JPH0455235Y2 - - Google Patents

Description

[Detailed Description of the Invention] <Industrial Application Field> The present invention relates to a learning control device for an air-fuel ratio feedback control system in an internal combustion engine having an electronically controlled fuel injection device.

<Prior art> A fuel injection valve used in an electronically controlled fuel injection device opens in response to a drive pulse signal given in synchronization with engine rotation, and injects fuel at a predetermined pressure during the valve opening period. It has become commonplace. Therefore, the fuel injection amount is controlled by the pulse width of the drive pulse signal, and if this pulse width is set as Ti and the control signal corresponds to the fuel injection amount, then in order to obtain the stoichiometric air-fuel ratio, which is the target air-fuel ratio, Ti is determined by the formula.

Ti=Tp・COEF・α+Ts However, Tp is the basic pulse width corresponding to the basic fuel injection amount and is called the basic fuel injection amount for convenience. Tp＝
K·Q/N, where K is a constant, Q is the engine intake air flow rate, and N is the engine speed. COEF is various correction coefficients such as water temperature correction. α is a feedback correction coefficient for air-fuel ratio feedback control (λ control) to be described later. Ts is a voltage correction amount, which is used to correct changes in the injection flow rate of the fuel injector due to changes in battery voltage.

Regarding λ control, an O ₂ sensor is installed in the exhaust system to detect the actual air-fuel ratio, and whether the air-fuel ratio is richer or leaner than the stoichiometric air-fuel ratio is controlled by the slice level. A coefficient α is determined and the stoichiometric air-fuel ratio is maintained by varying this α.

Here, the value of the feedback correction coefficient α is changed by proportional-integral (PI) control to ensure stable control.

In other words, the output voltage of the O ₂ sensor is compared with the slice level voltage, and if it is higher or lower than the slice level, the air-fuel ratio is rich (lean) without suddenly enriching or thinning the air-fuel ratio. ), the air-fuel ratio is controlled to be leaner (richer) by first lowering (raising) it by P, then gradually lowering (raising) it by I minutes (see Figure 7).

However, under conditions where λ control is not performed, α
The desired air-fuel ratio is obtained by clamping and setting various correction coefficients COEF.

By the way, the base air-fuel ratio under λ control conditions, that is, the air-fuel ratio when α=1, is the stoichiometric air-fuel ratio (λ
= 1), there is no need for feedback control; however, in reality, it is difficult to control due to variations in component parts (for example, air flow meters, fuel injection valves, pressure regulators, control units), changes over time, and fuel injection. Feedback control is performed because the base air-fuel ratio deviates from λ=1 due to factors such as non-linearity in the pulse width-flow rate characteristic of the valve and changes in operating conditions and environment.

However, if the base air-fuel ratio deviates from λ=1, it takes time to stabilize the step in the base air-fuel ratio to λ=1 through feedback control when the operating range changes significantly. For this purpose, the proportionality and integral constants (P/I) are increased, which causes overshoot and undershoot, resulting in poor controllability. In other words, if the base air-fuel ratio deviates from λ=1, the air-fuel ratio will be controlled within a range that deviates considerably from the stoichiometric air-fuel ratio.

As a result, the three-way catalyst is operated at a point where the conversion efficiency is poor, and not only does the cost increase due to an increase in the amount of precious metal in the catalyst, but the conversion efficiency further deteriorates as the catalyst deteriorates, making it necessary to replace the catalyst.

Therefore, by setting the base air-fuel ratio to λ = 1 through learning, it is possible to eliminate the deviation from λ = 1 caused by the step in the base air-fuel ratio during transients, and to reduce the P/I component, improving controllability. An air-fuel ratio learning control device for improving the air-fuel ratio was filed by the present applicant as Japanese Patent Application No. 1982-76221 (Japanese Unexamined Patent Publication No. 58-203828) or Japanese Patent Application No. 58-197499.

