the transition from the steady state to the transient state), the feedback injection amount stored in the memory is cleared to zero. The learning of the neural network 40 is always performed during the idle time of the CPU in the controller 7 with respect to the learning data for a plurality of operating states obtained in this way.

Next, the configuration of the feedback injection amount calculation section 30 will be described. The total injection amount calculator 30 adds the injection amount by feedback and the injection amount output from the neural network 40, and drives the injector 14 as the total injection amount.

Next, the configuration of the steady / transient determination section 60 will be described in detail. The steady / transient determination unit 60 detects a change in the throttle opening in a predetermined engine speed region, determines that the change is within a set range for a certain period of time, determines a steady state, and otherwise determines a transient state. .

Next, the configuration of the model base control unit 70 will be described in detail. 6 is a block diagram of a model-based control unit, FIG. 7 is a block diagram showing a forward model of an air system, and FIG. 8 is a block diagram showing a forward model of a fuel system.

The model-based control unit 70 models the engine (including the intake pipe) separately for the air system and the fuel system, obtains the estimated amount of air entering the cylinder by the forward model 71 of the air system, and obtains the forward model 72 of the fuel system. To obtain the estimated amount of fuel entering the cylinder,
It is input to the air-fuel ratio calculation unit 73. The air-fuel ratio calculation unit 73 calculates the estimated air-fuel ratio, and feeds it back to approach the target air-fuel ratio, and the feedback calculation unit 74 obtains the injection amount based on the model. Here, when the air amount and the injection amount are input, the model in which the air-fuel ratio is output as a result is a forward model, and when the target air-fuel ratio is input, the model in which the injection amount is output is the reverse model. I have.

Based on the engine speed and the throttle opening, an estimated amount of air entering the cylinder is output by a forward model 71 of the air system. The forward model 71 of the air system is configured as shown in FIG. 7, and the input of the throttle opening is advanced by the phase advance unit 710 to advance the phase of the air system by the dead time of the fuel system.
The air amount calculating unit 711 calculates the amount of air entering the cylinder by feeding back the intake negative pressure. The amount of air entering the cylinder is converted to pressure by a pressure conversion unit 712, and an intake negative pressure calculation unit 713 obtains an intake negative pressure based on the engine speed and the volume ratio of the intake negative pressure. The pressure is fed back to the air amount calculation unit 711 and sent to the volume efficiency calculation unit 714. Inputting the engine speed into the phase advance unit 7
Similarly, at 15, the phase of the air system is advanced by the dead time of the fuel system and is directly input to the intake negative pressure calculation unit 713 and is also input to the volume efficiency calculation unit 714. The volumetric efficiency is obtained from the data and the intake negative pressure, and sent to the intake negative pressure calculation unit 713.

Further, based on the injected fuel and the throttle opening, the estimated amount of fuel entering the cylinder is output by the forward model 72 of the fuel system. The forward model 72 of the fuel system is configured as shown in FIG. 8, and based on the injected fuel and the throttle opening, the injected fuel calculation unit 720 determines the adhesion to the intake passage and the delay time from the fuel injection to the fuel entering the cylinder. The amount of fuel entering the cylinder is calculated in consideration of the above.

As described above, the fuel system is modeled by the first-order delay and the dead time, and a model obtained by removing the dead time element from this model is used as a fuel system model. At the same time, the phases of both models are adjusted by advancing the phase of the air system by the dead time of the fuel system. Specifically, the phase advance is realized by estimating the future values of the engine speed and the throttle opening input to the air system by the least square method. Although the gain of the feedback control cannot be increased because the actual engine has a dead time and a high-order delay, when the feedback control is performed on the model created here, the fuel system is a simple first-order delay system. The feedback gain can be increased, and an inverse model of the engine can be configured. Since the air system can be regarded as a disturbance, the control period is set so that the air system can be regarded as steady.

As a part of the second control unit K2, an inverse model based on the dynamic behavior of the controlled object is provided, and when the engine 1 is in a transient state and the feedback control cannot be performed,
Alternatively, when accurate air-fuel ratio control with respect to the target air-fuel ratio is required as in the case of stoichiometric air-fuel ratio control, the injection amount for obtaining the fuel amount required by the engine using an inverse model based on the dynamic behavior of the engine 1 , The air-fuel ratio control is performed by using this value, and the injection amount for obtaining the fuel amount required by the engine is obtained, and the air-fuel ratio control can be performed by using this value.

