KR0176984B1 - Air-fuel ratio controller for engine - Google Patents

Abstract

(Objective) By introducing a learning value having a very slow learning speed from the conventional learning value, the learning function is fully operated even when there is a variation in air-fuel ratio beyond the learning range of the conventional learning value.

(Configuration) The air-fuel ratio learning value calculating means 35 reads out the learning value of the learning area and the learning value B of the memory 34 corresponding to the driving conditions at that time from the learning map 33, respectively, and learns these learning. The sum of the values A and B is calculated as the air-fuel ratio learning value. The injection amount calculation means 36 corrects the basic injection amount using this calculated air-fuel ratio learning value.

On the other hand, based on the air-fuel ratio feedback correction amount α, the learning value updating means 38 updates the learning value A with a narrow learning range and at a relatively fast learning speed. Based on the air-fuel ratio learning value, the learning value B is updated at a learning speed much slower than the learning value A.

Description

Air-fuel ratio control device of the engine

The present invention relates to learning for the control of an air-fuel ratio of a learning control device, in particular an engine.

In the so-called three-way catalyst system, in order to improve the conversion efficiency of the three components of the exhaust gas (CO, HC, NO _X ), the air-fuel ratio of the exhaust gas passing through the catalyst is within a narrow range arbitrarily selected based on the theoretical air-fuel ratio. Feedback control of the air-fuel ratio is performed so as to be maintained.

Even if the basic air-fuel ratio (air-fuel ratio determined by the basic injection pulse width (Tp) according to the output of the air flow meter and the engine speed) is set equal to the theoretical air-fuel ratio, the flow rate characteristics of the air flow meter or the fuel injection valve deviate from the prescribed values, and the like. For some reason, the air-fuel ratio may be wrong. In this case, the O2 sensor output changes as follows. → the computer makes small changes (increase or decrease) in fuel injection to reduce air-fuel ratio variations; → The output of the O₂ sensor gradually returns to its normal level; → The performance cost gradually returns to the theoretical performance cost. This process is repeated until the air-fuel ratio reaches a theoretical value or near the theoretical value.

However, even if the air-fuel ratio is caused to go wrong and feedback correction works, it takes some time until the air-fuel ratio returns to the air-fuel ratio near the theoretical air-fuel ratio, and during this time, an abnormal state (uneven state) continues. After that, the engine remains in a normal state until the engine is stopped. However, upon restarting, the O2 sensor detects an abnormality and the computer adjusts the injection volume. The feedback cycle of the above correction is repeated. In other words, an abnormal condition occurs for a while every time the engine is restarted. In addition, the air-fuel ratio in an abnormal state appears when the cooling water temperature is low at start-up or under operating conditions in which the air-fuel ratio feedback correction is stopped in a state such as a high load operation or the like.

Therefore, in order to improve the responsiveness of air fuel ratio correction, the learning function was employ | adopted (refer to Unexamined-Japanese-Patent No. 60-145443, July 72, 74th issue of Automotive Engineering.

In this learning function, when the correction amount (that is, the learning value) necessary for the learning control is obtained by observing the feedback correction, the learning value is stored even if the backup power of the computer is stopped even if the engine is stopped. According to the learning value, an appropriate increase (decrease) correction is performed from the beginning. Therefore, no transients such as feedback correction are produced. That is, the abnormal state is not gradually corrected gradually, but is suddenly corrected (abnormal state does not appear).

In addition, since the correction by the learning value continues to operate even under the operating conditions in which the feedback correction is stopped, an appropriate air-fuel ratio is compensated, so that no abnormal state is generated even at this time.

In addition, since it is not possible to correctly update the current learning value in comparison with the preceding data unless it is the same driving condition, the learning conditions are strictly determined as follows. First of all, the condition under which the feedback correction works. This is because under these conditions, even if the driving conditions are slightly different, all of them try to be the theoretical air-fuel ratio. As shown in FIG. 9 and FIG. 9, the learning area for the operating condition is set, and the output of the O 2 sensor is sampled several times in the same area.

By the way, since the above learning function aims at taming component unevenness in a new vehicle or eliminating the variation in the basic air-fuel ratio, the learning rate is relatively fast in a narrow learning range (for example, ± 10%). .