This is because if the base air-fuel ratio deviates from the stoichiometric air-fuel ratio during air-fuel ratio feedback control, the feedback correction coefficient α increases to fill the gap, so the engine operating state and α at this time are detected. , calculate the learning correction coefficient Kl based on α and store it, and when the same engine operating condition returns, the base air-fuel ratio is corrected using the stored learning correction coefficient Kl so that it is more responsive to the stoichiometric air-fuel ratio. do. Here, the learning correction coefficient Kl is stored for each region of a predetermined range obtained by dividing the RAM map into a grid according to appropriate parameters of the engine operating state such as engine speed and load.

Specifically, a map of the learning correction coefficient Kl corresponding to engine operating conditions such as engine speed and load is provided in RAM, and when calculating the fuel injection amount Ti, the basic fuel injection amount Tp is calculated as shown in the following formula. Correct using learning correction coefficient Kl.

T=Tp・COEF・Kl・α+Ts Then, learn Kl by following the steps below.

In the steady state, the region of the engine operating state at that time is detected, and the deviation Δα (=α−α ₁ ) of α from the reference value α ₁ during that period is detected as an average value. The reference value α ₁ is generally set to 1 as a value corresponding to λ=1.

The Kl that has been learned up to now corresponding to the region of the engine operating state is searched.

The value of Kl+M·Δα is calculated from Kl and Δα, and the memory is updated using the result (learning value) as a new Kl _(NEW) . M is a constant and 0<M<1.

<Problems to be solved by the invention> By the way, in such an air-fuel ratio learning control device, learning of the learning correction coefficient is performed when the engine operating state is in a steady state, that is, as described above. This is done only when the engine remains within the learning range to some extent, but under actual operating conditions, there are few opportunities for the engine operating state to remain in the same learning range, so the learning progresses slowly, and it is not possible to perform good air-fuel ratio control through learning. There was a time when it took me a while to get used to it.

If you reduce the number of grids that divide the area and expand the individual learning areas, the chances that the engine operating state will remain within the same learning area will increase and the speed of learning will speed up. Since the deviation in the engine operating state within the region becomes large, the amount of deviation in the air-fuel ratio becomes large, and learning control accuracy cannot be improved.

These relationships are shown in figures 10 and 11.
It will look like the figure.

The present invention was developed by focusing on these conventional problems, and provides an air-fuel ratio learning control device for an internal combustion engine that solves the above problems by changing the size of the learning area over time. The purpose is to provide.

<Means for Solving the Problems> In order to achieve the above object, the present invention provides an air-fuel ratio learning control device for an internal combustion engine, as shown in FIG. It is constructed by the means of (I), and in particular, the provision of the means of (I) is
The characteristics of

