JPH01106952A - Control device for learning of internal combustion engine - Google Patents

Abstract

PURPOSE:To improve the accuracy and speed of learning by analyzing the deviation of a feedback correction value from its reference value being classified by plural factors, renewing a learned value classified by factor, and correcting the feedback correction value based on the difference between a controlling quantity corrected by the learned value and a controlling quantity to which a feedback correction value is added. CONSTITUTION:The detected values of an airflow meter 13, a crank angle sensor 14, an O2 sensor 16, etc. are inputted into a control unit 12 to operatingly control the valve opening time of a fuel injection valve 6. The control unit 12 divides the deviation of an air-fuel ratio feedback correction value based on the O2 sensor 16 from its reference value into plural parameters classified by the factors of one caused by the airflow meter 13 and one caused by the fuel injection valve 6, etc. based on an engine speed and a basic fuel injecting quantity, to carry out the renewal of a learned value classified by factor. Based on the difference between a basic controlling quantity which is corrected by the learned value classified by factor and the controlling quantity to which the feedback correction value is added, the renewal of the feedback correction value is carried out in the direction of reducing this difference.

Description

[Detailed description of the invention] <Industrial application field> The present invention relates to an air-fuel ratio (fuel injection amount) for an internal combustion engine.

This invention relates to a learning control device for a feedback control system such as ignition timing and idle speed.

<Prior art> As a conventional learning control device for an internal combustion engine,
203828, JP-A-59-211738, and JP-A-60-90944.

There is one disclosed in Japanese Patent Application Laid-Open No. 61-190141.

These are feedback correction values that are set using proportional/integral control, etc., while comparing the basic control amount, which is set in accordance with the control target value such as the air-fuel ratio, based on the operating state of the engine, with the control target value and the actual value. In systems that perform feed-punk control of the air-fuel ratio, etc. to the control target value by correcting the control amount and calculating the control amount, the deviation from the reference value of the feedback correction value during feedback control is calculated as the engine The area-specific learning value is determined by learning for each area of the state, and when calculating the control amount, the basic control amount is corrected by the area-specific learning value, and the control amount obtained is calculated without correction using the feedback correction value. is made to match the control target value, and during feedback control, this is further corrected using a feedback correction value to calculate the control amount.

According to this, it is possible to eliminate the follow-up delay of feedback control during transient operation during feedback control, and it is possible to accurately obtain a desired control output when feedback control is stopped.

Therefore, it absorbs variations in component parts such as electronically controlled fuel injection devices, and also absorbs changes over time such as engine filling efficiency, atmospheric pressure, etc.
It is used to maintain the maximum performance of the engine over a long period of time by correcting changes in operating environment conditions such as temperature and humidity.

<Problems to be Solved by the Invention> However, such conventional learning control devices use a so-called iterative learning method using a data map, that is, data map grid divisions are set according to engine operating conditions, and feedback control deviation amount in each learning area is calculated. Since this method repeatedly updates the information based on learning experience, it has the disadvantage that if each learning area classification is set in detail in order to improve learning correction accuracy, the learning update speed becomes slow.

In other words, learning correction accuracy and learning speed are contradictory conditions.

SUMMARY OF THE INVENTION In view of these conventional problems, it is an object of the present invention to provide a learning control device for an internal combustion engine that can significantly improve learning speed while increasing learning correction accuracy.

<Means for Solving the Problems> In order to achieve the above object, the present invention configures a learning control device for an internal combustion engine including the following means A to J, as shown in FIG.

(A) Basic control amount setting means for setting the basic control amount corresponding to the control target value of the controlled object of the internal combustion engine. (B) Comparing the control target value and the actual value and moving the actual value closer to the control target value. Feedback correction value setting means for increasing or decreasing the feedback correction value by a predetermined amount (C) Rewritable factor-specific learned value storage means for storing a plurality of factor-specific learned values (D) Setting the basic control amount to the feedback correction value and (E) a control amount calculating means for calculating a control amount by correcting the plurality of factor-specific learned values and using arithmetic formulas set accordingly. Control means for controlling (F) Deviation detecting means for detecting the deviation of the feedback correction value from the reference value (G) Analyzing the causes of the deviation based on various information and detecting the deviation by factor based on the analysis result (H) Factor-specific learned value updating means that corrects and rewrites the factor-specific learning value in the factor-specific learning value storage means based on each of the plurality of parameters (I) The deviation detecting device feedback based on the difference between the control amount calculated by the control amount calculation means and the control amount calculated based on the plurality of factor-specific learning values based on the plurality of parameters separated this time by the factor analysis means. Feedback correction direction determining means (J) for determining a correction direction; a feedback correction value for correcting the feedback correction value set by the feedback correction value setting means in a direction that reduces the difference based on the determination result by the feedback correction direction determining means; Correction means (effect) The basic control amount setting means A sets a basic control amount corresponding to the control target value of the controlled object of the internal combustion engine, and the feedback correction value setting means B compares the control target value and the actual value. The feedback correction value is set to increase or decrease by a predetermined amount based on, for example, proportional/integral control in the direction of bringing the actual value closer to the control target value. Then, the control amount calculation means corrects the basic control amount with the feedback correction value, and further calculates the optimal value set according to the plurality of factor-specific learning values stored in the factor-specific learning value storage means C. The control amount is calculated by correcting it using a calculation formula.