Further, the controller 7 is provided as shown in FIG.
Can be configured. In this embodiment, the second control
The model parameters are stored in the model base controller 70 of the section K2.
A model correction unit 100 that corrects the engine speed.
Number of turns, throttle opening and O _TwoBased on the sensor signal,
Control may be performed such that the burning is always performed at the stoichiometric air-fuel ratio.

The controller 7 can be configured as shown in FIG. In this embodiment, a total injection amount calculating unit 10
The learning is performed on-line based on the engine speed and the throttle opening by the neural network 102 using the O ₂ sensor signal as teacher data, and the model-based control unit 7
The total fuel injection amount is calculated by the total fuel injection amount calculator 101 by adding the output from the neural network 102 and the output from the neural network 102, and the output is used to control the air-fuel ratio of the engine 1 via the control method switching unit 50. Is also good.

Next, the calculation of the feedback injection time and the total injection time of the state control method will be described.

FIG. 11 shows a flow chart for calculating the feedback injection time. If the crank angle is 1 at step a1, the angular velocity (v
The operation of 1) is performed and the processing is terminated. If it is not the crank angle 1 in step a1, it is determined whether or not the crank angle is 2 in step c1, and if it is not crank angle 2, the process is terminated. The angular velocity (v2) is calculated, and the angular velocity at two points in the expansion stroke of the engine 1 is calculated.
Next, in step e1, the angular acceleration (acc) = (v2-v
1) / Calculate the required time, and calculate the average value of the angular acceleration (acc_ave) in step f1.

Further, at step g1, an angular acceleration variation value (f1) = | acc-acc_ave | is determined, and at step h1, a combustion deterioration index (pnt) = pnt + f1 is determined. In step i1, the number of times of the combustion deterioration index (pnt) is N = 100? It is determined whether or not N = 100, and if N = 100 is not satisfied, N = N + 1 is set in step j1 and step c1
Repeat until it reaches 100 times. When the number of calculations of the combustion deterioration index (pnt) reaches 100, step k
1, N = 1, and the combustion deterioration index (pn
It is determined whether or not t) is pnt> allowable value (lim).

If the combustion deterioration index (pnt) does not exceed the allowable value (lim), the feedback injection time (t_fb) = t_fb-Δt is shortened in step m1 to reduce the injection amount, thereby reducing the combustion deterioration. If the index (pnt) exceeds the allowable value (lim), the feedback injection time (t_fb) = t_fb + Δt is increased in step n1 to increase the injection amount, and in step o1, the combustion deterioration index is set to pnt = 0. To end.

FIG. 12 shows a flow chart for calculating the total injection time. In the case of the steady state in step a2, the calculation of the injection time (t_nn) by the neural network 40 is performed in step b2, and the total injection time is calculated as the total injection time (t_tt1) = t_nn + t_fb in step c2, and the processing is terminated.

If it is not the steady state in step a2,
At step d2, it is determined whether or not the previous time is in a steady state. If so, at step e2, the total injection time t_tt1 is compared with the previous operation state by the neural network 4.
0 is set to the teacher data, and the total injection time is set to t_fb = 0 in step f2.

If the previous time is not the steady state at step d2, the injection time (t_m
The calculation of d1) is performed, and in step h2, the total injection time (t_tt1) is set to t_md1, and the process ends.

[0049]

As described above, according to the first aspect of the present invention, control is performed such that the state index of the controlled object becomes a predetermined allowable amount, and learning is performed based on the operation amount of the first control unit. Since the value obtained by adding the operation amount of the second control unit having the function is used as the final operation amount, control can be performed by constantly learning the state, and furthermore, state control with improved control responsiveness is possible. .

According to the second aspect of the present invention, simple and reliable control can be performed by feedback control so that the state index becomes a predetermined allowable amount.

According to the third aspect of the present invention, learning is performed by a neural network, and appropriate state control can always be performed.

According to the fourth aspect of the present invention, in a transient state or the like, appropriate state control can be performed using an inverse model, and for example, a response delay due to fuel adhesion to the intake pipe can be compensated. The degree of freedom is improved.

According to the fifth aspect of the present invention, the object to be controlled is the engine, and the operating state of the engine can be controlled.