(A) A first detection means for detecting the engine intake air flow rate, a second detection means for detecting the engine speed, and a third detection means for detecting the engine exhaust component and thereby detecting the air-fuel ratio of the engine intake air mixture. An engine operating state detecting means (B) including at least a detecting means, which calculates a basic fuel injection amount based on the engine intake air flow rate outputted by the first detecting means and the engine rotational speed outputted by the second detecting means. Basic fuel injection amount calculation means (C) Grid axes that divide the engine operating state into a plurality of regions according to their parameters, and learning for correcting the basic fuel injection amount for each region surrounded by these grid axes. a rewritable storage means (D) that stores a correction coefficient; a learning correction coefficient retrieval means (E) that searches the storage means for a learning correction coefficient in a corresponding area based on the actual engine operating state; and the third detection means. A feedback correction coefficient that increases or decreases by a predetermined amount a feedback correction coefficient for correcting the basic fuel injection amount by comparing the air-fuel ratio output by the controller with the target air-fuel ratio and bringing the actual air-fuel ratio closer to the target air-fuel ratio. Setting means (F) determines the fuel injection amount based on the basic fuel injection amount calculated by the basic fuel injection amount calculation means, the learning correction coefficient searched by the learning correction coefficient search means, and the feedback correction coefficient set by the feedback correction coefficient setting means. Fuel injection amount calculation means (G) for calculating the injection amount; Fuel injection means (H) that injects fuel into the engine on and off in response to a drive pulse signal corresponding to the fuel injection amount calculated by the fuel injection amount calculation means. Steady state detection means (I) for detecting that the engine operating state is in a steady state.Steady state detection means (I) for detecting that the engine operating state is in a steady state.Steady state detection means (I) learns the deviation from the reference value of the feedback correction coefficient for each region when the engine operating state is in the steady state. learning correction coefficient correction means (J) for correcting and rewriting the learning correction coefficient corresponding to the region of the engine operating state in the direction of decreasing the learning correction coefficient; Learning progress determining means (K) for determining the learning progress determining means (K) When the learning progress determining means determines that the learning progress is low, each area divided by the lattice axes of the storage means is relatively large. The lattice axis switching means sets the lattice axes as shown in FIG. , the basic fuel injection amount corresponding to the target air-fuel ratio is calculated from the engine intake air flow rate and the engine rotational speed according to a predetermined formula, and the learning correction coefficient retrieval means D retrieves the actual engine operating state from the storage means C. The feedback correction coefficient setting means E compares the actual air-fuel ratio with the target air-fuel ratio, and adjusts the feedback correction coefficient, for example, proportionally so that the actual air-fuel ratio approaches the target air-fuel ratio. It is set by increasing or decreasing a predetermined amount based on integral control. Then, the fuel injection amount calculation means F calculates the fuel injection amount by correcting the basic fuel injection amount using a learning correction coefficient (the one retrieved or the one modified as described later after the search) and further correcting it with a feedback correction coefficient. . Then, the fuel injection means G is actuated by a drive pulse signal corresponding to this fuel injection amount.

On the other hand, when the steady state detection means H detects a steady state, the learning correction coefficient correction means I
The storage means C learns the deviation of the feedback correction coefficient from the reference value, corrects the learning correction coefficient corresponding to the region of the engine operating state in a direction to reduce the deviation, and stores the deviation.
Rewrite the data.

Here, when the learning progress determining means J determines that the learning progress is low, the lattice axis switching means K determines that each region partitioned by the lattice axes of the storage means C is relatively large. Since the lattice axes are set in this way, the chances that the engine operating state will remain in the same region are increased, which speeds up the progress of learning in each region and allows learning to proceed evenly across all regions.

Further, when it is determined that the learning progress is high, the lattice axis switching means K switches the lattice axes so that each area of the storage means C is relatively small. As a result, learning is performed for each subdivided area, so learning control accuracy is improved.

<Example> An example of the present invention will be described below based on the drawings.

In FIG. 2, air is taken into the engine 1 through an air cleaner 2, an intake duct 3, a throttle chamber 4, and an intake manifold 5. As shown in FIG.

The intake duct 3 is provided with an air flow meter 6 as means for detecting the intake air flow rate Q, and outputs a voltage signal corresponding to the intake air flow rate Q signal. The throttle chamber 4 is provided with a primary throttle valve 7 and a secondary throttle valve 8 which are operated in conjunction with an accelerator pedal (not shown) to control the intake air flow rate Q. Further, an auxiliary air passage 9 is provided that bypasses these throttle valves 7 and 8, and an idle control valve 10 is interposed in this auxiliary air passage 9. The intake manifold 5 or the intake port of the engine 1 is provided with a fuel injection valve 11 as a fuel injection means. The fuel injection valve 11 is an electromagnetic fuel injection valve that opens when the solenoid is energized and closes when the energization is stopped, and opens when the solenoid is energized by a drive pulse signal.
The engine 1 is injected with fuel that is pressure-fed from a fuel pump (not shown) and controlled to a predetermined pressure by a pressure regulator.

Exhaust gas is exhausted from the engine 1 via an exhaust manifold 12, an exhaust duct 13, a three-way catalyst 14, and a muffler 15.