However, when the air fuel ratio deviates from this learning range occurs, it cannot be corrected by the learning value. If the air-fuel ratio is largely shifted to the rich side, for example, due to an unexpected failure of the fuel system components, the learning value decreases relatively quickly in order to return the air-fuel ratio to the lean side. It is attached to, and learning does not proceed from then on. Without this lower limit, the learning value will continue to decrease until equilibrium is reached. If the variation in the air-fuel ratio is excessive, the learning function will not work sufficiently, resulting in poor exhaust performance.

Therefore, in the present invention, in addition to a conventional learning value having a narrow learning range and a relatively high learning speed, the present invention introduces a learning value that is much slower than the learning value, so that a variation in air-fuel ratio beyond the learning range of the conventional learning value occurs. In addition, it aims to make learning function fully enough.

1st invention, as shown in FIG. 1,

An O₂ sensor 31 which outputs an output corresponding to the air-fuel ratio of the exhaust;

Air-fuel ratio feedback correction means (32) for performing air-fuel ratio feedback correction to maintain the air-fuel ratio close to the theoretical air-fuel ratio based on the O 2 sensor output;

A learning map 33 for storing and storing a learning value A for each learning region,

A memory 34 for storing and storing a learning value B different from the learning value A for each learning region;

From the learning map 33, the learning value A stored and stored in the learning area corresponding to the operation condition at that time and the learning value B stored and stored in the memory 34 are read, respectively. Air-fuel ratio learning value calculation means 35 for calculating the sum of two learning values A and B as air-fuel ratio learning value KBLRC;

Fuel injection amount calculation means 36 for calculating a fuel injection amount by correcting a basic injection amount Tp corresponding to an operating condition in accordance with the calculated air-fuel ratio learning value KBLRC and the air-fuel ratio feedback correction amount α used for the air-fuel ratio feedback correction. )and,

A fuel supply device 37 for supplying fuel of the injection amount to an intake pipe;

A learning value updating means 38 for updating the learning value A stored and stored in the learning map 33 at a narrow learning range and a relatively fast learning speed based on the air-fuel ratio feedback correction amount α;

Based on the air-fuel ratio learning value, the learning value B stored and stored in the memory 34 is updated at a learning rate very slower than the learning value A stored and stored in the learning map 33. Learning value updating means (39).

According to a second aspect of the invention, the learning value A and the learning value B different for each of the regions are one for all the operating conditions.

Both of the learning values A and B are in the initial values, and in this state, a deviation in the basic air-fuel ratio (determined from the basic injection amount) may occur such that the fuel gauge component breaks out of the learning range of the learning value A. At this time, for example, if the deviation is the rich side, the air-fuel ratio feedback correction amount α is shifted to the smaller side in order to return the air-fuel ratio to the lean side, so that the learning value A is also shifted relatively rapidly from the initial value to the smaller side. Learning does not progress.

On the other hand, in the present invention, another learning value B acts. Because the learning value B is a very low learning speed, it does not act before the learning value A reaches the lower limit. However, it acts after the learning value A reaches the lower limit, and shifts to the smaller value from the initial value at a lower speed. In other words, the learning progresses with respect to the deviation of the air-fuel ratio outside the learning range of the learning value A.

By introducing a separate learning value B having a different personality from the learning value A in this manner, even when there is a deviation of the air-fuel ratio exceeding the limit of the learning value A, the learning can be performed so that the exhaust performance is not deteriorated.

In the second aspect of the invention, if the learning value B is one for all the operating conditions, the frequency of learning is increased.

1 is a corresponding drawing of the first invention of the present invention.

2 is a system diagram showing a first embodiment of the present invention.

3 is a flowchart for explaining the calculation of the air-fuel ratio feedback correction coefficient α.

4 is a flowchart for explaining the update of the learning condition and the learning value in the learning value A. FIG.

5 is a flowchart for explaining the update of the learning condition and the learning value in the learning value B. FIG.

6 is a flowchart for explaining the calculation of the fuel injection pulse width Ti.

7 is a characteristic diagram showing a map value of a step component PR.

8 is a characteristic diagram showing a map value of a step component PL.

9 is an area diagram showing a learning area.

10 is a waveform diagram for explaining the operation of the embodiment of the present invention.

11 is a waveform diagram for explaining the operation of the embodiment of the present invention.

* Explanation of symbols for main parts of the drawings

3: intake pipe 4: fuel injection valve (fuel supply device)

6: three-way catalyst 7: air flow meter

10: crank angle sensor 12, 31: O₂ sensor

21 control device 32 air-fuel ratio feedback correction means

33: Learning Map 34: Memory

35: air-fuel ratio learning value calculation means 36: fuel injection amount calculation means

37: fuel supply device 38, 39: learning value updating means

EXAMPLE

In FIG. 2, 7 is an air flow meter for detecting the air flow rate Qa sucked from an air cleaner, 9 is an idle switch, and 10 is a reference for a signal and a crank angle per unit crank. A crank angle sensor that outputs a signal for each position (Ref signal), 11 is a water temperature sensor, 12 is an O₂ sensor whose value changes rapidly in response to the theoretical air-fuel ratio in response to the oxygen concentration of exhaust gas, 13 Is a knock sensor, 14 is a vehicle speed sensor, and the signals of these sensors are input to a control device 21 made of a microcomputer.

Since the injection of fuel is supplied from the fuel injection valve 4 provided at one point in the intake port when the quantity is large or small, the injection amount is adjusted to the injection time by the control device 21. If the injection time is long, the injection amount is large, and if the injection time is short, the injection amount is small. The concentration of the mixer, that is, the air-fuel ratio, shifts toward the rich side when the fuel injection amount to a predetermined amount of intake air increases, and shifts toward the lean side when the fuel injection amount decreases.

Therefore, if the basic injection amount of the fuel is determined so that the ratio with the intake air amount is constant, a mixer having the same air-fuel ratio is obtained even if the operating conditions are different. When the injection of fuel is performed once per revolution of the engine, the basic injection pulse width Tp (= KQa / Ne, where K is an integer) for the amount of air sucked in one revolution is the current intake air amount ( Qa) and the engine speed Ne. Usually, the air-fuel ratio (basic air-fuel ratio) determined by this Tp is near the theoretical air-fuel ratio in the air-fuel ratio feedback correction region.

The exhaust pipe 5 is provided with a three-way catalyst 6 for treating three toxic components such as CO, HC, and NO _X discharged from the engine 1. The three-way catalyst 6 maintains the conversion efficiency of all three components satisfactorily only when the atmosphere of the catalyst is in a narrow range (catalyst window) centered on the theoretical air-fuel ratio. If the air-fuel ratio is shifted to the rich side even a little more than this range, the conversion efficiency of CO and HC is lowered. On the contrary, if it is shifted to the lean side, the conversion efficiency of NO _X is lowered.

The control apparatus 21 feedback-corrects the fuel injection amount based on the output signal from the O2 sensor 12 so that the air-fuel ratio average value is maintained near the theoretical air-fuel ratio at which the three-way catalyst 6 can fully exhibit its capabilities.

If the output of the O 2 sensor 12 is higher than the slice level corresponding to the theoretical air-fuel ratio, the air-fuel ratio is on the rich side and low on the lean side. When the air-fuel ratio is reversed to the rich side from this determination result, the air-fuel ratio must be returned to the lean side. Therefore, as shown in the flowchart of FIG. 3, immediately after the air-fuel ratio is inverted to the rich side, the step component (PR) is subtracted from the air-fuel ratio feedback correction coefficient α and until the air-fuel ratio is immediately reversed to the lean side. Integral component (IR) is subtracted from (steps 2, 3, 7 of FIG. 3, steps 2, 3, 9).

On the contrary, when the air-fuel ratio is inverted to the lean side, the step component PL is added to α immediately after the inversion, and the integral component IL is added until immediately before the actual air-fuel ratio is inverted to the rich side next (step 3 in FIG. , 4, 12), steps (2, 4, 14)).

Further, the operation of α is synchronized with the Ref signal. This is because the fuel injection is synchronized with the Ref signal, and the system also operates in synchronization with the Ref signal.

The value of the above step components PR and PL is relatively much larger than that of the integral components IR and IL. This is to provide a large value step component immediately after the air-fuel ratio is reversed to the rich side or the lean side, so as to quickly change to the opposite side quickly. After the step change, the air-fuel ratio is gradually changed to the opposite side with a small integral component, thereby stabilizing the control.