According to the sixth aspect of the present invention, it is possible to control the air-fuel ratio to an arbitrary value required by the engine from the crankshaft rotation fluctuation information without using a linear air-fuel ratio sensor, and further to make the air-fuel ratio lean using feedback control. Even if it is attempted, since the air-fuel ratio can be controlled without deteriorating the driving performance and detecting the combustion state of the engine, the air-fuel ratio always required by the engine can be controlled without being affected by environmental changes or changes over time. When it is necessary to accurately control the air-fuel ratio to the target value, it can be dealt with by switching the control method. Further, since the air-fuel ratio control without using the exhaust gas sensor is possible, it is not only advantageous in terms of cost, but also there is no need to worry about the deterioration of the sensor or the installation location.

According to the seventh aspect of the present invention, when the engine is in a transient state and feedback control cannot be performed, or when accurate air-fuel ratio control for the target air-fuel ratio is required as in stoichiometric air-fuel ratio control, The fuel injection amount required by the engine is obtained, and the air-fuel ratio control is performed based on this value, and the responsiveness of the air-fuel ratio control in a transient state or the like is improved. Without
Lean air-fuel ratio at regular times.

According to the eighth aspect of the present invention, fuel can be supplied to the cylinder at a predetermined timing even if the injector is disposed in the intake passage. For example, by installing the injector upstream of the throttle, the fuel is heated. This prevents the viscosity from dropping due to the effect, making the fuel injection smoother,
Further, the flow velocity increases when passing through the throttle, so that atomization of fuel can be improved. Alternatively, the number of injectors can be reduced by using an SPI in which the injectors are arranged in the intake pipe collecting portion.

[Brief description of the drawings]

FIG. 1 is a system diagram of an engine mounted on a vehicle or a ship.

FIG. 2 is a control block diagram of a controller.

FIG. 3 is a block diagram of a rotation fluctuation detecting unit.

FIG. 4 is a block diagram of a feedback injection amount calculation unit.

FIG. 5 is a configuration diagram of a neural network.

FIG. 6 is a block diagram of a model base control unit.

FIG. 7 is a block diagram showing a forward model of an air system.

FIG. 8 is a block diagram showing a forward model of a fuel system.

FIG. 9 is a control block diagram of another controller.

FIG. 10 is a control block diagram of another controller.

FIG. 11 shows a feedback injection time calculation flow.

FIG. 12 shows a total injection time calculation flow.

[Explanation of symbols]

 Reference Signs List 1 engine 7 controller 6 crank angle sensor 13 throttle opening sensor 14 injector 20 rotation fluctuation unit 30 feedback injection amount calculation unit 40 neural network 50 control method switching unit 60 steady / transient determination unit 70 model base control unit 80 total injection amount calculation unit K1 first control unit K2 second control unit

──────────────────────────────────────────────────続 き Continuation of the front page (51) Int.Cl. ⁶ Identification number Office reference number FI Technical display location F02D 45/00 324 F02D 45/00 324 340 340C G05B 13/04 G05B 13/04

Claims (8)

[Claims] 1. A first method for calculating a state index of a control object from one or a plurality of pieces of information output from the control object, and performing a control such that the state index becomes a preset allowable amount. A state control method, comprising: a control unit, wherein a value obtained by adding an operation amount of a second control unit having a learning function to an operation amount of the first control unit is controlled as a final operation amount. 2. The state control method according to claim 1, wherein the first control section performs feedback control. 3. The state control method according to claim 1, wherein the second control unit implements the learning function by a neural network. 4. A control device according to claim 1, wherein said second control section has an inverse model based on a dynamic behavior of said control object, and performs control using said inverse model. The state control method according to claim 3. 5. The control object is an engine, and the state of the control object is an operating state.
The state control method according to claim 4. 6. An engine combustion state is detected from rotation fluctuation information of a crankshaft, feedback control of an air-fuel ratio is performed by using the rotation fluctuation information as a feedback signal, and a neural network is formed by using an injection amount by the feedback control as teacher data. Let them learn online,
6. The state control method according to claim 5, wherein the air-fuel ratio control is performed based on an added value of the injection amount by the feedback control and the injection amount by the neural network. 7. When the engine is in a transient state and feedback control cannot be performed, or when accurate air-fuel ratio control with respect to a target air-fuel ratio is required as in stoichiometric air-fuel ratio control, the dynamic behavior of the engine is determined. 7. The state control system according to claim 6, wherein an injection amount for obtaining a fuel amount required by the engine is obtained by using an inverse model based on the calculated amount, and the air-fuel ratio control is performed based on the obtained injection amount. 8. The state control system according to claim 6, wherein an injector for injecting fuel to be injected into the cylinder is disposed in the intake passage.