The exhaust manifold 12 is provided with an O ₂ sensor 16 . This O ₂ sensor 16 outputs a voltage signal according to the ratio of the oxygen concentration in the atmosphere (constant) and the oxygen concentration in the exhaust gas, and is a well-known method in which the electromotive force suddenly changes when the air-fuel mixture is combusted at the stoichiometric air-fuel ratio. It is a sensor. Therefore, the O ₂ sensor 16 is a means for detecting the air-fuel ratio (rich/lean) of the air-fuel mixture. Three-way catalyst 14
is a catalyst device that efficiently oxidizes or reduces CO, HC, and NOx in exhaust gas near the stoichiometric air-fuel ratio of the air-fuel mixture and converts them into other harmless substances.

In addition, a crank angle sensor 17 is provided. The crank angle sensor 17 is connected to the crank pulley 1
8 is provided with a signal disk plate 19,
For example, the teeth provided on the outer periphery of the plate 19
Outputs a reference signal every 120 degrees and a position signal every 1 degree. Here, the engine rotation speed N can be calculated by measuring the period of the reference signal. Therefore, the crank angle sensor 17 is a means for detecting not only the crank angle but also the engine speed N.

The air flow meter 6 and the crank angle sensor 1
7 and the O ₂ sensor 16 are both input to a control unit 30. Furthermore, the voltage of a battery 20 is directly applied to the control unit 30 via an engine key switch 21 as its operating power source and for detecting the power supply voltage. Furthermore, control unit 3
0 includes, as necessary, a water temperature sensor 22 that detects the engine cooling water temperature, a throttle sensor 23 including an idle switch that detects the throttle opening of the primary throttle valve 7, and a vehicle speed sensor 2 that detects the vehicle speed.
4. A signal from a neutral switch 25 or the like that detects the neutral position of the transmission is input. The control unit 30 performs arithmetic processing based on various input signals, outputs a drive pulse signal with an optimum pulse width to the fuel injection valve 11, and obtains a fuel injection amount for obtaining an optimum air-fuel ratio.

As shown in FIG. 3, the control unit 30 includes a CPU 31, a P-ROM 32, a CMOS-
It has a RAM 33 and an address decoder 34. Here, the RAM 33 is a rewritable storage means for learning control, and as an operating power source for the RAM 33, a battery 20 is used as the operating power source without using the engine key switch 21 in order to retain the memory contents even after the engine key switch 21 is turned off. Connect via a stabilized power supply.

Among the input signals to the CPU 31, each voltage signal from the air flow meter 6, O ₂ sensor 16, battery 20, water temperature sensor 22, and solo sensor 23 is an analog signal. 36. A/D converter 3
6 is controlled by the CPU 31 via an address decoder 34 and an A/D conversion timing controller 37. The reference signal and position signal from the crank angle sensor 17 are inputted via a one-shot multi-circuit 38. The signal from the idle switch built into the throttle sensor 23 and the signal from the neutral switch 25 are input via the digital input interface 39, and the signal from the vehicle speed sensor 24 is input via the waveform shaping circuit 40. It's getting old.

An output signal from the CPU 31 (a driving pulse signal for the fuel injection valve 11) is sent to the fuel injection valve 11 via a current waveform control circuit 41.

Here, the CPU 31 executes the flow chart shown in FIGS. 4 to 6 (fuel injection amount calculation routine,
(Learning subroutine and lattice axis switching routine) based program (stored in ROM 32)
Input/output operations and arithmetic processing are performed in accordance with the above, and the fuel injection amount is controlled.

The basic fuel injection amount calculation means, the learning correction coefficient search means, the feedback correction coefficient setting means, the fuel injection amount calculation means, the steady state detection means, the learning correction coefficient correction means, the learning progress determination means, and the grid axis switching means. The functions are accomplished by said program.

Next, the operation will be explained with reference to the flowcharts shown in FIGS. 4 to 6.