The step components PR and PL are maps having the basic injection pulse width Tp and the engine speed Ne as parameters (FIG. 7 is a map of the step component PR, and FIG. 8 is a map of the step component PL). To look up. In FIG. 7 and FIG. 8, the map values of PL and PR are different in some driving zones. In this driving zone, the output response of the O2 sensor is changed between the inversion to the rich side and the inversion to the lean side. Even if it is different, it is for the air-fuel ratio average value to be maintained near the theoretical air-fuel ratio.

In addition, the integral components IR and IL are determined in proportion to the fuel injection pulse width (corresponding to the engine load) Ti, which will be described later (steps 8 and 13 in FIG. 3). This is to make the amplitude of α substantially constant regardless of the control cycle of α because the amplitude of α may increase in the operating region in which the control period of α becomes long and may exceed the catalyst window. The values of the integral components IR and IL may be the same values.

In this way, if the air-fuel ratio of the exhaust gas is on the lean side than the theoretical air-fuel ratio, the fuel injection amount from the fuel injection valve 4 is increased so as to become the theoretical air-fuel ratio. Repeat to lose.

On the other hand, the learning area for air-fuel ratio learning is divided into areas (learning areas) divided by the basic injection pulse width Tp and the engine speed Ne, respectively, as shown in FIG. 9, and the learning values (KBLRC) for each of these areas. [%]) Is assigned. Specifically, each learning value is stored and stored in a two-dimensional map (learning map) using Tp and Ne as parameters. Needless to say, this map value backs up the battery so that it is not lost when the ignition switch is turned OFF.

The learning condition is performed when all of the following conditions are met, as shown in FIG.

1 Tp and Ne shall be in the same area (step 15 in FIG. 4),

2 air-fuel ratio feedback correction being performed (step 16 of FIG. 4),

3 The difference between the maximum and minimum of O₂ sensor output should be more than a certain value (step 4 of Fig. 4),

4 O2 sensor output will be sampled several times (step 18 in FIG. 4).

If the learning condition is established, the deviation amount ε [%] from the control center (100%) of α is obtained from Equation 1 below (step 19 in FIG. 4):

ε = (α _MAX + α _MIN ) / _2-100

Here, α _MAX is a value of α just before adding the step component PR, and α _MIN is a value of α just before adding the step component PL. Using this amount of deviation ε, the learning value KBLRC [%] is updated according to the following equation (2).

KBLRC = KBLRC + R #

Where R # is the learning update rate (value less than 1). The learning value of the area belonging to Tp and Ne when the learning condition is established is read from the map of FIG. 9, and the value (KBLRC on the right side of Equation 2) is modified to ε and the new value (Equation 2). KBLRC of the left side of is determined, and the new value is newly stored in the same area (steps 20 and 21 of FIG. 4).

However, the learning value KBLRC is limited between the lower limit value RLRMIN # and the upper limit value RLRMAX # (step 22 in FIG. 4).

In this case, the learning mainly updates the learning value KBLRC with a narrow learning range and at a high learning speed, mainly aiming at correcting the deviation of the parts of the new car and the deviation of the basic air-fuel ratio. For example, the learning range is set to ± 10% by setting the lower limit (RLRMIN #) and the upper limit (RLRMAX #) of the learning value to RLRMIN # = 90% and RLRMAX # = 110%, respectively. By setting the learning update rate R # to a relatively large value, the learning speed is set faster. It is manageable to keep the deviation of the parts non-uniformity and the basic air-fuel ratio in a predetermined range at the time of production in the case of a new car, and can cope with regular inspection when it goes out of range by deterioration by aging. Because.

However, when a large air-fuel ratio variation that is outside the learning range occurs, the above-described learning value cannot be corrected. Therefore, the above-described learning value (KBLRC) is different from the learning value in order to cope with this problem. Introduce a separate learning value.

If the learning value A is distinguished from the learning value B to be newly introduced, in this example, the air-fuel ratio learning value KBLRC can be obtained by Equation 3 below:

KBLRC = KBLRA + KBLRCB

On the other hand, the fuel injection pulse width Ti is given by Equation 4 below:

Ti = TpCo (α + A + B-200) + Ts

Where Tp is the default injection pulse width,

CO is the sum of 1 and various correction factors,

α is the air-fuel ratio feedback correction factor,

Ts is given by the invalid pulse width (Figure 6).

Here, the initial value of α and the learning value A is 100% as in the prior art, and the initial value of the learning value B is also 100%.

