JPH04175435A - Air-fuel ratio learning controller for internal combustion engine - Google Patents

Abstract

PURPOSE:To suppress an error learning of the air-fuel ratio slippage in a transition and to prevent generation of a large air-fuel ratio difference by prohibiting the learning when the average value of the air-fuel ratio feedback correction value is at a specific level or higher. CONSTITUTION:Depending on the engine operation condition detected by a means A, a standard fuel feeding amount is set by a means B, and an air-fuel ratio feedback correction value to correct the standard fuel feeding amount is set by a means D by comparing the air-fuel ratio detected by a means C and the object air-fuel ratio. And an air-fuel ratio learning correction value to correct the standard fuel feeding amount is stored in a means E to rewrite freely, and the air-fuel learning correction value is corrected by a means F by learning the deflection of the air-fuel feedback correction value from the object convergent value. Furthermore, depending on the fuel feeding amount set by a means G, a fuel feeding means I is driven and controlled by a means H. And when the variation width of the average value of the air-fuel ratio feedback correction value is at a specific level or more, the correcting rewriting of the air-fuel ratio learning correction value is prohibited by a means J.

Description

DETAILED DESCRIPTION OF THE INVENTION <Industrial Application Field> The present invention relates to an air-fuel ratio learning control device for an internal combustion engine, and more particularly to a device that can prevent erroneous air-fuel ratio learning due to air-fuel ratio deviations during engine acceleration/deceleration. Regarding.

<Prior art> Conventionally, in an internal combustion engine equipped with an electronically controlled fuel injection device having an air-fuel ratio feedback correction control function,
As disclosed in Japanese Patent Laid-Open No. 0-90944, Japanese Patent Laid-Open No. 1981-190142, etc., learning control of the air-fuel ratio is employed.

The air-fuel ratio feedback correction control is performed by setting a basic fuel injection amount Tp, which is calculated from engine operating state parameters (for example, intake air flow rate Q and engine rotational speed N) that are related to the amount of air taken into the engine, in the engine exhaust system. The actual air-fuel ratio is corrected by the air-fuel ratio feedback correction coefficient LMD, which is set by proportional/integral control based on rich/lean relative to the target air-fuel ratio (for example, stoichiometric air-fuel ratio) determined by an oxygen sensor. It performs feedback control on the air-fuel ratio.

Here, the deviation from the reference value (target convergence value) of the air-fuel ratio feedback correction coefficient LMD is learned for each predetermined plurality of operating regions, and the learning correction coefficient KBLRC is calculated.
is determined, and the basic fuel injection amount Tp is determined based on the learning correction coefficient KBL.
The base air-fuel ratio obtained without the correction coefficient LMD is corrected by RC so that it substantially matches the target air-fuel ratio, and during air-fuel ratio feedback control, the fuel injection amount Ti is calculated by further correcting with the correction coefficient LMD. It is.

As a result, it is possible to perform correction corresponding to the required value of air-fuel ratio correction that differs for each operating condition, and the air-fuel ratio feedback correction coefficient L
It is possible to stabilize MD near the reference value and improve air-fuel ratio controllability.

<Problem to be solved by the invention> By the way, during transient engine operation, the influence of fuel wall flow,
Due to the difference in intake air flow rate between when the fuel injection amount Ti is set and when the intake valve is opened, the air-fuel ratio becomes leaner during acceleration, and conversely, the air-fuel ratio becomes richer during deceleration. As the degree of acceleration/deceleration increases, the lean or rich tendency becomes more pronounced.

In this way, when a relatively large air-fuel ratio deviation occurs during transient operation of the engine, the air-fuel ratio feedback correction coefficient LMD is set to increase or decrease in a direction that eliminates the air-fuel ratio deviation.
When the air-fuel ratio is equalized during engine acceleration, the air-fuel ratio feedback correction coefficient LMD behaves as shown in FIG. Sometimes it is done and remembered. Therefore, when deceleration operation or steady operation is performed in the region where lean air-fuel ratio due to acceleration operation has been learned, the fuel injection amount is determined based on the learned correction value, which is significantly different from the request, as shown in Fig. 13. will be corrected, and using the learned correction value will actually aggravate the air-fuel ratio deviation, causing a large air-fuel ratio difference and increasing dOlHC or NOx in the exhaust. there were.

In particular, when the map for storing the learning correction coefficient KBLRC is configured with a learning map that divides the entire driving range into small unit areas and a learning map that divides the entire driving area into larger unit areas, it is possible to With a map that learns the air-fuel ratio at each operating range, there are few learning opportunities for each operating range, and because the operating range is narrow, air-fuel ratio deviations are picked up spot-on. Therefore, there is a problem in that the next time the erroneous learning region is entered, a larger air-fuel ratio difference occurs.

The present invention was developed in view of the above problems, and when a large air-fuel ratio deviation occurs due to transient operation of the engine,
Based on this, it is an object of the present invention to provide an air-fuel ratio learning control device that can avoid learning that is significantly different from the learning value corresponding to steady operation.

<Means for Solving the Problems> Therefore, an air-fuel ratio learning control device for an internal combustion engine according to the present invention is configured as shown in FIG.

In FIG. 1, the engine operating condition detecting means detects the engine operating condition including at least an operating parameter related to the amount of air taken into the engine, and the basic fuel supply amount setting means
A basic fuel supply amount is set based on the detected engine operating conditions.

Further, the air-fuel ratio detection means detects the air-fuel ratio of the engine intake air-fuel mixture, and the air-fuel ratio feedback correction value setting means compares the detected air-fuel ratio with a target air-fuel ratio to set the actual air-fuel ratio to the target air-fuel ratio. An air-fuel ratio feedback correction value for correcting the basic fuel supply amount is set to increase or decrease so that it approaches the air-fuel ratio.

On the other hand, the air-fuel ratio learning correction value storage means rewritably stores an air-fuel ratio learning correction value for correcting the basic fuel supply amount for each predetermined operating region based on engine operating conditions, and the air-fuel ratio learning correction value is stored in a rewritable manner. The value correction means learns the deviation of the air-fuel ratio feedback correction value from the target convergence value, and corrects the air-fuel ratio learning correction value stored in the air-fuel ratio learning correction value storage means in a direction to reduce the deviation. rewrite.

The fuel supply amount setting means sets the final fuel supply amount based on the basic fuel supply amount, the air-fuel ratio feedback correction value, and the air-fuel ratio learning correction value for the corresponding operating region stored in the air-fuel ratio learning correction value storage means. A fuel supply amount is set, and the fuel supply means drives and controls the fuel supply means based on the set fuel supply amount.

On the other hand, the learning prohibition means sets the air-fuel ratio learning correction value by the air-fuel ratio learning correction value correction means when the range of change of the average value of the air-fuel ratio feedback correction value set by the air-fuel ratio feedback correction value setting means is equal to or greater than a predetermined level. Modification and rewriting of is prohibited.

Here, as shown by the dotted line in FIG. 1, it is preferable to provide a learning prohibition level setting means. a predetermined level in the learning inhibiting means is set for each predetermined operating region of the air-fuel ratio learning correction value storage means based on the ratio data and data of operating conditions that separate operating regions in the air-fuel ratio learning correction value storage means; do.