In the fuel injection amount calculation routine shown in FIG. 4, in step 1 (S1 in the figure), the intake air flow rate Q obtained from the signal from the air flow meter 6 and the engine speed obtained from the signal from the crank angle sensor 17. Basic fuel injection amount Tp (=K・Q/
N) is calculated. This part corresponds to the basic fuel injection amount calculation means.

In step 2, various correction coefficients COEF are applied as necessary.
Set.

In step 3, a corresponding learning correction coefficient Kl is searched from the engine speed N representing the engine operating state and the basic fuel injection amount (load) Tp. This part corresponds to the learning correction coefficient retrieval means.

Here, the learning correction coefficient Kl is determined by partitioning the map with the engine speed N on the horizontal axis and the basic fuel injection amount Tp on the vertical axis using a grid of about 8 x 8, dividing the area, and storing each area on the RAM 33. A learning correction coefficient Kl is stored for each time. Note that all learning correction coefficients Kl are set to an initial value of 1 before learning has started.

In step 4, a voltage correction amount Ts is set based on the voltage value of the battery 20.

In step 5, it is determined whether the λ control condition is met.

Here, if the condition is not the λ control condition, for example, in a high rotation, high load region, etc., the feedback correction coefficient α is clamped to the previous value (or reference value 1), and the process proceeds from step 5 to step 10, which will be described later.

For λ control conditions, steps 6 to 8
The feedback correction coefficient α is set by integral control or proportional-integral control in which the output voltage V ₀₂ of the O ₂ sensor 16 and the slice level voltage V _ref corresponding to the stoichiometric air-fuel ratio are compared to determine whether the air-fuel ratio is rich or lean. This portion corresponds to the feedback correction coefficient setting means. Specifically, in the case of integral control, the air-fuel ratio = rich (V ₀₂ > V _ref ) by comparison in step 6.
When it is determined that the air-fuel ratio is lean (V ₀₂ <V _ref ), the feedback correction coefficient α is decreased by a predetermined integral (I) from the previous value in step 7. The feedback correction coefficient α is increased by a predetermined integral (I) with respect to the previous value. In the case of proportional-integral control, in addition to this, when reversing rich←→lean, an increase or decrease is performed by a predetermined proportional amount (P) larger than the integral (I) in the same direction as the integral (I).

In the next step 9, the learning subroutine shown in FIG. 5 is executed. This will be discussed later.

Thereafter, in step 10, the fuel injection amount Ti is calculated according to the following equation. This part corresponds to the fuel injection amount calculation means.

Ti=Tp・COEF・Kl・α+Ts However, as Kl, the one found in step 3 or the one modified in the learning subroutine shown in FIG. 5 is used.

When the fuel injection amount Ti is calculated, a drive pulse signal having a pulse width of Ti is output at a predetermined timing in synchronization with the engine rotation, and is applied to the fuel injection valve 11 via the current waveform control circuit 41. Fuel injection takes place.

Next, the learning subroutine shown in FIG. 5 will be explained.

Engine rotation speed N indicating the engine operating status in step 11
It is determined whether or not the basic fuel injection amount Tp and the basic fuel injection amount Tp are in the same range as the previous time. If it is the same area as the previous time, it is determined in step 12 whether flag F is set, and if it is not set, step 13 is performed.
In step 14, it is determined whether the output of the O ₂ sensor 16 has been reversed, that is, the direction of increase or decrease of the feedback correction coefficient α has been reversed, and each time this flow is reversed, the count value representing the number of reversals is incremented by 1 in step 14. When C=2, the process advances from step 15 to step 16 and flag F is set. This flag F is set when the output of the O ₂ sensor 16 inverts twice in the same area, as it is assumed that a steady state has been reached. After setting flag F, proceed to step 11.
If the area is the same as the previous one, proceed to step 12.
Proceed to step 17. This part of steps 11 to 16 corresponds to the steady state detection means, and it is determined that the engine operating state is in one of the divided regions, and that the direction of increase/decrease of the feedback correction coefficient α is a predetermined number of times (twice).
A steady state is detected when the state is reversed.