(1) learning area

In the learning value B, unlike the learning value A, no learning area is provided. In other words, in order to increase the frequency of learning, only one learning value is used for the entire driving area. This learning value B also backs up a battery so that it may not lose | disappear by turning off an ignition key switch.

(2) Update of learning level

The learning value B is also updated in the same manner as the learning value A in synchronization with the engine rotation (for example, every 16 rotations of the engine).

1 α + A-100 is compared with the air-fuel ratio lean-side limit value KBLRGH # at the start of learning, and when α + A-100? KBLRGH #, the learning value B is updated according to the following equation (5). 5 degree steps (32, 33)):

B = B + KBLB # × (KBLRC / 100)

Here, KBLB # is a learning update rate.

α + A-100 is an air-fuel ratio error (deviation from the theoretical air-fuel ratio) that remains uncorrected despite the correction operation based on α and the learning value A. When this value is greater than or equal to the lean side limit value KBLRGH # (for example, about 105 to 115%) at the start of learning (that is, greatly shifted to the lean side), the air-fuel ratio is returned to the rich side by updating the learning value B to the larger side.

On the other hand, if α + A-100 KBLRGH # and B ≥ 100, the air-fuel ratio error left uncorrected by α and the learning value A is less than the lean-side limit value of KBLRGH #. The learning value B is updated to the smaller one (steps 32, 36, 37 in FIG. 5).

B = B-KBLB # × (KBLRC / 100)

As a result of the update, if it is B100, it is limited to 100 (steps 38 and 39 in FIG. 5).

2 Similarly, α + A-100 is compared with the air-fuel ratio rich side limit value KBLGL # (for example, about 95 to 85%) at the start of learning, and if α + A-100 ≤ KBLGL #, α and the learning value A It is determined that the air-fuel ratio error left uncorrected by the value is equal to or larger than the rich limit value of KBLGL #, and the learning value B is updated to the smaller value by the above equation (6) (steps 40 and 41 in FIG. 5).

If α + A-100 KBLGL # and B ≤ 100, update the learning value B to the larger side by Equation 5 above (steps 40, 44, 45 in FIG. 5), and 100 when B ≥ 100 (Steps 46 and 47 of FIG. 5).

In order to prevent disturbances such as purge gas or blow-by gas from activated carbon canisters, the learning update rate (KBLB #) is as high as possible. It is set to a value small enough to slow down.

However, similarly to the learning value A, the learning value B is limited between the lower limit and the upper limit (steps 34 and 35 and steps 42 and 43 of FIG. 5).

(3) learning conditions

When all of the following conditions are satisfied, it is determined that the learning condition is satisfied (step 31 of FIG. 5). The following condition is similar to the learning condition for the learning value A.

1 Cooling water temperature (TW) is more than the lower limit, less than the upper limit,

When the TW is above the upper limit (at high temperature), there is a possibility of being affected by the purge gas, and when the TW is below the lower limit (at the low temperature), the basic air-fuel ratio is not stable due to the influence of fuel flowing through the wall. Do not learn.

2 Basic injection pulse width (Tp) is greater than the lower limit

Avoiding the influence of blow-by gas in the region where Tp is below the lower limit (low air flow region) (the blow-by gas is operated without reflux on the high load side, and then sucked in the low-air flow region such as children to enrich the air-fuel ratio). For example, learning is prohibited in the low air flow rate region.

3 The engine speed (Ne) should be more than the lower limit

4 When the water temperature (TWINT) is higher than the lower limit at start

5 Air-fuel ratio feedback correction

6 Not in clamp of air-fuel ratio feedback correction

7 When the idle switch is off,

Among the children, learning is stopped because of the large influence of the output unevenness of the blow-by gas or the air flow meter.

8 purge from canister should not be performed

Although the learning speed of the learning value B is slow, there is a possibility of mis-learning if it is influenced by the purge gas in front, so the learning is stopped during the purge.

Here, the operation of this example will be described.

The deviation of the air-fuel ratio such that the learning values (A, B) are both at the initial values (100% of all the learning values (A, B)) out of the learning range of the learning values (A) (for example, in the entire driving range). Think about when you have 15%).

At this time, since α shifts to the side smaller than 100% to return the air-fuel ratio to the lean side, the learning value A also shifts from the value of 100% to the smaller side at a relatively high speed, but sticks to 90% of the lower limit. Learning does not proceed.

