JPH05282007A - Method and device for digital adaptive control - Google Patents

Abstract

PURPOSE: To permanently secure constitution learning by resetting all learning counters, whose counter values are at least equal to or smaller than a prescribed value, to initial values corresponding to a maximum of a lower limit value in the case of the occurrence of a condition which periodically occurs. CONSTITUTION: A controlled system is an internal combustion engine 10 having an injection device 11 and random value measuring instruments 12 and 1, and the operation change, for example, a rotational frequency (n) and an angle of a throttle valve of the internal combustion engine 10 are measured by measuring instruments 12 and 2. Two address signals ADRI and ADR2 are outputted in accordance with the numerical area of this operation change. An operation value map 13 for open loop control outputs a preliminarily set operation value VS for control related to the operation state in accordance with these address signals. An adaptive device 15 is provided with a learnt value memory 17 accessed by address signals, a forcible learning device 19 which reads out a learnt value in accordance with address signals and uses a deviation signal to change it and stores it as a new learnt value, and a function which resets learning counters.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital adaptive control device (system) and method, and more particularly to a digital adaptive control device and method for an internal combustion engine. Such devices and associated methods have been used in practice for several years, for example to control the amount of air to be supplied to an internal combustion engine in order to minimize the production of harmful gases.

[0002]

2. Description of the Related Art In principle, each adaptive control device has the following functional groups:
That is, a measuring device that detects an operating variable of a controlled object including a closed loop control actual value, a closed loop control device that outputs a closed loop control operation value, an open loop control device that outputs a preset control operation value, a closed loop control Device for forming at least one adaptive value by using the operating value for control, and a first correcting the preset operating value for control by using the adaptive value
And a second correction device that corrects a preset control operating value that has already been corrected by the adaptive value by using the closed-loop controlling operating value.

In practice, the adaptive devices described above can each be formed in various types of methods, of which all of the methods briefly described below can be used in the present invention.

Normally, at least one adaptation value is calculated which is valid for all operating ranges. However, a large number of adaptive values are often obtained, in which case detection is carried out in different operating regions in which the respective errors appear particularly clearly, and the adaptive values thus detected are then used in the entire numerical region of the operating variables. Used for correction in. In that case, some of the adaptation values are combined, for example, in a multiplying manner with a preset control operating value, while the other part is additively combined.

The adaptive value used in the entire operating region of the controlled object is called a rough adaptive value. In addition, so-called structural adaptive values are known. It is used for devices with adaptive value memory or learning value memory, which memory values do not apply to a large number of memory locations for the entire operating range of the controlled object, but to a learning value which applies only to a limited, predetermined area. Are stored respectively. If the controlled object is an internal combustion engine, this memory is addressable, for example, via the engine speed and the value of the throttle valve angle.
In that case, a two-dimensional memory unit can be used.
However, it is also possible to use, for example, a one-dimensional memory unit having eight storage locations, in which case each storage location corresponds to a predetermined region of the operating variable, for example the intake pressure. Alternatively, the memory unit can be a memory unit higher than two-dimensional. In addition to the structural adaptation values read from the learning value memory, at least one general adaptation coefficient can be used.

In connection with the control device of the present invention, for example, N.
Tomizawa is SAE Paper No. 860594, 1986
Pp. 177-185, "Development of high-speed and high-accuracy learning control system for engine control (Development
ment of a High-Speed High
-Precision Learning Controller
ol System for the Engine
Of particular interest are digital adaptive control systems, such as those described under the name "Control". This system has the basic functional groups mentioned at the beginning of the digital adaptive controller, and in particular the following functional groups: at least one learning value calculated when the first condition occurs, respectively. There is a learning value memory to be stored in an addressable memory location through a numerical area of the driving variable, and a learning counter for each memory location of the learning value memory, and each time a new learning value is written to the relevant memory location. Even if the values of the learning counter unit where the learning counter is incremented and the related learning counter when the second condition occurs, the maximum limit value is the lower limit value, whereby the numerical value of the driving variable related to this memory location is obtained. A forced learning device that replaces the learning value of a memory location with the value obtained by using the learning value of another memory location when it is shown that learning has not been performed for a region. To have.

The corresponding known method stores the learned value in a memory location which is addressable via the following steps: at least one numerical area of the operating variable,
The learning counter for each memory location is incremented when a new learning value is written to the relevant memory location, and even if the value of the relevant learning counter reaches when the second condition occurs, the maximum limit value is the lower limit value. , When it is shown that learning has not been done for the numerical region of the driving variable related to this memory location, the learning value of the memory location is forced to the value obtained using the learning values of other memory locations. To replace each step with each step.

The first time learning occurs when it occurs
The condition is that the driving variable area related to the memory location is exited. The second condition for forced learning is that a new operating cycle of the internal combustion engine has started. In such a new operation cycle, whenever it is revealed that the learning counter has not reached the lower limit value, forced learning is performed according to a predetermined algorithm.

In the above-mentioned system, when the numerical area associated with the memory location is consecutively removed three times, the actual closed-loop control operation value differs from the learning value stored in the memory location by a predetermined threshold value. If it is revealed that there is more deviation, the corresponding learning counter is set to zero. Thereby, the stored value is amplified and changed by the new value. This is because in the above system, the degree of learning is done according to the counter value.

The above-mentioned device and the associated method have the disadvantage that the associated learning counter is not forcedly learned for memory locations whose lower limit value has been exceeded over time.

[0011]

SUMMARY OF THE INVENTION It is therefore an object of the present invention to provide a digital adaptive control device (system) and method in which permanent learning is guaranteed for memory locations associated with operating areas where the controlled object is rarely operated. Is to provide.

[0012]

The digital adaptive control apparatus according to the present invention has the above-mentioned system characteristics and at least the counter value is a predetermined value when the third condition that occurs periodically occurs. It is characterized in that a reset device for resetting the following learning counter to an initial value corresponding to a maximum lower limit value is provided.

The digital adaptive control method of the present invention has the characteristics of the above-mentioned method, and at least all counter values are equal to or less than a predetermined value when the third condition occurs periodically. The learning counter is reset to the initial value corresponding to the lower limit value at the maximum.

[0014]

In the device according to the invention and the method according to the invention, at least all learning counters associated with memory locations which have not been fully learned are sometimes permanently forced by resetting them to their initial state, preferably to the value zero. Guaranteed to do the learning. In that case, by selecting the third condition, it is possible to determine the interval at which forced learning is possible. That is, it can be checked whether the sum of all counter values exceeds a predetermined value. This value is determined according to the amount of operating areas corresponding to the memory locations, ie according to the number of memory locations. Furthermore, the dynamic characteristics of the controlled object are taken into consideration. This is because learning is typically performed when leaving the numerical range. When the controlled object is an internal combustion engine and there are 8 × 8 memory locations, the counter value to be reached can be set to several thousands. This counter value is obtained at a few driving time intervals when driving in urban areas.

In the simplest case, the second and third conditions match. That is, when the learning counter is reset, forced learning is also performed. In that case, forced learning must be done before reset. This is because the location where forced learning is performed is determined based on the counter value.

However, when the second and third conditions are different from each other, it is more flexible especially when the forced learning is already performed before the reset. That is, if, for example, all counter values in a thousand are summed, learning can be done in all memory locations where the counter value is still zero,
On the other hand, the counter is reset only when the above-mentioned sum reaches the value of 10,000. This method is especially fast first based on poorly adapted systems, for example after loss of data due to loss of voltage, whereas other forced learning steps are relatively large. It has the advantage that it is done at intervals.

Furthermore, the system is concerned with whether the interval at which forced learning is carried out by proper selection of conditions has already undergone a normal learning process once in a memory location of the operating region which is rarely entered. Sometimes it becomes even more flexible. That is, the flag is set when the learning is being performed normally. This flag is second
When the condition of occurs, it is reset immediately without forced learning. Only if the second condition occurs again and the flag is no longer set at that time,
In addition, forced learning is performed.

In addition to the structural learning values stored in the learning value memory, the apparatus and the method according to the present invention can be implemented with a rough learning value or an adaptation value as described in DE 3539395A1, for example. It does not matter if it is used. Furthermore, how the learned value is combined with the preset control operating value, that is to say for example multiplication or additive, and this combination is performed before correction with the closed-loop control operating value. It does not matter whether it is done or later, or if the joins are made in different ways when the learning values are large.

Also, only when and how the learning was performed, and therefore, for example, every time it deviates from a predetermined operating region, or only a predetermined operating value is stored in such a case, It is also unimportant whether the calculation of the learning values is carried out after summing a number of measured values in a statistical way, for example as described in EP-A-0379091. It is also conceivable to use the learning method considering the counter value together with the forced learning method performed at the start of the driving cycle shown in the SAE paper mentioned at the beginning. All that is important is that the learning counter is reset below a value that forms a limit to activate forced learning whenever a cyclically occurring condition occurs.

[0020]

Embodiments of the present invention will be described below with reference to the drawings.

The diagram of the functional group of FIG. 1 shows the following functional groups: a controlled object formed as an internal combustion engine 10 having an injection device 11 and a lambda value measuring device 12.1 and an internal combustion engine 1
A measuring device 12.2 which measures another operating variable of 0, for example the number of revolutions n and the angle α of the throttle valve, and outputs two address signals ADR1 and ADR2 according to the numerical range of this operating variable.
And an open-loop control operation value map 13 that is driven according to the above-mentioned address signal and outputs a preset control operation value VS related to the operating state, and a lambda closed-loop control device that outputs a closed-loop control operation value FR_IST. 14
An adaptive device 15 for calculating the adaptation coefficient FA from the closed-loop control operation value, and a first correction device 16.1 for combining the adaptation coefficient with a preset control operation value VS in a multiplying manner.
The corrected value described above is used as the operating value FR_IS for closed loop control.
A second corrector 16.2 is provided which is multiply-combined with T.

Further, the adaptive device 15 outputs the deviation signal AS by subtracting the target value FR_SOLL from the learning function memory 17 which can be addressed by the above-mentioned address signal and the closed loop control operation value FR_IST. The learning value is read from the subtraction point 18 and the storage location of the learning value memory 17 according to each address signal, this is changed using the deviation signal, and the changed value is stored as the new learning value in the corresponding storage of the learning value memory 17. A learning or forced learning device 19 to be stored again in a location, a learning counter section 20 addressable by the address signal having the same number of learning counters as the memory locations existing in the learning value memory 17, and a learning counter of the learning counter section 20. Check the value (counter content) and learn all if the specified condition occurs. And a zero-setting device 21 which operates as a reset device for resetting the counter to zero.

Next, the flow of a simple process carried out by the apparatus (system) or method shown in FIG. 1 will be described with reference to FIGS.

In step s1 (see FIG. 5), the measuring device 12.2 detects the measured value relating to the operating variable. After that, the measuring device checks which numerical range the measured value corresponds to and outputs the address signals ADR1 and ADR2 corresponding to the existing regions. If this address changes, it means that you have left the numerical range. In step s2, learning or forced learning device 19
Is removed using the address signals ADR1 and ADR2 supplied to it. If the previous numerical value area still exists, the process returns to step s2. Otherwise, in step s3, the learning value of the memory location of the learning value memory 17 associated with the previous numerical value area is calculated. For that purpose, the learning or forced learning device reads the above-mentioned learning value from the learning value memory 17 and corrects it by using the deviation signal AS from the subtraction point 18. For example, the deviation signal AS with a weight of 0.01 is added to the previous learning value with a weight of 0.99.

In this case, the weighting is performed by the learning counter section 20.
In relation to the related learning counter value of, the larger the counter value, and therefore, the more learning is already performed for each corresponding operating region, the smaller the weighting of the deviation signal becomes. When the learning or forced learning device 19 writes a new learning value in the above-mentioned memory location, the associated learning counter is also incremented.

The state of the learning value memory 17 is shown in FIG. 2 (a), and it can be seen from FIG. 2 that the learning value increases almost linearly from the beginning to the end across the six storage locations shown. Recognize. But the third and last values are off. FIG. 2B of the learning counter of the learning counter unit 20.
It is clear from the counter value shown in (1) that learning has not yet been performed for the third operating region and only little has been learned for the last operating region. The figure further shows that it is desirable to force the value of at least the third memory location to change. This value is about 15% lower than the adjacent value. If the internal combustion engine 10 enters the relevant operating range, an uncorrected preset operating value for control is output to the connection point 16.2, in which all corrections must be carried out using the operating value for closed-loop control. It would have to be done, but that would take a certain amount of time, which would temporarily set the internal combustion engine in an unsatisfactory state.

In order to form a condition for deciding when to perform forced learning, the learning or forced learning device 19 forms the sum S of the counter values of all the learning counters of the learning counter section 20 in step s4. Then (step s5) it is checked whether this sum is equal to the lower limit value GSU. In the display of FIG. 2B, the sum S of all counter values is just 120. It is assumed that this number is set as the lower limit value GSU. In that case, the condition to be checked in step s5 is satisfied, whereby the learning or forced learning device 19
Performs forced learning at step s6 for all memory locations where the contents of the associated learning counter are zero. According to FIG. 2B, this condition is satisfied for the third memory location.

As is apparent from FIG. 3, the learning value to be stored in the third memory location in the forced learning process is formed by the arithmetic mean from the learning values of two adjacent memory locations. If the value of the first memory location has to be force-learned, the value of the second memory location can be easily substituted, while if the last value has to be force-learned, The last value is substituted with the previous one. In that case, it is assumed that the counter value of the learning counter of the memory location where the learning value is used for correction is not zero. If this condition is not met, it continues until it encounters a learning counter with a value greater than zero. A linear interpolation then forms the learned values for the various adjacent memory locations whose associated learning counters all exhibit the value zero.

Since the related learning counters are incremented not only in the case of the normal learning process but also in the case of the forced learning process, even after the forced learning is performed once even if the related driving region is rarely reached. Will prevent further forced learning of the memory location. This can be avoided by not incrementing the learning counter during the forced learning process. However, in that case, especially if the forced learning conditions are selected to be able to be reached relatively early after a completely new start of the system, forced learning will occur at a higher frequency than originally desired. A new start-up of this kind is necessary, for example, when all learned values are set to their initial values after the voltage supply has been interrupted. In that case, it is desirable to perform forced learning of a rarely entered area after learning for a relatively short time in a frequently entered numerical area. Therefore, in a given situation, forced learning is done relatively early, but in normal situations it is done relatively rarely.

In order to satisfy the above conditions, step s7
At, the reset device 21 checks whether the sum S has reached the upper limit value GSO. In the example, this is the value 600. When this value is reached, the reset device 21 sets all learning counters to zero in step s8. This is shown by FIG. 4 (b). In that case, it is possible to perform the forced learning again for the area which has been known for a long time again in step s5.

At step s7, the upper limit value GSO
If it becomes clear that the end condition has not been reached yet, the final step se is proceeded to, where the termination condition is examined. If it is satisfied, the processing is terminated, and if not, the sequence is newly executed from step s1.

The conditional steps s2, s5 and s7 for determining the first, second or third conditions can be modified in various ways. That is, in step s5, it is determined whether a new operation cycle has started (the internal combustion engine 10 has started at an engine temperature of 50 ° C or lower) or has ended (the engine temperature has decreased to 50 ° C or lower). You can The latter is especially checked when a very computationally intensive learning process is carried out. In step s7, whether a predetermined operating period of the control target (in this case, the internal combustion engine 10) has elapsed since the last reset, whether the count value of one predetermined learning counter has exceeded the threshold value, or a predetermined selection is made. It is possible to check whether the counter value of each of the learned counters exceeded a predetermined threshold value.

Furthermore, since the same conditions can be checked in steps s5 and s7 respectively,
When this condition is satisfied, step s6 is first performed, and then step s8 is performed. This allows the system and method shown in FIG. 1 to be used with great flexibility. However, in all cases there is the possibility of repeating the forced learning process.

The learning counter value used for forced learning does not necessarily have to be zero. A value larger than that, for example, a value of 3 can be used. If the counter is reset, it must be set to at least this lower limit. For example, if the lower limit value is the value 3, it can be reset to the value 2. In that case, first, forced learning can be performed twice in a short period of time.
In that case, of course, it is necessary to make a different judgment in step s5, for example, it is necessary to judge whether or not there is a new operation cycle as already described. When the system is newly started, the counter is zero, so assuming that the condition of step s7 is met more rarely than the condition of step s5, all learning counters reach at least the lower limit and the step Until the condition 7 is satisfied, the forced learning can be continuously performed frequently at relatively short intervals until the condition is reset.

In the case of an internal combustion engine, the second and third conditions are always fulfilled as follows: the third condition is met at most as often as the second condition, preferably rather rarely. It is effective to select it so that it is satisfied only.
In other systems, the relationship of this condition can be reversed.

In the above-described embodiment, it is premised that all learning counters are reset to the initial state when the third condition occurs. However, when the counter value indicates that the relevant operating region is not learned too much, that is, when the counter value is less than or equal to the predetermined count value, the learning counter for that counter value It is also possible to reset only the initial state, for example to the value zero. This ensures that memory locations with very well adapted values are not endangered by forced learning. Such a danger is
Usually, it occurs when the controlled object happens to be in a driving state in which it is driven at a very high frequency, for a long time. For example, in the case of an internal combustion engine, when driving on a highway for a long time, the idling operating range which is often often reached is not reached for a long time.

By resetting the learning counter, the system and method of the present invention can perform the forced learning process repeatedly at a predetermined storage location with a simple function. By changing the second and third conditions, it is possible to very finely adapt the function to the operating characteristics of each controlled object.

[0038]

As is apparent from the above description, according to the present invention, forced learning is permanently ensured with respect to a memory location related to a driving region in which a controlled object is rarely driven.

[Brief description of drawings]

FIG. 1 is a block diagram showing a functional group of a system of the present invention and a device of the present invention operating by digital adaptive control.

2A and 2B are explanatory views showing the contents of a learning value memory or a learning counter unit having six storage locations corresponding to each other.

3A and 3B are explanatory views similar to FIGS. 2A and 2B, but the counter value indicates a time point after the forced learning.

FIGS. 4A and 4B are explanatory views similar to FIGS. 3A and 3B, but the counter value indicates a time point after the learning value memory is reset.

FIG. 5 is a flowchart illustrating a digital adaptive control method capable of continuously repeating forced learning processing.

[Explanation of symbols]

 10 Internal Combustion Engine 12 Operating Variable Measuring Device 13 Control Value Map 14 Lambda Closed Loop Controller 15 Adaptive Device 17 Learning Value Memory 19 Learning / Forced Learning Device 21 Reset Device

Front Page Continuation (72) Inventor Jürgen Kurlé, Federal Republic of Germany 7410 Reutlingen Werwaldweg 3

Claims (11)

[Claims] 1. A digital adaptive controller for calculating and storing a learned value when a first condition occurs, wherein the learned value is stored in an addressable memory location via at least one numerical area of the operating variable. Learning value memory (1
7), and a learning counter unit (2) that has a learning counter for each storage location of the learning value memory and that increments each learning counter when a new learning value is written in the associated storage location.
0) and even if the value of the related learning counter reaches when the second condition occurs, the maximum limit value is the lower limit value, so that the learning of the numerical range of the driving variable related to this memory location is not performed. A digital adaptive controller having a forced learning device (19) for replacing a learned value of a memory location with a value obtained by using a learned value of another memory location, And a reset device (21) for resetting at least the learning counter whose counter value is equal to or less than a predetermined value to the initial value corresponding to the lower limit value when the third condition occurs. And a digital adaptive controller. 2. The reset device (21) is configured to check, as a third condition, whether there is a start of a driving cycle of a controlled object driven in the driving cycle. 1. The device according to 1. 3. The reset device (21) is configured to check, as a third condition, whether or not a predetermined operation period of the controlled object has elapsed since the last reset. The device according to. 4. The reset device (21) is arranged, as a third condition, to check whether the sum of the values of all learning counters has reached an upper limit value. 1. The device according to 1. 5. Reset device (21) according to claim 1, characterized in that, as a third condition, it is arranged to check whether one of the counter values has reached at least a threshold value. apparatus. 6. The reset device (21) is configured to check whether the respective values of the predetermined number of learning counters selected as the third condition have reached at least the respective predetermined threshold values. The device according to claim 1, wherein 7. The forced learning device (19) is configured to use the same condition that the reset device uses as the third condition as the second condition. The apparatus according to any one of 1. 8. Device according to claim 1, characterized in that the forced learning device (19) is arranged to use the start of the driving cycle as the second condition. .. 9. The forced learning device (19) is arranged to check, as a second condition, whether the sum of all counter values has reached an upper limit value. The device according to any one of claims 1 to 6. 10. Device according to claim 1, characterized in that the reset device (21) uses the value zero as an initial value. 11. A learning value is calculated and stored when each of the first conditions occurs, and the learning value is stored in an addressable memory location through the following steps, namely at least one numerical area of the operating variable. Is stored, the learning counter for each memory location is incremented when a new learning value is written in the relevant memory location, and even if the value of the relevant learning counter reaches the maximum when the second condition occurs, It is a limit value, and when it is shown that learning has not been done for the numerical region of the driving variable related to this memory location, the learning value of the memory location is calculated using the learning values of other memory locations. In the digital adaptive control method in which the value is forcibly replaced with a predetermined value, at least when the third condition that occurs periodically occurs, the counter value is at least a predetermined value or more. Digital adaptive control method characterized by resetting the initial value corresponding to the lower limit value of all of the learning counter going on to a maximum.