In steady state, O ₂ is set in step 17.
It is determined whether the output of step 16 has been reversed, that is, the direction of increase or decrease of the feedback correction coefficient α has been reversed, and when this flow is repeated and reversed, it is the first time since it was determined to be steady in step 18, or the third time in the same area. Determine whether it is reversed or not, and if it is the third time, step
19 is the reference value α of the current feedback correction coefficient α ₁
The deviation Δα (α=α ₁ ) is temporarily stored as Δα ₁ . Thereafter, when the fourth reversal is detected, the process proceeds to steps 20 to 25 and learning is performed based on the data from the third reversal to the fourth reversal (see FIG. 6). Similarly, when a fifth or more reversal is detected, the program proceeds to steps 20 to 25 and performs learning based on the data from the previous reversal to the current reversal.

When reversing for the fourth time or more, in step 20, the deviation Δα of the current feedback correction coefficient α from the reference value α ₁ is calculated.
(=α−α ₁ ) is temporarily stored as Δα ₁ . The Δα ₁ and Δα ₂ stored at this time are the upper and lower peak values of Δα from the previous (for example, the third) reversal to the current (for example, the fourth) reversal, as shown in FIG.

Since the average value of the deviation Δα can be calculated based on these upper and lower peak values Δα ₁ and Δα ₂ ,
In step 21, the average value of the deviations Δα is calculated based on the following equation.

=(Δα ₁ +Δα ₂ )/2 Next, in step 22, the stored learning correction coefficient Kl corresponding to the current area is retrieved. However, in reality, the one searched in step 3 may be used.

Next, in step 23, the reference value α ₁ of the feedback correction coefficient α is set to the current learning correction coefficient Kl according to the following formula.
By adding a predetermined percentage of the average value of the deviation Δα (= α − α ₁ ) from
Calculate Kl _(oew) , correct and rewrite the learning correction coefficient data in the same area.

Kl _(oew) ←Kl+/M (M is a constant, M>1) After this, in step 24, the value of Δαα ₂ is substituted into Δα ₁ for the next calculation.

Next, in step 25, the learning update counter RC
Increment the value of RAM with backup
Remember it.

If it is determined in step 11 that the engine operating state is no longer in the same range as the previous time, the count value C is cleared and the flag F is reset in step 26.

Here, steps 17 to 24 correspond to learning correction coefficient correction means.

Next, the grid axis switching routine shown in FIG. 7 will be explained.

In step 31, the number of corrections RC of the learning correction coefficient stored in step 25 of the learning subroutine is calculated.
is compared with a predetermined value RCO, and if it is less than the predetermined value RCO, it is determined that the learning progress is low and the process proceeds to step 32, and if it exceeds RCO, it is determined that the learning progress is high and the process proceeds to step 33.

In step 32, the number of grid axes is reduced and set so that the area into which the operating conditions are divided becomes large. For example, as shown in Fig. 8, the operating conditions in which engine lambda control is performed are roughly divided into nine regions based on engine speed N, basic fuel injection price Tp,
Set three lattice axes for each.

Therefore, in step 11 of FIG. 5, the chance that the engine operating state (N, Tp) remains in the same region increases, and the speed at which the learning correction coefficient Kl is corrected, that is, the learning speed increases.

This allows learning to proceed evenly over the entire range, so that good air-fuel ratio control can be performed by learning control from the initial stage of engine operation.

On the other hand, if it is determined in step 31 that the learning progress speed is high and the process proceeds to step 33, the number of grid axes is increased and set so that the area into which the operating state is divided becomes small. For example, as shown in Figure 9, λ
Fifteen grid axes are set for each N ₁ Tp so that the operating state to be controlled is subdivided into 225 regions.

In this case, although the progress speed of learning is delayed by making the region smaller, the deviation in air-fuel ratio within the same region can be reduced, and the learning control accuracy can be improved.

In other words, at the beginning of learning, by proceeding with learning evenly, albeit somewhat roughly, in all areas, and then switching to high-precision learning, it is possible to perform good learning control corresponding to the elapsed driving time. be.