On the other hand, in this example, another learning value B acts. Because the learning value B is very slow in learning speed, as shown in FIG. 10, the learning value B does not act before the learning value A reaches 90% of the lower limit. However, it acts after the learning value A reaches the lower limit, and shifts slowly from 100% of the initial value to the smaller one. For the deviation of the air-fuel ratio outside the learning range of the learning value A, learning is conducted by the learning value B, whereby the air-fuel ratio can be quickly returned to the theoretical air-fuel ratio.

In addition, the reference air-fuel ratio may shift 15% to the rich side temporarily due to introduction of blow-by gas into the intake pipe. Also at this time, since learning progresses by the learning value B after learning value A reaches 90% of a lower limit like FIG. 11, the air fuel ratio has returned to theoretical air fuel ratio quickly. In addition, condition 2 of the above learning conditions is to cope with a case where the influence of blow-by gas is very large (when the deviation of the standard air-fuel ratio is 20% or more shifted toward the rich side), and it is relatively blow-by at 15% as shown in FIG. When the influence of the gas is small, the learning value B is updated.

In this way, by introducing another learning value B having a different personality from the learning value A, it is possible to proceed with learning so that the exhaust performance does not deteriorate even when the air-fuel ratio exceeds the limit of the learning value A.

On the other hand, since the learning value B is only one learning value for the whole driving area, the learning frequency is high. The reason why the learning area is provided in the learning value A is because the demand for the learning value varies depending on the driving conditions. On the other hand, since the learning value B aims to advance learning by the learning value B in the case of the deviation of the air fuel ratio exceeding the threshold value of the learning value A, it makes it easy to advance learning by increasing the frequency of learning. will be.

In addition to the conventional learning values that are updated at a narrow learning range and relatively fast learning speed, the first invention introduces a separate learning value that is updated at a learning speed very slower than this learning value. Even in the variation of the air-fuel ratio exceeding the limit value of the conventional learning value, it is possible to proceed so that exhaust performance does not deteriorate.

In the second aspect of the invention, if the additional learning value updated at the very slow learning speed is likely to be one for all the driving conditions, the learning frequency is increased and it is easy to proceed.

Claims (3)

An engine air-fuel ratio control system, comprising: an oxygen sensor 31 for generating an oxygen sensor signal indicating an air-fuel ratio of exhaust gas of an engine; Feedback correction means (32) for determining a feedback correction amount in accordance with said oxygen sensor signal and for performing a feedback correction operation for maintaining the air-fuel ratio near the theoretical air-fuel ratio; A first memory section (learning map) 33 for storing a value of a first learning variable; A second memory section 34 for storing a value of a second learning variable different from the first learning value; Fuel injection amount determining means (36) for determining a fuel injection amount by correcting a basic injection amount corresponding to an engine operating condition in accordance with the feedback correction amount and the first and second learning values; A fuel supply device 37 for supplying fuel of the injection amount to an intake pipe of an engine; First updating means (38) for updating the first learning value stored in the learning map at a high learning rate in accordance with the feedback correction amount; And second updating means 39 for updating the second learning value stored in the second memory unit at a learning rate slower than the fast learning rate, in accordance with the learning value. The memory portion of the means includes means for storing a value of the first learning value for each of the plurality of first learning regions, which are different portions of the engine operating region determined by the engine load and engine speed of the engine, wherein the second The memory portion is part of an engine operating region and includes a portion of a first region of the first learning region and at least a portion of a second region of the first learning region of the second learning value for the second learning region. Means for storing a value, the control system comprising a value of a first learning value of the first learning area corresponding to a current engine operating condition determined by engine load and engine speed; Reading from the first memory section and reading the value of the second learning value from the second memory section if the current engine operating condition becomes the second learning region, and reading the first and second learning values Calculating means 35 for calculating a learning amount that is a sum, wherein the fuel injection amount determining means includes means for determining the basic injection amount according to the learning amount and the feedback correction amount determined by the calculating means. Air-fuel ratio control device of the engine. 2. An air-fuel ratio control apparatus for an engine according to claim 1, wherein a learning value (B) different from the learning value for each learning region is one for all driving conditions. The air-fuel ratio of the engine according to claim 1 or 2, wherein the learning range by the second updating means 39 is set wider than the learning range by the first updating means 38. controller.