<Operation> According to this configuration, in air-fuel ratio learning in which the deviation of the air-fuel ratio feedback correction value from the target convergence value is learned and the air-fuel ratio learning correction value is corrected in a direction to reduce this deviation, the air-fuel ratio feedback correction value is Air-fuel ratio learning is prohibited when the range of change in the average value of is equal to or greater than a predetermined level.

Therefore, it is possible to prohibit learning when the air-fuel ratio feedback correction value changes significantly to compensate for a larger air-fuel ratio deviation caused by transient operation rather than a deviation in the base air-fuel ratio, and learning is not performed during transient operation. The learning results of the lever can be used later to prevent large air-fuel ratio differences from occurring.

Additionally, since the tendency of deviation in the base air-fuel ratio varies depending on operating conditions, there are changes in the air-fuel ratio feedback correction value due to transient operation that should prohibit learning, and changes in the air-fuel ratio feedback correction value based on fluctuations in the base air-fuel ratio that should be learned. It is desirable to change the predetermined level that distinguishes between Since the predetermined level is set for each predetermined operating region of the air-fuel ratio learning correction value storage means based on the data of the operating conditions that divide the operating regions in the means, the divided operating regions in the storage means can be changed in various ways. However, it is possible to easily set the predetermined level for each operating region.

<Example> The present invention will be explained in detail below.

In FIG. 2 showing one embodiment, an internal combustion engine 1 is connected to an air cleaner 2 through an intake duct 3. Air is sucked in through the throttle valve 4 and the intake manifold 5. A fuel injection valve 6 as a fuel supply means is provided in a branch portion of the intake manifold 5 for each cylinder. The fuel injection valve 6 is a normally closed type electromagnetic fuel injection valve that opens when the solenoid is energized and closes when the energization is stopped. When the valve is opened, fuel is injected and supplied under pressure from a fuel pump (not shown) and adjusted to a predetermined pressure by a pressure regulator.

Each combustion chamber of the engine 1 is provided with a spark plug 7, which ignites a spark to ignite and burn the air-fuel mixture. and,
From the engine 1, an exhaust manifold 8, an exhaust duct 9. Exhaust gas is discharged via the three-way catalyst lO and the muffler 11.

The control unit 12 includes a CPU and a ROM.

It is equipped with a microcomputer including a RAM, an A/D converter, and an input/output interface, and receives input signals from various sensors, performs arithmetic processing as described below, and controls the operation of the fuel injection valve 6. .

As the various sensors mentioned above, an air flow meter 13 is provided in the intake duct 3 and outputs a signal corresponding to the intake air flow rate Q of the engine 1.

Further, a crank angle sensor 14 is provided, and in the case of a four-cylinder engine, outputs a reference signal REF for each crank angle of 180 degrees and a unit signal PO8 for every crank angle of 10 or 2 degrees. Here, the engine rotation speed N can be calculated by measuring the period of the reference signal REF or the number of occurrences of the unit signal PO8 within a predetermined time.

Further, a water temperature sensor 15 is provided to detect the cooling water temperature Tw of the water jacket of the engine 1.

Here, the air flow meter 13. Crank angle sensor 14. The water temperature sensor 15 and the like correspond to engine operating condition detection means.

Further, an oxygen sensor 16 as an air-fuel ratio detection means is provided at the gathering part of the exhaust manifold 8, and detects the air-fuel ratio of the intake air-fuel mixture via the oxygen concentration in the exhaust gas. The oxygen sensor 16 is a known sensor that detects whether the actual air-fuel ratio is rich or lean with respect to the stoichiometric air-fuel ratio by utilizing the fact that the oxygen concentration in the exhaust gas changes suddenly after the stoichiometric air-fuel ratio.

Here, the CPU of the microcomputer built in the control unit 12 performs arithmetic processing according to the programs on the ROM shown in the flowcharts of FIGS. 3 to 9, and performs air-fuel ratio feedback correction control and air-fuel ratio The fuel injection amount Ti is set while executing the learning correction control, and the fuel supply to the engine l is controlled.

In this embodiment, basic fuel supply amount setting means, fuel supply amount setting means, fuel supply control means, air-fuel ratio feedback correction value setting means, learning correction value correction means, learning prohibition means, and learning prohibition level setting means are provided. The function is the third
As shown in the flowcharts in FIGS. 9 to 9, it is provided in terms of software, and the RAM of a microcomputer (not shown) built into the control unit 12 corresponds to the air-fuel ratio learning correction value storage means.

The program shown in the flowchart of FIG. 3 is a program that sets the air-fuel ratio feedback correction coefficient LMD (air-fuel ratio feedback correction value), which is multiplied by the basic fuel injection amount Tp, by proportional/integral control. (1 rev). Note that the initial value of the air-fuel ratio feedback correction coefficient LMD is l.

First, step l (indicated as Sl in the figure).

The same applies hereafter) reads the voltage signal output from the oxygen sensor (0□/S) 16 according to the oxygen concentration in the exhaust gas.

Then, in the next step 2, the voltage signal from the oxygen sensor 16 read in step 1 is compared with a slice level (for example, 500 mV) corresponding to the target air-fuel ratio (stoichiometric air-fuel ratio), and the engine intake air-fuel mixture is determined. It is determined whether the air-fuel ratio is rich or lean with respect to the target air-fuel ratio.

When the voltage signal from the oxygen sensor 16 is greater than the slice level and it is determined that the air-fuel ratio is rich, the process proceeds to step 3, where it is determined whether or not the current rich determination is the first.

When the rich determination is made for the first time, the process proceeds to step 4, where the air-fuel ratio foot hack correction coefficient LN·ID set up to the previous time is set to the maximum value a.

The fact that this is a rich determination or the first time means that a lean determination was made until the previous time, and as a result, an increase control of the air-fuel ratio feedback correction coefficient LMD (=increase correction of the fuel injection amount Ti) was performed. When the rich determination is made, the correction coefficient LMD is then controlled to decrease, so the value before the first reduction control of the rich determination is the maximum value of the correction coefficient LMD.

In the next step 5, a predetermined proportionality constant P is subtracted from the previous correction coefficient LMD to control the correction coefficient LMD to decrease. Further, in step 6, 1 is set to "P addition" which is a flag indicating that proportional control has been executed.

On the other hand, if it is determined in step 3 that it is rich or not the first time, the process proceeds to step 7, and the value obtained by multiplying the integral constant I by the latest fuel injection amount T1 is used as the correction coefficient L up to the previous time.
The correction coefficient LMD is updated by subtracting it from MD, and the reduction control by lXTi is repeated in step 7 every time this program is executed until the air-fuel ratio is eliminated from the rich state and reversed to lean.

In addition, when it is determined in step 2 that the air-fuel ratio is close to the target, in the same way as in the case of rich determination, first, in step 8, it is determined whether the current ratio is 1 or it is the first time. If it is the first time, the process proceeds to step 9 and sets the correction coefficient LMD up to the previous time, that is, the correction coefficient LMD that was controlled to decrease gradually at the time of rich judgment, to the minimum value. Correction coefficient LMD
By adding and updating the proportional constant P, the fuel injection amount Ti is increased and corrected, and in step 11, the P addition, which is a flag indicating that proportional control has been executed, is set to 1.