The function of this grid axis switching routine corresponds to a grid axis switching means.

In this example, the learning progress level was determined based on the total number of corrections of the learning correction coefficient Kl for all regions. A configuration may be adopted in which the degree of progress is determined for each area and the area is sequentially subdivided into small areas starting from the area with the highest degree of progress.

<Effects of the invention> As explained above, according to the invention, in the initial stage when the learning progress is low, the area for correcting and memorizing the learning correction amount is enlarged to increase the learning speed, and all areas are learned evenly. As the learning progress increases, highly accurate learning control can be performed by subdividing the region and performing learning for each small region, so it is possible to perform air-fuel ratio control using good learning control in response to elapsed driving time. can.

[Brief explanation of the drawing]

Figure 1 is a block diagram showing the configuration of the present invention;
3 is a block circuit diagram of the control unit shown in FIG. 2, FIG. 4 is a flowchart showing the fuel injection amount calculation routine of the same embodiment, and FIG. 5 is a block diagram showing an embodiment of the present invention. 6 is a flowchart showing the learning subroutine, FIG. 6 is a control characteristic diagram, FIG. 7 is a flowchart showing the grid axis switching routine, FIG. 8 is a region division diagram when the learning progress is low, and FIG. Figure 10 is a graph showing the relationship between the elapsed time and the number of learning times depending on the size of the area, and Figure 11 is a graph showing the space between the same areas depending on the size of the area. It is a graph showing the amount of deviation in fuel ratio. 1... Engine, 6... Air flow meter, 11...
...Fuel injection valve, 16...O ₂ sensor, 17... Crank angle sensor, 30... Control unit.

Claims (1)

[Claims for Utility Model Registration] A first detection means for detecting the engine intake air flow rate;
an engine operating state detecting means including at least a second detecting means for detecting the engine rotation speed, and a third detecting means for detecting an engine exhaust component and thereby detecting the air-fuel ratio of the engine intake air-fuel mixture; basic fuel injection amount calculation means for calculating a basic fuel injection amount based on the engine intake air flow rate outputted by the detection means and the engine rotational speed outputted by the second detection means; a rewritable storage means that stores grid axes dividing into a plurality of regions and learning correction coefficients for correcting the basic fuel injection amount for each region surrounded by these grid axes; and an actual engine operating state. learning correction coefficient retrieval means for retrieving a learning correction coefficient for a corresponding region from the storage means based on the learning correction coefficient; and a learning correction coefficient retrieval means for retrieving a learning correction coefficient for a corresponding region from the storage means; a feedback correction coefficient setting means for increasing or decreasing a feedback correction coefficient by a predetermined amount for correcting the basic fuel injection amount so as to bring it closer to the fuel ratio; a basic fuel injection amount calculated by the basic fuel injection amount calculation means; a fuel injection amount calculation means for calculating a fuel injection amount based on the learning correction coefficient searched by the learning correction coefficient search means and the feedback correction coefficient set by the feedback correction coefficient setting means; A fuel injection means for injecting fuel into the engine on and off in response to a drive pulse signal corresponding to the fuel injection amount; a steady state detection means for detecting that the actual engine operating state is in a steady state; Learning correction that learns the deviation from the standard value of the feedback correction coefficient for each corresponding region when in a steady state, and corrects and rewrites the learning correction coefficient corresponding to the region of the engine operating state in a direction to reduce this deviation. coefficient modifying means; learning progress determining means for determining the degree of learning progress based on the number of times the learning correction coefficient is modified by the learning correction coefficient modifying means; When it is determined that the learning progress is high, the lattice axes are set so that each region partitioned by the lattice axes of the storage means is relatively large, and when it is determined that the learning progress is high, each region is relatively large. A learning control device for an air-fuel ratio of an internal combustion engine, comprising: a lattice axis switching means for switching the lattice axis so that the lattice axis becomes smaller;