Then, the control means E operates according to this control amount to control the controlled object of the internal combustion engine.

On the other hand, the deviation detection means F detects the deviation of the feedback correction value from the reference value. And factor analysis means G
The system inferentially analyzes the factors that led to the deviation based on various information (for example, at least one of the following: engine operating condition, deviation amount, deviation direction, deviation speed, deviation direction, etc.) according to predetermined analysis rules. , the deviation is separated into a plurality of parameters for each factor based on the analysis results, and the learning value for each factor updating means H corrects the learned value for each factor stored in the storage means C based on each of the plurality of parameters. rewrite.

The feedback correction direction determination means I determines the difference between the control amount set when the deviation is detected and the control amount calculated using the learned values for each factor based on the plurality of parameters analyzed by the factor analysis means G this time. Based on this, the direction of correction based on the feedback correction value is determined when using the learning values for each factor obtained through the current factor analysis. Then, the feedback correction value modification means J determines, based on the determination result of the feedback correction direction determination means (2), that the control amount calculated using the factor-specific learning value set this time is the control amount at the time of deviation detection (equivalent to the control target). The feedback correction value is corrected so that it approaches the control amount).

That is, when the factor analysis means G separates the deviation of the feedback correction value from the reference value into a plurality of parameters for each factor, the feedback correction direction determination means I calculates the control amount based on the learning value for each factor based on the analysis result. Then, the control value (control amount at the time of deviation detection) corrected to the control target by feedback correction is compared. Here, if the two control amounts are approximately equal, control will be achieved to the control target by using the learning value for each factor obtained through the current factor analysis and the current feedback correction value. 2
When there is a difference in the control amount, using the learned values for each factor obtained through factor analysis indicates that the control target is not controlling and needs to be followed by feedback correction, so the learned values for each factor are used. The feedback correction value is corrected in advance so that the control is immediately achieved approximately to the control target even when the control target is controlled.

In this way, the deviation (error amount) of feedback control is detected, and this is inferred and factor-analyzed using various information and a database, and the control result when using this factor analysis result is determined to be close to the control target. By increasing/decreasing the feedback correction value in advance so that the feedback correction value is corrected in advance and performing correction with high accuracy using an arithmetic formula suitable for each factor, both learning correction accuracy and learning speed can be achieved.

<Embodiment> An embodiment in which a learning control device according to the present invention is applied to an air-fuel ratio feedback control system of an internal combustion engine having an electronically controlled fuel injection device will be described below.

In FIG. 2, an engine 1 is connected to an air cleaner 2 through an intake duct 3. Air is taken in via the throttle valve 4 and the intake manifold 5. A fuel injection valve 6 as a control means is provided in a branch portion of the intake manifold 5 for each cylinder. The fuel injection valve 6 is an electromagnetic fuel injection valve that opens when the solenoid is energized and closes when the energization is stopped. Fuel is injected and supplied by a pump and adjusted to a predetermined pressure by a pressure regulator. Although this example is a multi-point injection system, it may also be a single-point injection system in which a single fuel injection valve is provided in common to all cylinders, such as upstream of a throttle valve.

An ignition plug 7 is provided in the combustion chamber of the engine 1, which ignites a spark to ignite and burn the air-fuel mixture.゛And,
From engine 1, exhaust manifold 8° exhaust duct 9. Exhaust gas is discharged via a three-way catalyst 10 and a muffler 11. The three-way catalyst 10 oxidizes Co and HC in the exhaust components,
It is also an exhaust purification device that reduces NOx and converts it into other harmless substances, and the reconversion efficiency is the best when the air-fuel mixture is combusted at the stoichiometric air-fuel ratio.