If it is determined in step 8 that the lean determination has not been made for the first time, the process proceeds to step 12, where the value obtained by multiplying the integral constant I by the latest fuel injection amount Ti is added to the previous correction coefficient LMD, and the correction coefficient LMD is Increase gradually.

When the proportional control of the correction coefficient LMD is executed in the first rich/lean determination, various processes related to the air-fuel ratio learning correction control, which will be described later, are further performed.

In this embodiment, as shown in FIG. 10, two learning maps for the air-fuel ratio learning correction value are provided, which divide the entire operating range using the basic fuel injection amount Tp and the engine rotational speed N as parameters. One is a 4x4 grid that divides the entire operating area into 16 unit areas, and the other is a 16x16 grid that divides the entire operating area into 256 unit areas. One operating region is further subdivided into 16 4×4 regions using a 16×16 grid learning map. Further, as described above, the air-fuel ratio learning correction value is also set to be applied to the entire operating range without dividing the operating range.

First, in step 13, the range of change Δα of the average value (a+b)/2 of the air-fuel ratio feedback correction coefficient LMD is calculated according to the following formula.

In the above calculation formula, A B dmy is the average value (='(a+b)/2) of the correction coefficient LMD obtained when the proportional control of the correction coefficient LMD was performed last time, and the average value is calculated every time the proportional control of the correction coefficient LMD is performed. The value A B dmy is found, and
The absolute value of the difference between the previous value and the current value of this average value ABdmy is determined as the range of change Δα.

As will be described later, the range of change Δα is set to prohibit learning when the air-fuel ratio feedback correction coefficient LMD, which is used to correct air-fuel ratio deviations caused by transient operation, changes significantly.

In the next step 14, the latest average value of the correction coefficient LMD=(a+b)/2 is set to
Set to dmy.

In the next step 15, a count value cnt is determined to determine whether or not the vehicle is stably stationary in one operating region on the 4×4 grid map.

In the flowchart of FIG. 4, which will be described later, when the corresponding operating region is changing at predetermined minute intervals among the operating regions divided by 4×4 grids, the count value cnt
is set to a predetermined value (for example, 4), and if it is determined in step 15 that the count value cnt is not zero, the process proceeds to step 16 to decrease the count value cnt by 1. After coming to stop in one driving area in the 4x4 grid, the count value cnt
is reduced by 1 every time the correction coefficient LMD is proportionally controlled, and when the count value cnt is zero, it can be considered that it is stably stopped in one operating region on the 4 x 4 grid. This is so that it can be done.

By determining whether or not the count value cnt is zero, it is determined whether or not to perform learning update as described later, and at the initial stage when the corresponding operating region on the 4×4 grid changes. This prevents learning from occurring.

In step 17, as will be described later, a flag is determined, which is set to 1 when most of the air-fuel ratio learning correction values corresponding to the respective operating regions of the 4×4 lattice learning map have been learned.

When the flag flag is l and the learning of the 4×4 lattice learning map is almost completed, the process advances to step 18.

In step 18, it is determined whether or not the map remains in the same area as the previous time on the 4×4 grid map, and only if it is not the same area as the previous time, the process proceeds to step 19.

In step 19, find the difference between the absolute value of the deviation of the latest correction coefficient LMD average value from the target convergence value Target (= 1.0>) and the absolute value of the previous deviation,
Based on this value, Δstress is set by referring to a Δstress map that indicates the degree of inappropriateness of the learned value. Here, the Δ stress is set to a larger value when the latest average value of the correction coefficient LMD changes significantly.

In step 17, it is determined that most of the air-fuel ratio learning correction values corresponding to the respective operating regions of the 4x4 lattice learning map have been learned.
Should the correction coefficient L I'vi D be approximately stable even if the corresponding operating region changes on the grid map?
When the operating area changes on the 4-grid map, the correction coefficient L
If MD changes significantly, it is assumed that there is a defect in learning, and the Δ stress is set larger.

Furthermore, in step 19, the obtained Δ stress is integrated,
The integration result is stored as stress.

As will be described later, when this stress exceeds a predetermined value, it is determined that the already learned air-fuel ratio learning correction coefficient is inappropriate, and the learning is performed again.

The program shown in the flowchart of FIG. 4 is an air-fuel ratio learning program for each operating region, and is executed at predetermined minute intervals (for example, 10 m5).

In step 21, the flag "P distribution force old" which is set to 1 when the air-fuel ratio feedback correction coefficient LMD is proportionally controlled in the flowchart of FIG. Proceed to step 22 P
After resetting the minute addition to zero, various processes are performed by this program, and when it is zero, the program is immediately terminated.

When the P addition is reset in step 22, in the next step 23, a flag indicating whether the learning correction coefficient KBLRCφ (initial value l), which is an air-fuel ratio learning correction value common to all operating ranges, has been learned or not. Determine Fφ.

Here, if the flag Fφ is set, the learning correction coefficient KBL
I? If learning of Cφ has not been completed, step 24
Proceed to the maximum and minimum values a, b of the correction coefficient LMD.
It is determined whether the average value (-(a + b)/2) is approximately l.

If it is not (a + b)/2 or approximately I, the process proceeds to step 26, and the predetermined coefficient The value multiplied by the previous learning correction coefficient KBL
RCφ and the addition result as a new learning correction coefficient KB.
Set as LRCφ.

Also, in step 26, a 4x4 grid map and a 16x
Learning correction coefficient KBLRC1, learning correction coefficient KBLRC stored in each operating area of the 16 grid map
2 are all reset to the initial value l. Therefore, when learning and updating the learning correction coefficient KBLRCφ, even if learning values are stored in the 4×4 grid map and the 16×16 grid map, learning of the learning correction coefficient KBLRCφ is performed with the data reset. It is something.

When it is determined in step 24 that (a+b), /2 is approximately 1, the flag Fφ is set to 1 in step 25, and the learning correction coefficient KBLRC corresponding to the entire driving range is set.
It is possible to determine that the learning of φ has been completed, in other words, that the air-fuel ratio feedback correction coefficient LMD has converged to approximately 1 as a result of learning and updating the learning correction coefficient KBLRCφ.

On the other hand, if it is determined in step 23 that the flag is Fφ or l, the learning correction coefficient K corresponding to the entire operating range
Since this indicates that the learning of BLRCφ has been completed, the operating region is divided into a plurality of regions based on the basic fuel injection amount Tp and the engine rotational speed N, and air-fuel ratio learning is performed for each region.

First, from step 28 onwards, it is determined whether the current operating condition is included in the area in the learning map in which the operating area is divided into 16x16 grids based on the basic fuel injection amount Tp and the engine rotational speed N.

Here, in step 28, a cello is set to a counter value 1 for determining which number of the 16 grids the current basic fuel injection amount (basic fuel supply amount) Tp is included in, and in the next step 29, determines whether the counter value l exceeds 15 or not, and if it does not exceed 15, step 3
0, a preset threshold value Tp (i) of the basic fuel injection amount Tp corresponding to the count value i is compared with the most recently calculated basic fuel injection amount Tp.

When it is determined in step 30 that the latest basic fuel injection amount Tp is smaller than the threshold value Tp (i), the process proceeds to step 33 and the count value i at that time is determined to include the latest basic fuel injection amount Tp. Set to I as the region position. That is, the maximum basic fuel injection amount Tp of each region is set in advance as a threshold value Tp (i), and the decisive basic fuel injection amount Tp and the threshold value Tp (i) are set on the smaller side (or thicker side). Tp (i)
This is set to ■ to indicate the number of the i or Tp block at the time when >Tp.

Further, when it is determined in step 30 that Tp (i)≦Tp, the process proceeds to step 31, where the count value i is incremented by 1, and Tp(i), which is one step larger, is compared with the latest Tp. Make it.

Then, when the count value i is counted up to 16 in step 31, 16 grids from 0 to 15 (16
In this state, a basic fuel injection amount Tp larger than the initial maximum value of the basic fuel injection amount Tp range divided into blocks) has been calculated, and at this time, i is set to the maximum value 15 in step 32, and then step Proceed to step 33,
Maximum Tp area of initialized Tp block Current T
Assume that p is included.

Next, to divide into 16 blocks based on engine speed N,
Similarly to the block determination of the basic fuel injection amount Tp,
The block number containing the latest engine rotational speed N is determined based on the count value. First, in step 34, the count value k is initially set to zero, and until it is determined in step 35 that the count value k exceeds 15, the comparison with the threshold value N (k) is performed in step 36. and set the count value k when N (k)>N for the first time to K indicating the block number of N in step 39.
When N(k)≦N, the count value k is incremented by 1 in step 37. Also, the count value k is 1
When it exceeds 5, the process proceeds to step 38, sets the count value k to 15, and proceeds to step 39.

In this way, the basic fuel injection amount Tp and the engine rotation speed N
In which operating region of the learning map divided into 256 operating regions of 16×16 grids includes the current operating condition, the coordinates ( K, ■).

The corresponding position of the 16X16 grid is as shown above (see below).
If it is clear, as shown in FIG. 10, one operating area in the 4 x 4 grid operating area map is divided into 4 x 4 = 16 areas in the 16 x 16 grid operating area as one program, so the above I. , based on K4
The location corresponding to the current driving condition can be identified on the ×4 grid learning map.

That is, in step 40, the block number ■ of Tp is divided by 4, and the integer value obtained by rounding down the decimal point of the result is set to A, and in step 41, the block number K of N is similarly divided. Divide by 4, and set the resulting integer value, rounded down to the decimal point, to B. As a result, the area position of the 4x4 grid map to which the current operating conditions apply is CB.

A] coordinates.

In the next step 42, using CB, A) indicating the position of the corresponding area on the 4x4 grid map, the above-mentioned Add the value obtained by multiplying A by 16 and B and use the result as A
Set to B.

Then, in step 43, by comparing ABOLD, which is the AB calculated last time, with the latest AB calculated this time, it is determined whether or not the current operating range is the same as the previous time. AB: AB OLD and 4
If the operating region is different from the previous time on the x4 grid, a predetermined value (for example, 4) is set to the count value cnt in step 44.

In step 45, AB calculated in step 42 this time is set to AB,L as the previous value for the next determination in step 43.

In step 46, in the 4x4 grid map, [B, A
) is used as the coordinates to determine whether or not the air-fuel ratio learning has been completed in the current operating condition or operating range that is included.
) or zero and learning has not been completed in one driving region on the 4×4 grid map that includes the current driving conditions, the process advances to step 47.

In step 47, it is determined whether or not the count value cnt is zero, and if the count value cnt is not zero and there is a change in the corresponding area in the 4x4 grid map, the program is terminated and the count value cnt is The process proceeds to step 48 only when the corresponding operating region on the 4×4 grid map is stable.

In step 48, the average value (a+b)/2 of the maximum and minimum values a and b of the air-fuel ratio feedback correction coefficient LMD sampled in the flowchart of FIG. 3, that is, the center value of the correction coefficient LMD, or the target convergence value Target
The progress of learning is determined based on whether or not the value is near the initial value (=1.O), and if it is not recognized as being approximately 1.0 and learning has not been completed, the process proceeds to step 50 with a hegejump, and when the value is approximately 1. If there is, and it is recognized that the learning has been completed, the flag F (B, A) is set to 1 in step 49, and the process proceeds to step 50.

In step 50, a slice level 5ysE (predetermined level) is used to determine whether the change width Δα of the average value of the air-fuel ratio feedback correction coefficient LMD obtained in the flowchart of FIG. 3 is greater than or equal to a predetermined level.
CB, which includes the current operating conditions in a 4x4 grid, A) Data M stored corresponding to the area.
Set idE (B, A).

Then, in the next step 51, the change range Δα calculated most recently in the flowchart of FIG. 3 is compared with the slice level 5ysE set in the step 50,
When the variation width Δα is greater than or equal to the slice level 5ysE, the current operating condition is included (B.

A] Skip forward to step 53 without performing learning corresponding to the territory.

Here, as mentioned above, if learning is prohibited when the range of change △α of the average value of the air-fuel ratio feedback correction coefficient LMD is larger than a predetermined level, it is possible to prevent response delays in fuel control during transient operation. Learning will not be performed when the air-fuel ratio deviates significantly from the target air-fuel ratio due to the influence of air flow, etc., and the air-fuel ratio feedback correction coefficient LMD changes significantly to correct such air-fuel ratio deviation. Therefore,
It is possible to avoid learning the air-fuel ratio deviation due to transient operation. For example, it may be a result of learning the air-fuel ratio deviation during acceleration (leaning), or it may be used in deceleration driving (riching operation) where the correction direction is completely different. This can prevent the occurrence of large air-fuel ratio differences.

Further, error rate data M determined for each operating region at the slice level 5ysE for determining the range of change Δα
If idE (B, A) is set, it is possible to accurately distinguish between changes in the correction coefficient LMD due to deviations in the base air-fuel ratio and changes in the correction coefficient LMD due to air-fuel ratio deviations caused by transient operation. , it is possible to secure learning opportunities while preventing erroneous learning of air-fuel ratio deviations due to transient operation.

Note that the slice level 5y used to determine the variation range Δα is
The setting of Mi dE (B, A) set to sE will be explained in detail later according to the flowchart of FIG.

On the other hand, in step 51, it is determined whether the change range Δα or the slice level 5
If it is determined that it is less than ysE, the process proceeds to step 52 to perform air-fuel ratio learning in the CB, A) region of the 4×4 grid map.

In step 52, the current C
B. For the learning correction coefficient KBLRCI stored corresponding to the A) area, the target convergence value Target (l, 0 in this example) is subtracted from the average value of the maximum and minimum values a and b. Add the values multiplied by a predetermined coefficient
A new learning correction coefficient KBLRCI corresponding to .

During the learning of the CB on such a 4x4 grid map, the learning correction coefficient KBLRC2 of each region included in the CB and A) region on a finer 16x16 grid map during the learning of the A) region is calculated in step 53. are all set to initial value l.

As mentioned above, when there is a region in the 4x4 grid map where learning has not been completed, when the operating region becomes stable, the predetermined percentage of the deviation from the target value Target of (a+b)/2 is calculated by Stored learning correction coefficient KBL
By adding and updating the RCI, the learning correction coefficient KBLR is used instead of the air-fuel ratio feedback correction coefficient LMD.
It is assumed that the target air-fuel ratio is obtained through correction by CI, and that the learning of the operating region is completed when the air-fuel ratio feedback correction coefficient LMD substantially converges to the initial value 1, which is the target convergence value.

On the other hand, in step 46, if it is determined that the flag is FEB, A) or l, and if the learned correction coefficient KBLRCI is currently stored in the corresponding driving region of the 4x4 grid map, then the learning correction coefficient KBLRCI is stored. The operating area (B, A
) is further subdivided into 16 regions using a 4x4 grid.
Move on to learning the lattice learning map.

In step 54, the average value of the correction coefficient LMD (a
+b)/2 determines whether or not it approximately matches the target convergence value of 1, and determines whether (a + b)/2 is not approximately 1 and requires correction by the air-fuel ratio feedback correction coefficient LMD. If so, the process advances to step 55.

In step 55, "
DatE (K, I) stored as a predetermined level for determining the variation range Δα is set to slice level 5.
Set to ysE.

Then, in step 56, similarly to step 51, the range of change Δα obtained from the flowchart of FIG.
The slice level 5ysE set in step 55 is compared, and only when the variation width Δα is less than the slice level 5ysE, the process proceeds to step 57 to perform learning in the corresponding area of the 16×16 lattice map, and the variation width Δα If the slice level is 5ysE or higher, the process is terminated without performing learning.

Therefore, in the learning of the 16x16 grid map, as in the case of the 4x4 grid map learning described above, it is possible to prevent the air-fuel ratio deviation that occurs during transient operation, which is different from the deviation of the base air-fuel ratio, from being learned. It is.

In step 56, it is determined that Δα<5ysE, and the process proceeds to step 57. In step 57, (a+b)/
2 to the target convergence value Target (1.0 in this example)
) is multiplied by a predetermined coefficient X2, and the value is 16X
It is added to the learning correction coefficient KBLRC2 (K, I) stored corresponding to the operating region (K, I) that includes the current operating conditions of the 16-lattice learning map, and this addition result is added to the corresponding operating region (K, I). Set as a new correction coefficient KBLRC2(K, I) in I).

On the other hand, if it is determined in step 54 that (a+b)/2, which is the average value of the correction coefficient LMD, substantially matches 1, which is the target convergence value Target, the process proceeds to step 58;
A flag FP (K, I) is set to 1 so that it is determined that learning of the current operating condition of the 16×16 grid map or the operating region (K, I) included in the map has been completed.

Then, from step 59 onward, it is determined that the current learning has been completed, and the corresponding flag FF (K, 1) is set to 1 in the predetermined operation area (K,
Based on I), if there is an operation area on the 16x16 grid map adjacent to this area (K, I) for which learning has not been completed (see Figure 11), the current corresponding area is added to that operation area. Control is performed to store the learning correction coefficient KBLRC2 stored corresponding to (K, T).

In step 59, in order to indicate the position of the area containing the current operating conditions in the 16x16 grid map,
The values obtained by subtracting l from `` are set to m and n, respectively, and in the next step 60, it is determined whether m=+2 or not.

When the process advances from step 59 to step 60, a negative determination is made in step 60, so the process proceeds to step 61 to complete the learning of one operating region of the 16×16 grid indicated by (m, n). It is determined whether the flag FF(m, n) is 1 or zero.

Here, if the flag FF (m, n) is zero and learning has not been completed, the process advances to step 62.

In this step 62, the area address (m, n) on the 16x16 grid is converted to the area address [m, n) on the 4x4 grid.
/4. area CB on the 4×4 grid map, which is currently determined to be applicable.

A].

In other words, (K, I) is a region included in CB, A], or the region surrounding (K, I) is included in a region adjacent to CB, A) on a 4x4 grid map. If the area is included in the same (B, A) ((m/4.n/4)=CB, A), the process proceeds to step 63, and it is determined that the area has been learned this time. (K
, I) Let the learning correction coefficient KBLRC2 corresponding to the region be the learning value of the (m, n) region as it is.

On the other hand, in step 62, (m) adjacent to (K, I).

When it is determined that the area (m, n) is included in a different area on the 4x4 grid map, the process proceeds to step 64, and the learning correction coefficient KBL calculated by the following formula is applied to the area (m, n).
Store RC2.

KBLRC2Cm, n) ← KBLRCI (B, A) +KBLRC2(K, I
) - KBLRCI (m/4.n/4) The formula for calculating the above KBLRC2 (m, n) is:
[K, I) and (m, n) are adjacent areas on a 16x16 grid, so the final correction request is based on the assumption that they should be similar.
The learning correction coefficient K of each included 4×4 grid map
Since the BLRCI is different, each KBLRCI is different.
(B, A), KBLRCI (m/4,
n/4) or approximated as shown in the following formula.

KBLRCl(B, A) +KBLRC2(K,
I) = KBLRCI (m/4, n/4
) 10 KBLRC2 (m, n) As described above, when the area (m, n) has been learned, the learned value is not updated, and when it is unlearned, niha, K
BLRC2(K, I) based on KBLRC2(
When m, n) are updated, in step 65, m is incremented by 1, and the process returns to step 60 again, until m=+2, that is, with n constant, m is a range of +1 centered around K. , and determines whether it has been learned or not for each driving region.

Then, as a result of the 1-up processing of m in step 65, when it is determined in step 60 that m=+2, the process proceeds to step 66, and it is determined whether n=1+2, and n≠1+2. If so, m is set to -1 again in step 67, and n is incremented by 1 in step 68, after which the process proceeds to step 61.

Therefore, initially, let n=1-1 and m is centered at K±
Next, we set n=I and moved m within a range of ±1 around K, and then set n=I+1 to distinguish adjacent areas. The learning correction coefficient KBLRC2 (K, I) is changed when the eight operating ranges surrounding (K, I) (see Figure 11) have not been learned.
) as the learning correction coefficient KBLRC for that operating region.
2(m, n).

In this way, if the learning results of the learned area are applied to the surrounding unlearned areas, even if the driving area is subdivided like a 16x16 grid and there are few learning opportunities for each driving area, It is possible to prevent differences in air-fuel ratio controllability from occurring between operating ranges.

When it is determined in step 66 that n=I+2, it means that the determination processing for all eight operating regions surrounding (K, I) has been completed, so at this time, step 5
Proceed to step 5 and select the learning correction coefficient KBLRC2, which is determined to have already been learned in this area (K, I).
The learning update is performed depending on whether the range of change Δα of the average value of the air-fuel ratio feedback correction coefficient LMD is greater than or equal to a predetermined value.

In this way, in this embodiment, first, the learning correction coefficient KBLRCφ corresponding to all driving regions is learned, and then learning is performed for each driving region divided by a 4×4 grid learning map. For areas that have already been learned using a 4 x 4 grid learning map, we have divided that area into 4 x 4 grids for learning, so learning progresses from large driving areas to small driving areas. Therefore, convergence of the air-fuel ratio is ensured by learning over a large operating range, and as learning progresses, learning is performed for each operating range in detail, so the difference in required correction values due to different operating ranges is Can respond with high precision.

Here, DatE (K, I), Mi
The setting of dE (B, A) will be explained according to the flowchart of FIG.

The program shown in the flowchart of FIG. 9 is executed every time the power is turned on to the control unit 12. First, in step 151, i and j for counting grid positions in a 16×16 grid are each set to zero. Reset.

Then, in step 152, it is determined whether the above-mentioned i is 14 or less, and if i is 14 or less, the process proceeds to step 153.

In step 153, threshold values 'rp(i) and N(j) of the basic fuel injection amount Tp and the engine rotational speed N at the grid point (j, i) indicated by the counters i and j are set to Tpdmy and Ndmy, respectively. At the same time, the threshold values Tp (i+1) and N (j+1) of Tp and N at the grid point indicated by the counter i+1.j+1 are set to Tpdm2.Ndm2, respectively.

In step 154, the intake air flow rate Qdmy at the grid point indicated by the counters i, j, and the counter i+l,
Intake air flow rate Qdm at the grid point indicated by j+1
2 is calculated according to the following formula.

The basic fuel injection amount 'rp is T p-K x Q/
Since it is calculated as N (K is a constant), the intake air flow rate Q corresponding to each grid position is estimated and set using the above equation.

In the next step 155, the battery voltage correction numerator s is added to the basic fuel injection amounts Tpdmy and Tpdm2 at the grid points i and i+1, respectively, to obtain the fuel injection amount T i dmy at each grid point.

Set Tidm2.

In step 156, the fuel injection amount Tidmy, Ti
Error rate ETi corresponding to m2.

Find ETi2 by searching.

The map of the predicted maximum error rate (predicted error rate data) ETi is created by determining in advance the maximum predicted error rate of the actual injection amount with respect to the drive pulse signal given to the fuel injection valve 6. In step 156, each fuel injection amount Tidmy
, Tidm2, the air-fuel ratio error rate that is predicted to occur due to the fuel injection valve 6 (fuel supply system) is determined.

In step 157, the intake air flow rates Qdmy and Qdm2 obtained in step 154 are calculated from the map of the predicted maximum error rate EQ that is preset according to the intake air flow rate Q.
Search and find the error rates EQ and EQ2 corresponding to .

The map of the predicted maximum error rate (predicted error rate data) EQ is obtained by calculating in advance the maximum detection error rate predicted in the air flow meter 13, and in step 157, each intake air Flow rate Qd
The air-fuel ratio error rate that is predicted to occur due to the detection error of the air flow meter 13 in my, Qdm2 is determined.

Then, in step 158, the absolute value ΔQ of the error rate difference between the grid points is calculated from the predicted maximum error rate data at the adjacent grid points obtained as described above.
, ΔTi is the error rate ETi.

It is obtained from ETi2, EQ, and EQ2 as follows.

ΔQ-I EQ-EQ21 ΔTi←1ETi-ETi21 The above ΔQ and ΔTi indicate the expected variation in error rate in the operating region between grid points, and if the operating conditions change within the region, , said ΔQ,
The error corresponding to ΔTi indicates the possibility that the error may occur as a deviation in the base air-fuel ratio due to the fuel injection valve 6 or the air flow meter 13.

In other words, each learning region includes different operating conditions, and the required correction amount is slightly different depending on each operating condition. There is a possibility that an error equivalent to , ΔTi may occur. Conversely, if an air-fuel ratio error exceeding the error equivalent to ΔQ, ΔTi occurs, it is not due to a shift in the base air-fuel ratio, but rather due to transient operation that the air-fuel ratio has changed significantly. It can be assumed that there is a deviation.

In step 159, the magnitude relationship between ΔQ and ΔTi is determined, and in step 160 or step 161, the relationship between ΔQ and ΔTi is determined.
The larger of Ti is set as the final error rate Er.

Then, in step 162, in the area (j, i) on the 16×16 grid map indicated by the current i, j, the maximum error rate DetE (j, i) that can occur due to the deviation of the base air-fuel ratio is determined by Set.

In the next step 163, i is incremented by 1, and the process returns to step 152 again.

When i is increased by 1 while j is fixed at zero and the maximum error rate Er of the base air-fuel ratio predicted in each operating region is set as described above, the process proceeds from step 152 to step 164. .

Here, until j exceeds 14, in step 165, i
is reset to zero, j is incremented by 1 in step 166, and the process returns to step 152.
As a result, the operation of fixing j and counting up i from zero to 14 is repeated, and the maximum error rate Er of the base air-fuel ratio predicted in the entire area of the 16×16 grid map is determined.

maximum error rate E in a 16x16 grid map
When the setting of r is completed, proceed to step 167 and subsequent steps to create each area of the 4×4 grid map by the same operation (however, i and j are upped three times at a time in step 179 or step 182). Err corresponding to the predicted maximum variation amount of the base air-fuel ratio is set for each region, and M i d E (j / 4, i /
4) is determined (steps 167 to 182).

When comparing the range of change Δα of the average value of the air-fuel ratio feedback correction coefficient LMD with the slice level 5ysE, the range CB includes the operating conditions at that time.

A) Mi stored corresponding to (K, I) respectively
dE (B, A) and DetE (K, I) will be set to the slice level 5ysE, and the range of change in the correction coefficient LMD based on the base air-fuel ratio deviation caused by the fuel injection valve 6 and air flow meter 13. It is possible to determine the movement of the correction coefficient LMD due to transient operation exceeding Δα. This prevents the air-fuel ratio deviation due to transient operation from being learned instead of the deviation of the base air-fuel ratio, and as mentioned above, the movement of the correction coefficient LMD in each operating region is caused by the air-fuel ratio deviation due to transient operation. Since it is possible to accurately determine whether or not the cause is caused by the problem, it is possible to prevent learning opportunities from being lost.

In this embodiment, the program shown in the flowchart of FIG. 9 is executed every time the control unit 12 is powered on, but the same result is obtained each time
E (B, A), De tE (K.

■], so the Mid
E (B, A), De tE CK, I) may be stored in the ROM.

Three learning correction coefficients KBL learned as above
The final air-fuel ratio learning correction coefficient KBLRC based on RCφ, KBLRCI, and KBLRC2 is set according to the flowchart in FIG.

The flowchart in FIG. 8 is processed as a pack ground job, and first, in step 71,
The learning correction coefficient KBLRCI stored in area 1:B, A) of the 4x4 lattice map is read out, and in the next step 72, the learning correction coefficient KBLRCI stored in area [K, I) of the 16x16 lattice map is read out. Read out. In addition, B×40A and KX16+I shown in the flowchart
converts the address indicating each grid position into an address in memory.

In step 73, KBLRCφ+KBLRC1+K
The final learning correction coefficient KBLRC is set as BLRC2-2,0→KBLRC.

The learning correction coefficient KBLRC set in the flowchart of FIG. 8 is used to calculate and set the fuel injection amount Ti in the program shown in the flowchart of FIG.

The program shown in the flowchart of FIG. 5 is executed at predetermined minute intervals (for example, tOms). First, in step 81, the intake air flow rate Q detected by the air flow meter 13 and the crank angle sensor 14 are calculated. Input the engine rotation speed N calculated based on the detection signal.
゛Then, in the next step 82, based on the intake air flow rate Q and engine rotational speed N input in step 81, the basic fuel injection amount T corresponding to the intake air flow rate Q per unit rotation is calculated.
p (←KxQ/N; is a constant) is calculated.

In the next step 83, various corrections are made to the basic fuel injection amount Tp calculated in step 82 to calculate the final fuel injection amount (fuel supply amount) Ti.

Here, the correction value used for correcting the basic fuel injection amount Tp is the learning correction coefficient KBLRC which is learned and set according to the flowchart in FIG. 4, the air-fuel ratio feedback correction coefficient LMD which is calculated according to the flowchart in FIG.
Various correction coefficients C0FF are set including a basic correction coefficient based on the cooling water temperature Tw detected by the water temperature sensor 15, a post-start increase correction coefficient, etc., and changes in the effective injection time of the fuel injection valve 6 due to changes in battery voltage are A correction molecule s for correction, Ti4-TpxLMDxKBLRCxc
The final fuel injection amount Ti is updated at predetermined intervals by calculating OEF+Ts.

At a predetermined fuel injection timing, the control unit 12 outputs a drive pulse signal with a pulse width corresponding to the latest calculated fuel injection amount Ti to the fuel injection valve 6 to control the amount of fuel supplied to the engine 1. do.

The program shown in the flowchart of FIG. 6 is a program that performs processing based on the stress sampled according to the flowchart of FIG. 3, and is executed as a background job (BGJ).

In step 91, the determination is made when most of the 16×16 grid maps have a learned disease and the flag flag (setting of this flag will be explained in detail later based on the flowchart of FIG. 7) is set to 1. stress (degree of air-fuel ratio deviation) and a predetermined value (for example, 0.
8) to determine whether the degree of air-fuel ratio deviation when learning is almost completed is greater than or equal to a predetermined value.

Here, when the stress exceeds a predetermined value, it is determined that although learning has almost completed, an air-fuel ratio deviation has occurred due to the learning result, and learning is performed again (learning repetition). The process advances to step 92 in order to do so.

In step 92, a flag Fφ is set for determining whether or not air-fuel ratio learning for each operating region has been completed.

F (0,0)~F (3,3), FF(0,O)~
All FFs (16, 16) are reset to zero, and 1
A flag that is set to 1 when most of the operating regions of the 6×16 grid map have been learned; and the subdivision operating regions (16 regions) of the 16×16 grid map included in the range of one operating region of the 4×4 grid map. The flag fIag (0°0) to (3,3
) is reset to zero.

Furthermore, since learning is repeated again from all operating ranges as described above, stress is also reset to zero.

In this way, if the degree of deviation (stiffness) of the air-fuel ratio feedback correction coefficient LMD from the reference value becomes greater than a predetermined value, relearning can be performed to prevent accidents such as holes in the intake system. When the air-fuel ratio changes suddenly, the learning for each large operating range is re-performed, so the air-fuel ratio can be quickly converged.

Next, a description will be given of correction of the learning correction coefficient for a larger segmented operation area based on the subdivision area, which is performed according to the program shown in the flowchart of FIG.

The program shown in the flowchart of FIG. 7 is executed as a background job (BGJ). First, in step 101, when the learning correction coefficient KBLRCφ corresponding to all operating ranges has been learned, 1 is set. The flag Fφ that is set is determined, and the flag Fφ is set.
If the number is zero, the program is terminated as is, or if the number is 1, the program proceeds to step 102 and subsequent steps.

In step 102, various parameters used in this program, such as Sum, W, X, and Y, are initialized. Here, Sum is the learning correction coefficient KBLR stored in the area that has been learned in each driving area of the 4X4 grid map.
It is the integrated value of CI and 1. 1.0 is set as the initial value, and the W is used to count up the number of samples in the integration calculation, and l is set as the initial value. Further, X and Y designate grid positions in the 4X4 grid map, and zero is set to each as an initial value.

In the next step 103, a flag F (X
, Y) (4x4 lattice learned determination flag) and searches for a learned area in the 4x4 lattice driving area indicated by X and Y. Since the initial values of X and Y are zero, if the operating region of (0, O) has not been learned, in step 104, X is increased by 1 to become (1, O), and in step 105, it is determined that it is not X or 4. Once the determination is made, the determination in step 103 is performed again.

In this way, when X is incremented to 4 in step 104, X is reset to zero in step 106 and Y is incremented by 1. Returning to step 103 and proceeding to step 104, X is increased by 1, so as a result, fixing Y and changing X is repeated, and 4
Flag F (X
, Y).

Here, flag F (X
, Y) is determined to be 1, the process advances to step 108.

In step 108, one learning correction coefficient KBLRCI which is determined to be flag F (X, Y) or l and which is stored in the learned driving region is added to the accumulated value Sum up to that point.
At the same time, the sampling number W is increased by 1 in order to count the number of additions (in other words, the cumulative number of samples).

Then, in the next step 109, the (
In order to discriminate one by one the 16 regions in the 16X16 lattice included in the X, Y) region, initial values of zero are set for each of α and β, and learning correction is performed for the regions that have already been learned among the 16 regions. An initial value of 1.0 is set to Sump, which sets the integrated value of the coefficient KBLRC2, and an initial value of 1 is set to Z, which counts the number of samples of the integrated value Sump.

Then, in steps 110 to 115, the four
FF (α
+4X, β+4Y), and the learning correction coefficient KBLRC2 stored in the learned area is calculated as
ump is accumulated, and the accumulated sample number Z is counted up.

In this way, when a learned area among the 16 areas of the 16x16 grid map that is included in the learned area of the 4x4 grid map is picked up, the next step is to determine whether or not it exceeds 12). do.

Here, if the number of samples Z is less than a predetermined value, the (X
As for the Y) area, the learning of the 16 more subdivided areas has not progressed sufficiently, so in step 117 the flag F (X, Y) is set to zero and relearning is performed. .

After setting the flag (X, Y) to 7 in step 117, the process returns to step 104 and the 4×4
Search for other areas that have been learned on the grid map.

On the other hand, if it is determined in step 116 that the number of samples Z exceeds the predetermined value, the flag f is set in step 118.
lagF (X, Y) is set to 1, and in the next step 119, the deviation between the average value (S ump/Z ) of the learning correction coefficient KBLRC2 stored in the learned area and the target convergence value 'rarget is calculated. Absolute value or predetermined value (
For example, it is determined whether the value is less than 0.04).

If the deviation exceeds a predetermined value, it should be
The correction to be borne by the learning correction coefficient KBLRCI for a 4-grid map, or the learning correction coefficient KB for a 16x 16-lattice map.
Assuming that it is added to LRC2, step 12 is performed to transfer the added amount to the learning correction coefficient KBLRCI side.
Go to 0.

In step 120, the 4
The learning correction coefficient KBLRCI (X, Y) stored in the area (X, Y) on the X4 lattice map is learned and updated according to the following formula.

KBLRCI (X, Y)← ump KBLRCI (X, Y) 10 (--Target)
In other words, using the above calculation formula, the average correction amount borne by the 16x16 lattice map is transferred to the 4x4 lattice map corresponding to the area including them, and the 16x16
This is to bring the correction amount by the lattice map closer to the target convergence value.

In this way, when the average correction amount borne by the 16X16 grid map is transferred to the 4x4 grid map corresponding to the area including them, the learning correction coefficient KBLRC2 on the 16XI6 grid map is adjusted accordingly. Since reverse correction is necessary, the processing from step 121 onwards is performed.

That is, the learning correction coefficient KBL corrected in step 120
The amount added to the learning correction coefficient KBLRCI is subtracted from the learning correction coefficient KBLRC2 stored in each of the 16 regions on the 16×16 grid map included in the region of RCI (X, Y) (step 121).
~Step 126).

In this way, after the correction amount for the learning correction coefficient KBLRCI is corrected by inversely correcting the learning correction coefficient KBLRC2, the process returns to step 104, searches for a learned area on the 4×4 grid map, and When a region is detected, the above-described processing is repeated each time.

Then, after checking the learned areas for all areas on the 4x4 grid map, steps 107 to 12
We will move on to 7.

In step 127, W, which counts up the number of learned areas on the 4x4 grid map, is set to a predetermined value (for example, 12
), and it is determined whether most areas of the 4×4 grid map have been learned.

Here, if the W is less than or equal to the predetermined value, the flag is set to zero in step 128 and the process ends; however, if the W exceeds the predetermined value, the flag is set to zero in step 129. After setting 1 to
After step 130, the learning correction coefficient K corresponding to all operating regions is calculated based on the data of the learned region of the 4×4 grid map.
Processing to correct BLRCφ is performed.

In step 130, the learning correction coefficient KBLRCI in the area determined to have been learned on the 4×4 grid map is
The average value (S um/W ) and the target convergence value Targ
It is determined whether the absolute value of the deviation from et is less than a predetermined value. If the deviation is less than a predetermined value, the program is terminated, but if it is greater than or equal to the predetermined value, the learning correction coefficient K corresponding to all operating regions should be
The correction amount to be borne by the BLRCφ side is the learning correction coefficient KB.
It is assumed that this has been transferred to the LRCI side, and the process proceeds to step 131 in order to have the learning correction coefficient KBLRCφ side bear the burden of this transfer.

In step 131, the learning correction coefficient K is determined according to the following formula.
The learning update of BLRCφ is performed.

BLRCφ← In other words, the average correction amount in the learned area on the 4×4 grid map is transferred to the learning correction coefficient KBLRCφ that is corrected in all operating areas, and the learning correction coefficient KBLRCI becomes the target convergence value. It is intended to be learned in the vicinity.

Here, similarly to the case where the learning correction coefficient KBLRCI is corrected based on the learning correction coefficient KBLRC2 of the same area,
It is necessary to reversely correct the learning correction coefficient KBLRCφ by the amount corrected to each learning correction coefficient KBLRCI, and in steps 132 to 137, from the learning correction coefficient KBLRCI of all driving regions of the 4×4 grid map, The amount of correction added to the learning correction coefficient KBLRCφ (S un/Z
-Target) Correct by subtracting Xγ2.

<Effects of the Invention> As explained above, according to the present invention, learning is prohibited when the range of change in the average value of the air-fuel ratio feedback correction value is greater than or equal to a predetermined level. It is possible to prevent errors from being learned incorrectly, and what is incorrectly learned during acceleration (deceleration) is used during deceleration (acceleration) or during steady state, resulting in large air-fuel ratio differences. This has the effect of preventing this.

In addition, by setting a predetermined level for determining the range of change for each operating region based on predicted error rate data for air-fuel ratio control that is preset according to engine operating conditions, the range of change due to transient operation and the base It is possible to accurately discriminate between the range of change due to air-fuel ratio deviation, and it is possible to secure learning opportunities while preventing erroneous learning.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of the present invention, FIG. 2 is a system schematic diagram showing an embodiment of the present invention, FIGS. 3 to 9 are flowcharts showing control details in the above embodiment, and FIG. 11 and 11 are diagrams for explaining the state of the learning map in the above embodiment, respectively, and FIGS. 12 and 13 are time charts for explaining the problems of conventional learning control, respectively. 1. Engine 6... Fuel injection valve 12... Controller/L, unit 13... Air flow meter 91
4...Crank angle sensor 15...Water temperature sensor
166.・Oxygen sensor patent applicant Fujio Sasashima, agent and patent attorney for Japan Electronics Co., Ltd. BLRC in Figures 6, 10, and 12

Claims (2)

[Claims] (1) An engine operating condition detection means for detecting engine operating conditions including at least operating parameters related to the amount of air taken into the engine, and a basic fuel supply amount based on the engine operating conditions detected by the engine operating condition detection means. basic fuel supply amount setting means for setting the air-fuel ratio; air-fuel ratio detection means for detecting the air-fuel ratio of the air-fuel mixture taken into the engine; and comparing the air-fuel ratio detected by the air-fuel ratio detection means with a target air-fuel ratio to determine the actual air-fuel ratio. air-fuel ratio feedback correction value setting means for increasing or decreasing an air-fuel ratio feedback correction value for correcting the basic fuel supply amount so as to bring the fuel ratio closer to the target air-fuel ratio; an air-fuel ratio learning correction value storage means for rewritably storing an air-fuel ratio learning correction value for correcting a basic fuel supply amount; and learning a deviation of the air-fuel ratio feedback correction value from a target convergence value, and reducing the deviation. an air-fuel ratio learning correction value correcting means for correcting and rewriting the air-fuel ratio learning correction value stored in the air-fuel ratio learning correction value storage means in a direction in which the basic fuel supply amount, the air-fuel ratio feedback correction value, and the air-fuel ratio learning correction value are a fuel supply amount setting means for setting a final fuel supply amount based on an air-fuel ratio learning correction value for a corresponding operating region stored in a fuel ratio learning correction value storage means; a fuel supply control means for driving and controlling the fuel supply means based on the fuel supply amount; and when a range of change in the average value of the air-fuel ratio feedback correction value set by the air-fuel ratio feedback correction value setting means is equal to or greater than a predetermined level. An air-fuel ratio learning control device for an internal combustion engine, comprising: learning inhibiting means for prohibiting the air-fuel ratio learning correction value modification means from rewriting the air-fuel ratio learning correction value. (2) Based on the prediction error rate data of the air-fuel ratio control preset according to the engine operating conditions and the data of the operating conditions that demarcate the operating range in the air-fuel ratio learning correction value storage means, the learning inhibiting means Claim 1, further comprising learning prohibition level setting means for setting a predetermined level for each predetermined operating range of the air-fuel ratio learning correction value storage means.
The air-fuel ratio learning control device for the internal combustion engine described above.