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 ink face, 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 described above, a hot wire type or flap type air flow meter 13 is provided in the intake duct 3, and outputs a voltage signal according to the intake air flow rate Q.

Further, a crank angle sensor 14 is provided, and in the case of a four-cylinder engine, outputs a reference signal for every 180 degrees of crank angle and a unit signal for every 1 degree or 2 degrees of crank angle. Here, the engine speed N can be calculated by measuring the cycle of the reference signal or the number of unit signals generated within a predetermined time.

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

Furthermore, the 0□ sensor 16 is located at the collecting part of the exhaust manifold 8.
is provided to detect the air-fuel ratio of the air-fuel mixture taken into the engine 1 via the 0° concentration in the exhaust gas. If the 0□ sensor 16 is equipped with a NOx reduction catalyst layer, as proposed in Japanese Patent Application No. 62-65844, more accurate detection will be possible.

Here, the CPU of the microcomputer built in the control unit 12 performs calculations according to programs (fuel injection amount calculation routine, air-fuel ratio feedback control routine, optimal learning routine) on the ROM shown as flowcharts in FIGS. 3 to 5. processing and control fuel injection.

The functions of the basic control amount setting means, the feedback correction value setting means, the control amount calculation means, the deviation detection means, the factor analysis means, the learning value updating means for each factor, the feedback correction direction determination means, and the feedback correction value correction means are as follows. This is achieved by the program. Further, a RAM is used as the factor-specific learning value storage means, and the stored contents are retained even after the engine key switch is turned off using a backup power source.

Next, the state of the arithmetic processing of the microcomputer in the control unit 12 will be explained with reference to the flowcharts of FIGS. 3 to 5.

FIG. 3 shows a fuel injection amount calculation routine, which is executed at predetermined time intervals.

In step 1 (denoted as Sl in the figure; the same applies hereinafter), the intake air flow rate Q is detected based on the signal from the air flow meter 13, and the engine speed N is calculated based on the signal from the crank angle sensor 14. .

Water temperature T detected based on the signal from the water temperature sensor 15
Enter w etc.

In step 2, the basic fuel injection amount Tp corresponding to the intake air amount per unit rotation is determined from the intake air flow rate Q and the engine rotation speed N.
Calculate -Q/N (K is a constant) on =. This step 2 corresponds to the basic control amount setting means.

In step 3, the water temperature correction coefficient KT□1 according to the water temperature Tw
Various correction coefficients C0EF=1+KT including air-fuel ratio correction coefficient KMR according to engine speed N and basic fuel injection amount Tp
1. Set □+... to l+.

In step 4, the latest air-fuel ratio feedback correction coefficient α (reference value 1) set by the air-fuel ratio feedback control routine shown in FIG. 4, which will be described later, is read.

In step 5, the voltage correction numerator s is calculated based on the battery voltage.
Set. This is to correct changes in the injection flow rate of the fuel injection valve 6 due to changes in battery voltage.

In step 6, the factor-specific learning values XI and X2 are read from a predetermined address of the RAM serving as the factor-specific learning value storage means. Note that when learning has not started, the initial value is
, =O, X2 =1.

In step 7, the fuel injection amount Ti is calculated according to the following equation. This step 7 corresponds to the control amount calculation means.

Ti=Xz・'rp −C0EF・αten(Ts+X,)
In step 8, the calculated Ti is set in the output register. This allows predetermined engine rotation synchronization (for example, 1
At the fuel injection timing (for each rotation), a drive pulse signal having a pulse width of Ti that has been set latest is applied to the fuel injection valve 6, and fuel injection is performed.

FIG. 4 shows an air-fuel ratio feedback control routine, which is executed in rotational synchronization or time synchronization, and thereby sets the air-fuel ratio feedback correction coefficient α. Therefore, this routine corresponds to feedback correction value setting means.

In step 11, it is determined whether predetermined air-fuel ratio feedback control conditions are satisfied. Here, the predetermined air-fuel ratio feedback control condition is that the engine speed N is equal to or less than a predetermined value, and the basic fuel injection amount Tp representing the load is equal to or less than a predetermined value. If such conditions are not met, this routine is terminated. In this case, the air-fuel ratio feedback correction coefficient α is the previous value (or reference value 1)
is clamped, and air-fuel ratio feedback control is stopped. This stops the air-fuel ratio feedback control in the high rotation or high load region, obtains a rich output air-fuel ratio using the air-fuel ratio correction coefficient KMH, and suppresses the rise in exhaust temperature.
This is to prevent seizure of the engine 1 and burnout of the three-way catalyst 10.

When the air-fuel ratio feedback control conditions are met, step 1
Proceed to step 2 and beyond.

In step 12, the output voltage of the 02 sensor 16 is ■. 2 is read, and in the next step 13, it is compared with the slice level voltage V rof corresponding to the stoichiometric air-fuel ratio to determine whether the actual air-fuel ratio is richer or leaner than the stoichiometric air-fuel ratio. That is, in this embodiment, the control target is the air-fuel ratio of the engine intake air-fuel mixture, and the control target is the stoichiometric air-fuel ratio.

When the air-fuel ratio is lean (■o2〈■r,, f), the process proceeds from step 13 to step 14, where it is determined whether or not it is the time of reversal from rich to lean (immediately after the reversal). After proceeding to step 16, the deviation from the reference value 1 of the previous air-fuel ratio feedback correction coefficient α is stored as a=α−1 for the optimum learning routine shown in FIG. 5, which will be described later. α is increased by a predetermined proportionality constant of 2 with respect to the previous value. Otherwise, the process proceeds to step 17, where the air-fuel ratio feedback correction coefficient α is increased by a predetermined integral constant of 1 with respect to the previous value, and thus the air-fuel ratio feedback correction coefficient α is increased at a constant slope. Note that P>>1.

When the air-fuel ratio is rich (Voz>V,,!f), proceed from step 13 to step 18 to determine whether it is the time of reversal from lean to rich (immediately after reversal), and when the ratio is reversed, proceed to step 19. For the optimum learning routine shown in FIG. 5, which will be described later, the deviation of the previous air-fuel ratio feedback correction coefficient α from the reference value 1 is stored as b=α−1, and then the process proceeds to step 20 where the air-fuel ratio feedback correction coefficient α is calculated. is decreased by a predetermined proportionality constant of 2 minutes from the previous value. Otherwise, the process proceeds to step 21, where the air-fuel ratio feedback correction coefficient α is decreased by a predetermined integral constant of 1 minute from the previous value, and thus the air-fuel ratio feedback correction coefficient α is decreased at a constant slope.

Figure 5 shows the optimal learning routine, which is executed at predetermined intervals,
As a result, factor-specific learning (+!X+, Xz is set and updated.

In step 31, it is determined whether a predetermined learning condition is satisfied. Here, the predetermined learning condition is that the air-fuel ratio is under feedback control and that the rich/lean signal of the 02 sensor 16 is inverted at an appropriate period. If such conditions are not met, this routine is terminated.

If the predetermined learning condition is satisfied, the process proceeds to step 32 and the output voltage V of the 0□ sensor 16 is set. 2 is reversed, and if it is not reversed, the process proceeds to step 33, where the engine speed N and basic fuel injection amount Tp are sampled as data on the engine operating state at that time.

When the output voltage of the 0□ sensor 16 is inverted, the process proceeds to step 34 to obtain the average value of a and b described above for optimal learning. At this time, a and b are the upper and lower peak values of the deviation of the air-fuel ratio feedback correction coefficient α from the reference value 1 from reversal to reversal of the increase/decrease direction of the air-fuel ratio feedback correction coefficient α, as shown in FIG. , by calculating these average values, the reference value 1 of the air-fuel ratio feedback correction coefficient α is determined.
The average deviation Δα from Δα is detected.

Therefore, step 15.19 in FIG. 4 and step 34 in FIG. 5 correspond to the deviation detection means.

Next, proceed to step 35 and output voltage of 02 sensor 16■
. Engine speed N and basic fuel injection amount while 2 is reversed
Movement of rp (N I, N 2..., Tp+, TPz
...) and specify the engine operating state (N, TP).

Next, proceed to step 36 and check the engine operating state (N, Tp).
The learning weight assigned to each area is searched for by referring to the map from the area, and the learning weight assigned to each area is the parameter. however,
2 for K and + is 1 or less.

Here, the factors that led to the deviation Δα are mainly caused by the fuel injection valve 6 (hereinafter referred to as F/I critical factor), and factors caused by the air flow meter 13 including changes in air density (hereinafter referred to as F/I critical factors). Q factor) and the proportion of each factor is expressed as 2, , and 2.

Then, based on a rule of thumb, we estimate that the F/■ factor is large in the low-speed, low-load area, and the Q factor is large in the high-speed, high-load area, so we assign + and KzO values to each area, and refer to this map. By doing so, factor analysis is performed based on the engine operating status.

As a result, the deviation Δα is set to 1 in the F/I prisoner factor parameter.
・Δα and 2 ・Δα can be separated into Q factor parameters, and in the next step 37, ΔαI
Separate into each parameter as =KI・Δα, Δα2=2・Δα.

Therefore, steps 35 to 37 correspond to factor analysis means.

In addition to being performed based on the engine operating state as described above, the factor analysis may also be performed based on the amount of deviation, direction of deviation, speed of deviation, direction of deviation change, etc., and by inference from these databases.

Next, the process proceeds to step 38, where the learned values X, , X, for each factor are stored in a predetermined address on the RAM, and the deviation Δα, M5
The difference Δ is added to the learning value X2 of the other Q factor.
Update α2 by adding M2. M+ and Mt are learning weighting coefficients.

X+ = X+ +M+ ・Δα.

X2 = x, +M2 ・Δα2 Next, the process proceeds to step 39, and when the deviation Δα is detected, the fuel is calculated based on the stored data of the current factor-based learned values XI and Xz calculated in the fuel injection amount calculation routine of FIG. 3. Read the injection amount Ti and set this value as MTi. The fuel injection amount Ti when the above-mentioned deviation Δα is detected is the fuel injection amount Ti that has been corrected and set so that the air-fuel ratio becomes near the stoichiometric air-fuel ratio by the air-fuel ratio feedback correction coefficient α. This is the average value of the fuel injection amount Ti when the upper and lower peak values of the correction coefficient α are taken.

In the next step 40, the fuel injection amount Ti is calculated using the factor-specific learning values X, , X2 updated in the step 38 above. However, the various numerical values used in this calculation are the learning values for each factor X+, Xz among the various numerical values used in the calculation of MTi.
This is the only thing that has been replaced.

T i =X2・Tp−COEF・α+(T s +
X Comparing the feedback-corrected fuel injection amount MTi, the factor-specific learning value X set this time
1. The direction of increase/decrease in the feedback correction coefficient α when using X2 is determined. Therefore, steps 39 to 41 correspond to feedback correction direction determining means.

If it is determined in step 41 that Ti<<MTi or Ti>>MTi, the current air-fuel ratio feedback correction coefficient α and the factor-specific learning values X, , X, If the fuel injection amount Ti is calculated using , the target stoichiometric air-fuel ratio cannot be obtained. Value X+
When fuel injection amount control is started using , . Therefore, in this embodiment, the factor-based learning value X1. The air-fuel ratio feedback correction coefficient α is increased/decreased in advance by the amount that X2 is corrected, and the factor-specific learning value X1. Even if X2 is used, the air-fuel ratio is immediately controlled near the stoichiometric air-fuel ratio.

That is, if it is determined in step 41 that Ti>>MTi, the current learning values for each factor are X, .

If , the amount of fuel will be larger than the fuel injection amount Ti corresponding to the stoichiometric air-fuel ratio, and the air-fuel ratio will become richer, so step 42
Set the air-fuel ratio feedback correction coefficient α to a predetermined value α. Then, the process returns to step 40 again, and the fuel injection amount Ti calculated using the reduced air-fuel ratio feedback correction coefficient α becomes MT.
The reduction correction in step 42 is repeated until the distance between i and the road is reached.

On the other hand, if it is determined in step 41 that Ti<<MTi, if the current factor-specific learned values X, , becomes lean, so in step 43 the air-fuel ratio feedback correction coefficient α is set to a predetermined value α. Then, the process returns to step 40 again until the fuel injection amount Ti calculated using the increased air-fuel ratio feedback correction coefficient α becomes between MTi and the road. Repeat the increase correction in step 43.

Therefore, the above steps 42 and 43 correspond to the feedback correction value correction means.

The air-fuel ratio feedback correction coefficient α corrected as described above is used in the air-fuel ratio feedback control routine shown in FIG. 4, and is set to increase or decrease based on the normal air-fuel ratio rich/lean.

Here, if the air-fuel ratio feedback correction coefficient α is corrected by the correction in step 42 or step 43 such that it is determined as T i ', MT i in step 41, or the deviation Δα is small and the difference between the previously learned value and the The learning values X+ and Xz for each factor such as time are set and step 4
1, if it is determined that TiζMTi, step 44
Proceed to step 3 and write the factor-specific learning values X1°X2 set in step 38 to the specified address on the RAM to rewrite the data, and this rewritten data is read in step 6 of the routine in Figure 3 to determine the fuel injection amount. It is used for calculating Tt.

The RAM is a backup memory, and the stored contents are retained even after the engine key switch is turned off.

Therefore, the portion of step 44 corresponds to a factor-specific learning value updating means.

In this way, the F/I prisoner requirement learning value X and the learning value X2 of the Q factor are determined, but correction based on these is based on the factor This is done using a separate optimal calculation formula.

That is, an arithmetic expression is set for the learned value X of the F/I factor as an addition term to the basic fuel injection amount Tp, and for the learned value X2 of the Q factor as a multiplication term to the basic fuel injection amount Tp. corrections will be made.

Figure 7 shows that the effect of this learning control is +16 for the old seal.
The engine with a rich tendency of % approaches the engine with the median variation of ∆ mark after about 4 learnings, and the engine with a lean tendency of 16% with ∆ mark after about 3 learnings.
This shows how the dispersion of the mark approaches the median value engine, clearly showing the improvement in learning speed by this learning control.

In this embodiment, the so-called L-Jetro type electronically controlled fuel injection system that has an air flow meter and detects the intake air flow rate is shown, but the so-called D-Jetro type that detects the intake manifold negative pressure is used. It can be applied to various systems such as the so-called α-N method based on the throttle valve opening (α) and the engine speed (N).

Moreover, it can be applied not only to air-fuel ratio feedback control, but also to ignition timing control based on knocking detection and idling speed feedback control via an auxiliary air valve.

<Effects of the Invention> As explained above, according to the present invention, instead of learning by area as in the past, by analyzing the factors that led to the deviation and using the learning values for each factor as a result of the analysis. Since the feedback correction value is increased or decreased in advance so that the control is approximately in line with the control target, the learning speed can be greatly improved, and even if the learning value for each factor is updated, the control is immediately adjusted to the control target. The accuracy of learning correction can be improved. In addition, such learning control can reduce the number of matching steps, simplify the management of two parts, and make maintenance free. Furthermore, the capacity of the backup memory can also be reduced.

[Brief Description of the Drawings] Fig. 1 is a functional block diagram showing the configuration of the present invention, Fig. 2 is a system diagram showing an embodiment of the present invention, Figs. 3 to 5 are flow charts showing control contents, FIG. 6 is a diagram showing how the air-fuel ratio feedback correction coefficient changes, and FIG. 7 is a diagram showing the effect of learning control. 1... Engine 6... Fuel injection valve 12... Control unit 13... Air flow meter
14...Crank angle sensor 16...0□Sensor patent applicant Japan Electronics Co., Ltd. Agent Patent attorney Fujio Sasashima Figure 1

Claims (1)

[Scope of Claims] Basic control amount setting means for setting a basic control amount corresponding to a control target value of a controlled object of an internal combustion engine, and comparing the control target value and the actual value to bring the actual value closer to the control target value. feedback correction value setting means for increasing or decreasing the feedback correction value by a predetermined amount in the direction; rewritable factor-specific learning value storage means for storing a plurality of factor-specific learning values; and setting the basic control amount to the feedback correction value. and a control amount calculation means that calculates a control amount by correcting the plurality of factor-specific learned values using calculation formulas set accordingly, and operates in accordance with the control amount to control a controlled object of the internal combustion engine. a control means for detecting a deviation of the feedback correction value from a reference value, a deviation detection means for detecting a deviation of the feedback correction value from a reference value, and analyzing the cause of the deviation based on various information and converting the deviation into a plurality of parameters for each factor based on the analysis result. factor analysis means for separating; factor-specific learning value updating means for correcting and rewriting the factor-specific learning value in the factor-specific learning value storage means based on each of the plurality of parameters; and when the deviation detection means detects a deviation. Determining a feedback correction direction based on the difference between the control amount calculated by the control amount calculation means and the control amount calculated by the factor analysis means based on a plurality of factor-specific learning values based on the plurality of parameters separated this time. feedback correction direction determining means for determining the feedback correction direction; and feedback correction value modifying means for modifying the feedback correction value set by the feedback correction value setting means in a direction that reduces the difference based on the determination result by the feedback correction direction determining means. A learning control device for an internal combustion engine, comprising: