JP2002099890A - Automatic program generation method and automatic program generation device - Google Patents

Abstract

(57) [Problem] To provide an automatic program generation method and apparatus with high versatility, with few restrictions on the configuration of an apparatus capable of using an automatically generated program. SOLUTION: Each gene constituting the chromosome constitutes one step of the program, and the content of the gene is determined by the operation amount of the actuator of the automatic control device and the value of the sensor of the automatic control device using the automatically generated control program. The coding format of the chromosome is determined so that it can be determined, a chromosome of a predetermined evaluation is obtained by a so-called genetic algorithm, and the chromosome is decoded, whereby a control program of a desired control characteristic is automatically generated. Is to be done.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for automatically generating a program, and more particularly to a method for automatically generating a control program in an automatic control device having an arithmetic processing unit represented by a microcomputer. Then, it relates to the application of a genetic algorithm.

[0002]

2. Description of the Related Art A so-called automatic control device having an arithmetic processing unit represented by a microcomputer, a plurality of actuators, and a plurality of sensors generally has a plurality of operation units. The control program is controlled by using the signals detected by the plurality of sensors to obtain a desired operation result. Conventionally, a control program in such an automatic control device has been generally created by trial and error work of a so-called programmer.

[0003]

However, as the number of actuators to be controlled is increased, the structure of the automatic control device is complicated, and as the size of the automatic control device is increased, so-called manual operation of the control operation is required. Creating a control program not only requires a lot of effort, but also
As a result, there is a problem that the price of the apparatus is increased. On the other hand, although there has been no proposal of a device for automatically generating a program, conditions for a device that can use an automatically generated program are often specified, and the device lacks versatility and is satisfactory. There were few. The present invention
An object of the present invention is to provide an automatic program generation method and an automatic program generation apparatus which have little restrictions on the configuration of an apparatus which can use an automatically generated program and have high versatility.

[0004]

In order to achieve the above object, an automatic program generation method according to the present invention is an automatic program generation method for automatically generating a control program for controlling the operation of an automatic control device. Generate an initial population of chromosomes coded in a predetermined coding format, based on a predetermined evaluation function, calculate the fitness of each chromosome constituting the initial population, the chromosomes satisfying the predetermined fitness is It is determined whether or not there is, and when it is determined that there is a chromosome that satisfies a predetermined fitness, a program obtained by decoding the chromosome is used as a control program to be obtained, while a predetermined fitness is determined. When it is determined that there is no chromosome to be satisfied, a selection process, a crossover process, and a mutation process are sequentially performed on a chromosome group that does not satisfy the predetermined fitness, After that, performing a fitness calculation based on a predetermined evaluation function, determining whether there is a chromosome that satisfies the predetermined fitness, and performing a selection process, a crossover process, and a mutation process on a chromosome population, the predetermined fitness Iteratively until a chromosome satisfying is obtained, and if a chromosome satisfying a predetermined fitness is obtained, a program obtained by decoding the chromosome is a control program to be obtained,
The selection processing is to randomly select two chromosomes from a chromosome population under a first predetermined condition, and the crossover processing is a second predetermined chromosome selection for the two chromosomes selected by the selection processing. Under the conditions, the contents of each other are crossed to generate two new chromosomes, and the mutation process randomly selects a chromosome at a predetermined ratio from the chromosomes subjected to the crossover process, and One of the selected chromosomes is randomly selected, the content of the gene is rewritten, and the chromosome is mutated.

In such a configuration, in particular, a chromosome is coded in a predetermined format so that a gene constituting the chromosome corresponds to one step of the program, thereby enabling automatic generation of a program by a so-called genetic algorithm. It is. As the coding format of the chromosome, a plurality of genes constituting the chromosome are each composed of one function fragment and a plurality of element fragments, and the function fragment is a portion in which an element for determining the program character of the entire gene is set. Wherein the element is selected from a plurality of pre-selected ones of a programming language used for the control program, and the plurality of element fragments are the number of actuators of the automatic control device which is the object of the control program. And are provided corresponding to the sum of the number of sensors,
It is preferable that the operation amount of the corresponding actuator or the value of the sensor is set. With this configuration, there is little restriction on the configuration of an apparatus that can use an automatically generated program, and a highly versatile program automatic generation method can be provided.

[0006]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below with reference to FIGS. The members, arrangements, and the like described below do not limit the present invention, and can be variously modified within the scope of the present invention. First, a configuration of an automatic program generation device (hereinafter, referred to as “this device”) according to an embodiment of the present invention will be described with reference to FIG. The present apparatus is for automatically generating a control program for an automatic control device as described later in detail. Here, the automatic control device includes an arithmetic processing unit typified by a microcomputer, a plurality of actuators, and a plurality of sensors. A control program is executed in the arithmetic processing unit, and The operation of the plurality of actuators is controlled based on detection signals, data, and the like, so that a desired operation state is obtained. Note that the actuator is not necessarily provided with the name of the actuator as described in an example of an automotive air conditioner described later, and performs an operation by control. For example, a blower motor in an automotive air conditioner is used. (Not shown) can also be regarded as the actuator referred to herein. In the embodiment of the present invention, the automatic control device is assumed to be an automotive air conditioner, and the automatic generation of the program will be described below.

First, the present apparatus is roughly divided into an input unit 1 as input means, a central processing unit 2, and a display unit 3 (see FIG. 1). The input unit 1 is for inputting commands, data, and the like to the central processing unit 2. For example, an interface circuit (not shown) for interfacing a so-called keyboard (not shown) with the central processing unit 2. And known as the main components
It has a well-known configuration. The display unit 3 displays commands and the like input from the input unit 1 to the central processing unit 2 and displays automatically generated control programs and the like.
For example, it has a known / known configuration using a so-called CRT, a liquid crystal display element, or the like. The central processing unit 2 is for automatically generating a control program for an automatic control device based on a so-called genetic algorithm, as described later. The central processing unit 2 according to the embodiment of the present invention includes an arithmetic control unit 4 as arithmetic control means.
And a selection / crossover processing unit 5 as selection / crossover processing means
, A mutation processing unit 6 as a mutation processing unit, an evaluation operation unit 7 as an evaluation unit, and a simulation unit 8 as a simulation unit. Can be realized by executing software in accordance with a processing procedure described later using the microcomputer.

The arithmetic and control unit 4 generates a chromosome as a control program and controls the overall operation of the apparatus according to a processing procedure described later. The selection / crossover processing unit 5 performs a selection process and a crossover process on chromosomes in accordance with a processing procedure described later. The mutation processing unit 6 performs a mutation process on the chromosome that has been subjected to the selection / crossover process in accordance with a processing procedure described later. The evaluation calculation unit 7 calculates the fitness based on a predetermined evaluation function as described later. The simulation unit 8 simulates the automatically generated control program and generates data necessary for the calculation of the evaluation function. In the embodiment of the present invention, although details will be described later, a so-called neural network is used. It has been built.

Next, an air conditioner for a vehicle which is an object of automatic generation of a program by the present device, in other words, an air conditioner for a vehicle using a control program automatically generated by the present device will be described with reference to FIG. explain. This automotive air conditioner is generally referred to as a so-called reheat air mix system, which cools all the air taken in from the intake once and then reheats (reheats) a part of it to produce high-temperature air. This is mixed with the remaining low-temperature air to obtain air at a desired temperature. That is, a blower 12 is provided near an inlet (not shown) formed at one end of the air-conditioning duct 11, and guides air from the inlet into the air-conditioning duct 11 to blow air. An evaporator 13 is provided downstream of the blower 12 so that the air blown by the blower 12 is cooled. And the evaporator 13
A mix door 14 is provided on the downstream side of the
On the downstream side of the mix door 14, the air conditioning duct 11 is divided into two flow paths over an appropriate length, and a heater core 15 is provided on one of the two flow paths.

In accordance with the opening of the mixing door 14, a part of the air that has passed through the evaporator 13 is passed through the heater core 15 and reheated, and is mixed with air that has not been passed through the heater core 15. Thus, a temperature-adjusted mixed air can be obtained.
Further, near the other end of the air conditioning duct 11, a mode door 16 is provided, and depending on the position of the door, the air distribution ratio between the upper and lower sides of the vehicle compartment is changed, and the preceding mixed air is adjusted to the air distribution ratio. The air is distributed up and down in the passenger compartment accordingly.

The automotive air conditioner has an arithmetic processing unit 17 for controlling the rotation speed of a blower motor (not shown) for rotating the blower 12 and the position of the mix door 14 and the mode door 16. , Are controlled by the execution of the control program by the arithmetic processing unit 17. First, the blower motor (not shown)
Through a digital / analog converter (denoted as “D / A” in FIG. 2) 18 and a pulse width modulation signal output circuit (denoted as “PWM” in FIG. 2) 19 provided together with the arithmetic processing unit 17, The rotation speed is controlled by executing a control program in the arithmetic processing unit 17. That is, a control signal is output from the arithmetic processing unit 17 by executing the control program in the arithmetic processing unit 17, the control signal is converted into an analog signal by the digital / analog converter 18, and the pulse width modulation signal output circuit 1
9 is input. And a pulse width modulation signal output circuit 1
From 9, a pulse width modulation signal corresponding to the input analog signal is output and applied to a blower motor (not shown), whereby the rotation speed is determined.

Also, the mix door 14 and the mode door 1
For the position control of No. 6, a first motor actuator (indicated as “ACT (1)” in FIG. 2) 20 and a second motor actuator (“ACT (2)” in FIG. 2)
2), and an input / output circuit (FIG. 2) for interfacing the first and second motor actuators 20, 21 with the arithmetic processing unit 17.
, "I / O") 22 is provided.
The execution of the control program in the arithmetic processing unit 17 outputs a control signal for door position control from the arithmetic processing unit 17, and the first motor actuator 20 and the second motor When the actuators 20 and 21 are applied to the actuator 21 and operate in accordance with the control signals, the door position is controlled.

Further, an evaporator sensor 23 for detecting the substantial temperature of the evaporator 13 is provided, for example, on the downstream side near the evaporator 13 or on the evaporator 13.
And the heater core 1
5 for detecting a substantial temperature of the heater core sensor 24
Are provided in the same manner, and these detection signals are input to the arithmetic processing unit 17. Further, a vehicle interior temperature sensor 25 for detecting the temperature of the vehicle interior is provided in the vehicle interior, and an external temperature sensor 26 for detecting the external air temperature is provided outside the vehicle at appropriate positions. Is also input to the arithmetic processing unit 17. The detection signals of these sensors 23 to 26 are used as so-called feedback information in the execution of the control program in the arithmetic processing unit 17.

Next, a procedure for automatically generating a control program for the vehicle air conditioner having the above-described configuration by the present apparatus will be described with reference to FIGS. First, the program automatic generation method according to the embodiment of the present invention generally generates a control program for an automatic control device automatically based on a so-called genetic algorithm. FIG. 3 shows a specific procedure. Hereinafter, an automatic program generation procedure based on a genetic algorithm will be described with reference to FIG. When this apparatus is started, first, an initial population of chromosomes is generated, and the generation counting variable N is set to “1” (see step S100 in FIG. 3). That is, a population of chromosomes to be first targets of the respective processes of selection, crossover, and mutation described below is generated. Here, in the embodiment of the present invention, the chromosome that is the solution candidate is a candidate for the control program to be obtained, and therefore, the coding format of the chromosome is defined as described below. That is, first, as shown in FIG. 4 (A), the point that the chromosome is composed of a plurality of genes is as follows.
It is the same as the conventional genetic algorithm, but in the embodiment of the present invention, the number of genes on one chromosome is variable. However, at the time of generation of the initial group, a predetermined number is set.

The data structure of this gene is unique to the present invention. Specifically, as shown in FIG. 4 (B), the inside of one gene is divided into a plurality, and one function fragment located at the top is
It is composed of a plurality of element fragments that follow. The number of element fragments is a total number of the number of sensors and the number of actuators of the automatic control device to be automatically generated. That is, in the case of the automotive air conditioner shown in FIG. 2 for which a program is to be generated in the embodiment of the present invention, the evaporator sensor 23, the heater core sensor 24, the vehicle interior temperature sensor 25 are used as temperature sensors and data. And five of the outside air temperature sensor 26 and the target temperature correspond to the blower motor (not shown) and the first actuator 20, respectively, as actuators, and therefore, the number of element fragments is set to seven. (See FIG. 4B). The function fragment determines the overall character of the gene corresponding to one step of the program. In the embodiment of the present invention, the function fragment includes “IF”,
One of "ENDIF" and "SET" is set,
In the case of "IF", the gene is decoded as a conditional statement, and in the case of "SET", the gene is
This is decoded as a program for setting the operation amount of the actuator.

In the embodiment of the present invention, the seven element fragments are, specifically, the head (the next element fragment of the function fragment)
From the blower rotation speed (Blower speed), evaporator temperature (Temp. (Evaporator)), mix door opening (Mixdoo
r opening), heater core temperature (Temp. (heater)), vehicle interior temperature (Temp. (room)), outside air temperature (Temp. (outdoor)), target temperature (Temp. (target)) (See FIG. 4B).

The interior of the element fragment is further divided into a plurality of parts, and in the embodiment of the present invention, the element fragment is divided into six parts. In the embodiment of the present invention, each of the six divided element fragments is referred to as A value, B value,
The meaning and content are defined as follows as values, D values, E values, and F values. That is, the A value is set to either “0” or “1”,
"0" means that the value of the element fragment is the value of the sensor, and "1" means that the value of the element fragment is the operation amount of the actuator ( FIG. 5 (A)).

In the B value, the type of element fragment, that is, the name or abbreviation of the actuator or sensor is set. The C value indicates whether or not to use an element fragment. If “TRUE”, the “TRU”
"E" means to use the element fragment in which it is written, and if "FALSE", this "FALS"
"E" means that the element fragment in which the information is written is not used (see FIG. 5B). The D value is expressed as “I
This is used when "F" is set, and a logical operator in C language is set. Then, when this element value is used in an element fragment in a conditional statement (IF statement),
This is for defining whether to take AND (logical product) or OR (logical sum) with the condition indicated by the other element fragments. That is, if the D value is “1”, it means “&&” in C language that represents a logical product (AND), and if the D value is “2”, the logical sum (OR ) In the C language, and the D value is “3”.
, It means "&&!" In C language, which represents negation of logical product (NAND), and if the D value is "4", it represents negation of logical sum (NOR). This means "||!" In the C language (see FIG. 5C).

For the E value, a comparison operator is set. That is, when the E value is “0”, it means that they are equal,
When the value is “1”, it means that the inequality sign is “>”. When the E value is “2”, it means that the inequality sign is “<”, and when the E value is “3”. "Means that the absolute value is small, and when the E value is" 4 ", it means that the absolute value is large (FIG. 5).
(D)). When the function fragment is set to “IF”, the F value is set to a specific numerical value that is used as a reference value in the conditional statement, and when “SET” is set, the actuator A specific numerical value indicating the amount of operation of is set.

The generation of an initial population of chromosomes having such a configuration is performed using, for example, random numbers. That is, first, a mutual relationship (a so-called conversion table) is determined in advance so that each value of the above-described function fragment and element fragment is determined according to the value of the random number. It is stored in an appropriate storage area.
Then, the arithmetic control unit 4 sequentially generates random numbers,
From the obtained values, the contents of the function fragments and the element fragments are sequentially determined by the conversion table, and an initial population of chromosomes is generated.

Next, an example of a chromosome gene coded according to the above rules and its decoding will be described with reference to FIG. The gene shown in FIG. 6 is an example in which “IF” is set in the function fragment. Therefore, the decoding result of this gene is IF
Will be treated as a sentence (conditional statement). First, looking at the element fragment immediately after the function fragment, since the A value is “1”, this first element fragment is decoded as corresponding to the actuator. And B
The value specifically describes what kind of actuator the actuator is. As described above, since the first element fragment is related to the blower rotation speed, "Blow
er speed ". For the C value, "FALS
Since "E" is set, this first element fragment is decoded as not being used in the gene in this example. Therefore, for the first element fragment, the remaining D value, E value and F
Regardless of the content of the value, it will not be used as a control program.

The second element fragment in FIG. 6 has the A value set to “0” and is decoded as an element fragment relating to the sensor. And the B value is "TEva"
And an abbreviation of the evaporator temperature are set, and this element fragment is decoded as being related to the evaporator temperature. Then, the C value of this element fragment is “TRUE”
Thus, this element fragment is decoded as being used. Next, since the D value is set to “1”, the value is decoded to “AND”. Here, the logical operator decoded from the D value specifies a logical operation with the content of the immediately preceding element fragment. In this example, the immediately preceding element fragment is described in the above. As described above, since the C value is “FALSE”, in this case, “AND” has no meaning. Next, the E value is set to “1” and is decoded as “>”. The E value comparison operator specifies a comparison operation between the contents of the C value and the contents of the F value. In this example, the F value is “0.45
6 ", the second element fragment is eventually decoded as"TEva> 0.456 ".

Next, looking at the contents of the third element fragment, first, the A value is set to "1", and it is decoded as an element fragment relating to the actuator. In the B value, “MixOpen” and the abbreviation of the mixed door opening are set, and the third element fragment is decoded as being related to the mixed door opening.
Next, since the C value is set to “FALSE”, the third element fragment in this case is decoded to be unused. In the following description, the description of the contents after the D value will be omitted for the unused element fragments.

Next, the A value of the fourth element fragment is “0”.
And is decoded to be an element fragment for the sensor. Then, “THeat” and an abbreviation of the heater core temperature are set in the B value, and the element fragment is decoded as being related to the heater core temperature. Next, since the C value is set to “TRUE”, this element fragment is decoded as being used. Next, since "4" is set in the D value, it is decoded as "||!".
This is because the second element fragment and the fourth element fragment are “OR” and “NOT”, in other words, “NOR”.
Means to be joined. The E value is decoded to mean that the absolute value is small because “3” is set. Further, since the F value is set to “0.012”, the fourth element fragment part is eventually decoded as “||! Fabs (THeat) <0.012”. Become. Here, “fabs (x)”
Represents the absolute value of x, and represents | x | in C language.

Next, the A value of the fifth element fragment is “0”.
And is decoded to be an element fragment for the sensor. In the B value, “TRoom” and an abbreviated name of the vehicle interior temperature are set, and this element fragment is decoded as being related to the vehicle interior temperature. Then C
Since the value is set to “FALSE”, the fifth element fragment in this case is decoded to be unused.

The A value of the sixth element fragment is set to "0", and is decoded as an element fragment relating to the sensor. In the B value, “TOut” and an abbreviated name of the outside air temperature are set, and the element fragment is decoded as being related to the outside air temperature. Next, the C value becomes “TR
Since "UE" is set, this element fragment is decoded as being used. next,
Since "3" is set for the D value, "&&
! Is decoded. This is because the fourth element fragment and the sixth element fragment are “AND” and [NO
T ", in other words, it is connected by" NAND ". Since “0” is set in the E value, it is decoded as “==”. Further, the F value is “0.6
78 ", the sixth element fragment eventually becomes"&&! (TOut == 0.67
8) ".

Next, the A value of the seventh element fragment is “1”.
And is decoded as an element fragment for the actuator. And the B value includes “TTarge
"t" and an abbreviation of the target temperature in the passenger compartment are set, and this element fragment is decoded as relating to the target temperature in the passenger compartment. Next, since the C value is set to “FALSE”, the seventh element fragment in this case is decoded to be unused.

Therefore, the decoding result of the gene shown in FIG. 6 as a whole is as follows.

If ((TEva> 0.456) ||! (Fab
s (THEat) <0.012) &&! (TOut = 0.67
8)) ｛

[(TEva> 0.4) in the first parenthesis ()
56) ｜｜ (Fabs (THeat) <0.012) &&!
If (TOut == 0.678) is rewritten with a normal logical operation expression, “(TEva> 0.456) NOR (fabs (T
Heat) <0.012) NAND (TOut == 0.67)
8) ". In other words, the meaning is
"(TEva> 0.456) and (fabs (THeat)
<0.012 and not (TOut = 0.678). " After all, considering the IF set in the function fragment, the program on the gene in this example is "if,
(TEva> 0.456) and (fabs (THeat) <
If it is not 0.012 and (TOut = 0.678), the following is executed. "

Next, an example of a gene in which “SET” is set in a function fragment will be described with reference to FIG. First, as a common rule when decoding a SET sentence, the D value and the E value are ignored. This is because they make sense in the case of an IF statement (when the function fragment is a gene in which “IF” is set).

Looking specifically at the example in FIG.
Since the first element fragment has the A value set to “1”, this element fragment is decoded as relating to the actuator, and the B value is set to “Blower speed”. It is decoded as relating to the rotation speed. Since “TRUE” is set in the C value, this element fragment is decoded as being used. In addition, the F value includes
Since it is set to “0.123”, the first element fragment is eventually “SET” together with the function fragment.
(Blower speed, 0.123) ". In other words, it means "set the blower rotation speed to 0.123".

Next, the second element fragment has a C value of “FA
Since it is set to "LSE", it is not used as a program. Next, since the A value of the third element fragment is set to “1”, this element fragment is decoded as relating to the actuator, and the B value is set to “MixOpen”. , It is decoded as relating to the mix door opening. Also, since the C value is set to “TRUE”,
This element fragment will be decoded as being used. Then, since the F-number is set to “0.789”, this third element fragment eventually ends up with “SET (MixOpen,
0.789) ”. In other words, it has the meaning of “set the mix door opening to 0.789”.

For the fourth and fifth element fragments,
In both cases, the C value is set to “FALSE”, so that it is not used as a program. The sixth element fragment pertains to the sensor and does not make sense in this SET sentence, so it is not decoded.

For the seventh element fragment, the C value is "F
Since it is set to "ALSE", it will not be used as a program. As a result, the decoding result of the gene shown in FIG. 7 as a whole is as follows.

SET (Blower speed, 0.123); SET (MixOpen, 0.789);

Here, returning to the description of FIG. 3, after the generation of the initial population, whether the individual chromosomes of the initial population are sufficient (in other words, as a desired control program, An evaluation process for determining whether or not it is sufficient is performed by the evaluation calculation unit 7 and the simulation unit 8 (see step S200 in FIG. 3). That is, although the details will be described later, the fitness of the chromosome as a desired control program is calculated using a preset evaluation function. After the evaluation processing is performed according to the predetermined procedure, the arithmetic control unit 4 determines whether the obtained fitness is the predetermined fitness (see step S300 in FIG. 3). Then, when it is determined that the predetermined fitness is obtained (in the case of YES), a series of processing is ended, while when it is determined that the fitness is not the predetermined fitness (in the case of NO) Then, the selection / crossover process is performed by the selection / crossover processing unit 5 as one process of the genetic algorithm (see step S400 in FIG. 3). That is, first, the selection process is one of processes for generating a next-generation population based on a population of chromosomes that did not satisfy the predetermined fitness, together with a crossover process and a mutation process described below, In general, it is a process of selecting a chromosome under a predetermined condition from a group that does not satisfy a predetermined fitness. Here, as a predetermined condition for selecting a chromosome from a group, a chromosome is selected from a descending order of fitness with a high probability. Here, in order to select the one with the higher fitness with a high probability, for example, each chromosome is so-called weighted in advance, a random number is generated, and the chromosome is selected according to the random number. In this case, the weight may be selected in consideration of the weighting, and such a method is a known / known method in a field called so-called data processing or numerical processing, and a detailed description thereof is omitted here. I decided to.

After the selection process is performed as described above, the selected chromosome is subjected to a crossover process. That is, the crossover process involves exchanging genes at certain positions of two chromosomes selected at a time. More specifically, a position (intersection point) at which gene exchange is performed is determined in advance, and the contents of the latter half are exchanged between two chromosomes from this position, and the intersection point is basically arbitrary. It may be determined in.

Next, the mutation processing is performed by the mutation processing unit 6 on the chromosome subjected to the selection / crossover processing (see step S500 in FIG. 3). That is, in the mutation process, a chromosome to be subjected to the mutation process at a predetermined ratio (predetermined mutation rate) is selected from the chromosome population subjected to the crossover process, and one gene is assigned to the chromosome. This is performed by arbitrarily selecting and arbitrarily rewriting the contents. Here, the content of rewriting differs depending on the selected gene,
It is preferable that the content of the rewriting is provided with a selection range for each gene, and this is determined by, for example, a random number. After the completion of the mutation process, the generation counting variable N is incremented by one and updated (see step S600 in FIG. 3). Then, returning to the previous evaluation process, the chromosome of the new generation is evaluated, and thereafter, each process is repeated until a chromosome of a predetermined fitness is generated as described above. .

Next, the specific contents of the evaluation processing will be described with reference to FIGS. First, in the evaluation of chromosomes in the genetic algorithm, an evaluation function is determined in advance according to the content of the chromosome to be generated by the genetic algorithm, and for the chromosome to be evaluated,
This evaluation function is calculated, the fitness as a result of the calculation is obtained, and the suitability of the generated chromosome is determined depending on whether a predetermined fitness is obtained. The basic idea of this evaluation is to model a target system (automatic control device) on a computer and evaluate the response of the system when a program on a chromosome is executed at a certain step. For example, when generating a control program at the time of starting air conditioning, the time to reach the target temperature may be used as the evaluation value. From this point of view, in the embodiment of the present invention, as the evaluation function for calculating the fitness, the target temperature in the passenger compartment by the control program obtained in the form of chromosomes and the control program were executed in the vehicle air conditioner. In this case, the smaller the deviation from the estimated vehicle interior temperature, the higher the fitness. Specifically, what is expressed by the following equation 1 is set as the evaluation function.

[0041]

(Equation 1)Where T_target ^{( m )}Is the target temperature at a certain time m
And T_cab ^{( m )}Is the room bill at a certain time m
Table temperature. And T_cab ^{( m )}Is given by the following equation 2.
Is required.

T _cab ^{( m )} = (T _head ^{( m )} + T _foot ^{( m )} ) / 2 (Equation 2)

In Equation 2, T _head ^{( m )} is a certain time m
_{Is the} temperature near the head in the vehicle cabin, and T _foot ^{( m
)} Is the temperature near the legs in the vehicle cabin at a certain time m. The evaluation function is obtained by calculating the difference (T _target ^{( m ))} between the target temperature T _target and the indoor representative temperature T _cab within a predetermined time period (30 minutes in the embodiment of the present invention) at predetermined time intervals. −T _cab ^{( m )} ) is determined as the reciprocal of the integrated value of _{も の} of the squared value of the fitness value. Here, the predetermined time interval is a 5-second interval in the embodiment of the present invention.

Then, in order to obtain a deviation between the target temperature in the passenger compartment by the control program obtained in the form of a chromosome and the estimated temperature in the passenger compartment obtained when the control program is executed in the vehicle air conditioner, In the embodiment of the present invention, a neural network based on data at the time of actual operation of the vehicle air conditioner is set in advance (details will be described later), and a control program is transmitted to the vehicle air conditioner by the neural network. The simulation of the vehicle interior temperature change at the time of application is performed for a predetermined time (30 minutes in the embodiment of the present invention). That is, first, in the embodiment of the present invention, a first neural network 31 (see FIG. 8A) for obtaining the estimated temperature of the evaporator 13 and the estimated temperature of the heater core 15 and the head of the occupant in the vehicle compartment Two of the second neural network 32 (see FIG. 8B) for obtaining the estimated temperature near the part and the estimated temperature near the leg are constructed in the simulation part 8.
The basic configuration of each of the neural networks 31 and 32 is based on a well-known and well-known neural network. In the embodiment of the present invention, any of the input layer 6, the output layer 2, and the intermediate layer 10 is used. This is a so-called hierarchical neural network.

Two neural networks 31, 32
Are used to input six data: blower rotation speed, evaporator temperature, mix door opening, heater core temperature, indoor representative temperature, and outside air temperature. Generally, as is well known, a neural network needs to be learned in advance. In the case of the first neural network 31, as training data in this learning, a predetermined time interval (embodiment of the present invention) At 5
The evaporator temperature and the heater core temperature for the input data at every second interval). In the case of the second neural network 32, the vicinity of the head for the input data at the predetermined time interval (5 second interval in the embodiment of the present invention). Temperature and temperature near the leg. In order to obtain these data, in the embodiment of the present invention, the operating conditions of the actual air conditioner for a vehicle are variously changed, and 3710 kinds of data (about 5 minutes) are obtained at a sampling interval of 5 seconds. The data corresponding to the input data and the output data of the first and second neural networks 31 and 32 were measured.
Then, from the measured data, inconsistent data such as data having different output values with respect to the same input was deleted, and 2793 types of data were actually used for learning.
The learning was performed 500,000 times by the back propagation method. In particular, in the embodiment of the present invention, each time one pattern to be learned is presented, the so-called successively corrected moment method that repeats the operation based on the error back propagation method is used for the intermediate layer and the output. Made necessary corrections to layers and hidden layers.

FIGS. 9 and 10 show the first and second neural networks 31 and 3 obtained as described above.
In order to confirm the accuracy of No. 2, estimated temperatures and measured values obtained by the first and second neural networks 31 and 32 are shown. In FIGS. 9 and 10,
“NN” means a neural network. That is, FIG. 9 and FIG. 10 show 30 points using the first and second neural networks 31 and 32 for which learning has been completed.
Over a period of 5 minutes, the respective characteristic lines showing changes when estimating values of the evaporator temperature, the heater core temperature, the temperature near the head and the temperature near the leg are obtained every 5 seconds, and the first and second characteristic lines.
Characteristic lines showing changes in the actually measured values corresponding to the above-described respective estimated temperatures when the actual vehicle air conditioner is operated under the same input data conditions as when the neural networks 31 and 32 are operated. It is shown.
In each of the figures, the characteristic line indicating the change characteristic of each estimated value and the characteristic line indicating the change characteristic of the actually measured value almost coincide with each other, and both the first and second neural networks 31 and 32 perform the estimation. It can be confirmed that the temperature is sufficient to determine the temperature.

Here, a procedure for obtaining the fitness of the chromosome by the above-described equation 1 using the first and second neural networks 31 and 32 as described above is shown in FIG.
This will be described again with reference to FIG. First, the overall flow will be described with reference to FIG. 11. First, the timing operation of the first timer T1 is started (see step S202 in FIG. 11). At the same time, one step of the control program on the chromosome is executed, and if there is one corresponding to the input data of the first and second neural networks 31 and 32 in the control program, the value is changed to the first and second neural networks. 2 are input as input data to the second neural networks 31 and 32 (see step S210 in FIG. 11). In FIG. 11, “NN” means “neural network”. Next, it is determined whether or not a predetermined time tb has elapsed from the start of the timing of the second timer T2 (see step S250 in FIG. 11). Here, T2 is a program execution process described in detail later (step S2 in FIG. 11).
10) is a second timer started to measure time, and tb is set to 5 seconds in the embodiment of the present invention.

If it is determined in step S250 that the predetermined time tb has elapsed (in the case of YES), the above-described calculation of Expression 1 is performed by the evaluation calculation unit 7 (step S252 in FIG. 11). reference). Next, it is determined whether or not a predetermined time ta has elapsed from the start of the timing of the first timer T1 (step S2 in FIG. 11).
54). Here, ta is set to 30 minutes in the embodiment of the present invention. This step S25
In 4, when it is determined that the predetermined time ta has elapsed (in the case of YES), a series of processing ends, and the process returns to the main routine (see FIG. 3). On the other hand, if it is determined in this step S254 that the predetermined time ta has not yet elapsed (NO), the process returns to the previous step S210, and the same processing is repeated for the next step of the control program. Becomes

As described above, in the embodiment of the present invention, as the chromosome evaluation process, the control program on the chromosome is executed step by step at intervals of 5 seconds, and the first and second control programs are executed by executing the control program. Data is input to the second neural networks 31 and 32, and estimated values of the evaporator temperature, the heater core temperature, the temperature near the head and the temperature near the legs are obtained at 5-second intervals. Then, each time, the half of the square of the deviation between the target temperature and the indoor representative temperature is obtained for 30 minutes, and the fitness according to Equation 1 is obtained. Here, since the indoor representative temperature obtained by Expression 2 is based on the estimated values of the temperature near the head and the temperature near the legs as described above, in other words,
It can be said that it is the estimated temperature in the cabin.

FIG. 12 shows a specific processing procedure in the previous step 210, and its contents will be described below with reference to FIG. The control program on the chromosome is executed step by step (in other words, for each gene) at regular time intervals (in the embodiment of the present invention, every 5 seconds) according to the procedure shown in FIG. . That is, when the process is started, first, an instruction (command) on the gene is read out (see step S212 in FIG. 12). That is, the contents of the function fragment are read. Next, it is determined what the read command is, and when it is determined that the command is “IF”, the process proceeds to step S216, while it is determined that the command is other than “IF” (Other). If so, the process proceeds to step S222.

First, in step 216, the content of the element fragment following the function fragment is decoded, the content of the IF statement is determined, and a jump to the gene at the branch destination is performed according to the content ( Step S21 in FIG.
8). Next, it is determined whether or not the gene at the branch destination is at the end of the chromosome. When it is determined that the gene is at the end (in the case of YES), the process proceeds to step 232 to be described later, while not at the end. Is determined (in the case of NO), a series of processing ends (FIG. 12).
Step 220).

On the other hand, in step 222, it is determined whether the command is “SET” or “ENDIF”. If it is determined that the command is “ENDIF”, the process in step S224 is not performed. The process proceeds to step S226 described below. When it is determined in step 222 that the command is “SET”, the C value of the element fragment is 1, that is, the value of the actuator to be used as described above is read out. And this is the first and second neural networks 3
1 and 32 (step S in FIG. 12).
224). In addition, in step S224 of FIG. 12, "NN" means a neural network. Next, the timer operation of the second timer T2 is started (see step S226 in FIG. 12), and one step is advanced (see step 228 in FIG. 12). Then, it is determined whether or not it is the end of the chromosome, and if it is determined that the end is the case (YES), the first gene is designated as a processing target, and a series of processing is temporarily terminated. (See step S232 in FIG. 12). On the other hand, if it is determined in step 230 that it is not the end of the chromosome (NO), the above-described step S232 is performed.
A series of processes is temporarily terminated without performing the process of (1).

[0053]

As described above, according to the present invention,
Coding the chromosome in a predetermined format so that the genes constituting the chromosome correspond to one step of the program,
Since a desired program can be automatically generated by a so-called genetic algorithm, a desired program can be obtained without manually creating a program, thereby reducing work time and, consequently, automatically generating a control program. It is possible to reduce the price of the device using the device. In addition, chromosome coding is based on an automatic control device having an actuator and a sensor and an arithmetic processing unit, thereby providing a versatile program automatic generation method and apparatus. This has the effect of being able to do so.

[Brief description of the drawings]

FIG. 1 is a configuration diagram illustrating a configuration example of an automatic program generation device according to an embodiment of the present invention.

FIG. 2 is a configuration diagram showing a configuration of an automotive air conditioner for which a control program is automatically generated in the embodiment of the present invention.

FIG. 3 is a main flowchart showing a processing procedure of automatic program generation by the automatic program generation device shown in FIG. 1;

FIG. 4 is an explanatory diagram illustrating a coding format of a chromosome according to the embodiment of the present invention.

FIG. 5 is an explanatory diagram illustrating values set in element fragments.

FIG. 6 is an explanatory diagram showing a first example of chromosome coding.

FIG. 7 is an explanatory diagram showing a second example of chromosome coding.

FIG. 8 is a schematic diagram schematically illustrating a configuration of first and second neural networks according to the embodiment of the present invention. FIG. 8A schematically illustrates a configuration of the first neural network. FIG. 8B is a schematic diagram schematically showing the configuration of the second neural network.

FIG. 9 is a characteristic diagram showing a time change between the evaporator temperature and the heater core temperature obtained by the first neural network and the measured values thereof according to the embodiment of the present invention.

FIG. 10 is a characteristic diagram showing a time change between the temperature near the head and the temperature near the leg obtained by the second neural network according to the embodiment of the present invention and the measured value thereof.

FIG. 11 is a subroutine flowchart showing a procedure of an evaluation process according to the embodiment of the present invention.

FIG. 12 is a subroutine flowchart showing a processing procedure for executing a program in FIG. 11;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Input part 2 ... Central processing part 3 ... Display part 4 ... Calculation control part 5 ... Selection / crossover processing part 6 ... Mutation processing part 7 ... Evaluation calculation part 8 ... Simulation part 31 ... First neural network 32 ... No. 2 neural networks

Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat II (Reference) G06F 9/44 G06N 3/12 G06N 3/12 G06F 9/06 620A (72) Inventor Einobu Ito Konan, Osato-gun, Saitama 39, Higashihara, Chiyo-cho, Machio-machi Inside of Xexel Valeo Climate Control (72) Inventor Keiichi Watanuki 4-12-27-215, Tsukakoshi, Warabi-shi, Saitama Terms (reference) 3L060 AA06 AA08 CC02 CC03 CC04 CC05 DD08 EE05 EE07 5B076 DD09 5H215 BB01 BB05 JJ02 JJ13 JJ14

Claims (10)

[Claims] 1. An automatic program generation method for automatically generating a control program for controlling the operation of an automatic control device, comprising: generating an initial population of chromosomes coded in a predetermined coding format, based on a predetermined evaluation function. The fitness of each chromosome constituting the initial population is calculated, and it is determined whether there is a chromosome satisfying the predetermined fitness.It is determined that there is a chromosome satisfying the predetermined fitness. In such a case, the program obtained by decoding the chromosome is used as the control program to be obtained.On the other hand, when it is determined that there is no chromosome satisfying the predetermined fitness, the chromosome population not satisfying the predetermined fitness is determined On the other hand, selection processing, crossover processing, and mutation processing are sequentially performed, and thereafter, fitness is calculated based on a predetermined evaluation function, and chromosomes satisfying the predetermined fitness are calculated. The determination of noness and the selection process for the chromosome population, the crossover process and the mutation process are repeatedly performed until a chromosome satisfying the predetermined fitness is obtained.If a chromosome satisfying the predetermined fitness is obtained, ,
A program obtained by decoding the chromosome is a control program to be obtained. The selection process is to randomly select two chromosomes from a chromosome group under a first predetermined condition. The two chromosomes selected by the selection process are crossed with each other under a second predetermined condition to generate two new chromosomes, and the mutation process includes the chromosome subjected to the cross process. Randomly select a chromosome at a predetermined ratio from among the chromosomes, randomly select one gene from the selected chromosome, rewrite the contents of the gene and mutate the chromosome Characteristic program automatic generation method. 2. A plurality of genes constituting a chromosome are each composed of one function fragment and a plurality of element fragments, and the function fragment is a portion in which an element for determining a program character of the entire gene is set. Therefore, the element is a program language used in the control program,
The plurality of element fragments are selected from a plurality of preselected ones, and the plurality of element fragments are provided corresponding to the sum of the number of actuators and the number of sensors of the automatic control device that is the object of the control program, and the operation of the corresponding actuator 2. The automatic program generation method according to claim 1, wherein an amount or a value of a sensor is set. 3. The element fragment is divided into six sections, a first section having an A value, a second section having a B value, a third section having a C value, and a fourth section having a C value. The D value is set in the section, the E value is set in the fifth section, and the F value is set in the sixth section. The A value takes either 0 or 1, If 0, it means that the element fragment determines the value of the sensor, and if it is 1, it means that the element fragment determines the operation amount of the actuator. B value is the name of the actuator or sensor. The C value is set to either use or unused, the D value is set to a logical operator, the E value is set to a comparison operator, and the F value is the actuator operation amount or sensor. 3. The automatic program generation method according to claim 2, wherein the value is set. 4. When the control program is for driving an air conditioner of an automotive air conditioner, the predetermined evaluation function is a target temperature in the vehicle interior set by the control program, and the target temperature of the vehicle interior by the control program. 4. The automatic program generation method according to claim 3, wherein the fitness is set to be higher as the deviation from the estimated temperature in the vehicle compartment when the operation of (1) is controlled becomes smaller. 5. The evaluation function is: fitness = ｛Σ (T_target−
T_cab)²/ 2)｝ ^-1And the above T_targetIs the cabin set by the control program
Is the target temperature within_cabIs the interior of the cabin
A representative temperature, and in the equation for calculating the fitness, the target temperature and the chamber
The sum of the deviation from the representative temperature in the
The target temperature and the indoor representative temperature obtained at regular time intervals
It is the sum of the deviation from the temperature, the temperature near the head for obtaining the room representative temperature and
The temperature near the leg is determined by decoding the chromosome to be evaluated.
The control program obtained by the
The neural network during the predetermined time
Network and operate the neural network at the predetermined time interval.
The neural network is the first neural network.
Network and the second neural network.
All neural networks are used for automotive air conditioning.
Blower rotation speed, evaporator temperature,
Door opening, heater core temperature, indoor representative temperature and outside air
Temperature as input data, the first neural network comprises an evaporator temperature
Output an estimated value of the temperature and an estimated value of the heater core temperature,
The second neural network estimates the temperature near the head
And outputs an estimated value of the temperature near the leg and the estimated value of the evaporator temperature and the heater core temperature.
Is estimated by the first and second neural networks.
Returned to the input as input data, the temperature near the head
The average of the estimated value and the estimated value of the temperature near the leg is determined by the
First and second neural networks as representative temperatures
5. The process according to claim 4, wherein the feedback is made to the input of
Automatic program generation method. 6. An automatic program generation device for automatically generating a control program for controlling the operation of an automatic control device, comprising: input means for inputting data and commands; and an operation start command input by the input means. By
Calculating a chromosome constituting the initial population in a predetermined coding format, and performing a predetermined operation and operation control; selecting a chromosome from a plurality of chromosomes under a first predetermined condition; Selecting crossovers for exchanging the contents of the selected chromosomes under a second predetermined condition.
Crossover processing means, mutation processing means for performing mutation for arbitrarily rewriting a part of the gene of the chromosome processed by the selection / crossover processing means, and fitness of the chromosome to be evaluated are predetermined. Evaluation means for calculating by a given evaluation function, the arithmetic control means, after the initial population of chromosomes is generated, the chromosome is evaluated by the evaluation means, the evaluation means to calculate the fitness, the calculation It is determined whether the obtained fitness satisfies a predetermined value, and when it is determined that the predetermined fitness is obtained, a control program for obtaining a program obtained by decoding the chromosome is used. If it is determined that the fitness of No. is not obtained, the processing by the selection / crossover processing means, the processing by the mutation processing means, and the processing by the evaluation means are considered It is repeated until a chromosome with a certain fitness is obtained, and when it is determined that the predetermined fitness is obtained, a control program for obtaining a program obtained by decoding the chromosome is automatically generated. apparatus. 7. The coding format of a chromosome is such that a plurality of genes constituting a chromosome are each composed of one function fragment and a plurality of element fragments, and the function fragment is a program of the entire gene. This is the part where the element that determines the personality is set, and the element is one of the programming languages used for the control program.
The plurality of element fragments are selected from a plurality of preselected ones, and the plurality of element fragments are provided corresponding to the sum of the number of actuators and the number of sensors of the automatic control device that is the object of the control program, and the operation of the corresponding actuator 7. The automatic program generation device according to claim 6, wherein an amount or a value of a sensor is set. 8. An element fragment is divided into six sections, the first section having an A value, the second section having a B value, the third section having a C value, and the fourth section having a C value. The D value is set in the section, the E value is set in the fifth section, and the F value is set in the sixth section. The A value takes either 0 or 1, If 0, it means that the element fragment determines the value of the sensor, and if it is 1, it means that the element fragment determines the operation amount of the actuator. B value is the name of the actuator or sensor. The C value is set to either use or unused, the D value is set to a logical operator, the E value is set to a comparison operator, and the F value is the actuator operation amount or sensor. 8. The automatic program generation device according to claim 7, wherein a value is set. 9. When the control program is for driving an air conditioner for an air conditioner for an automobile, the predetermined evaluation function is a target temperature in a vehicle compartment set by the control program, and the air conditioner for the automobile based on the control program. 9. The automatic program generation device according to claim 8, wherein the fitness is set so as to be higher as the deviation from the estimated temperature in the vehicle compartment when the operation is controlled becomes smaller. 10. A simulation means for simulating execution of a control program is provided, and the evaluation function is: fitness = ｛Σ (T _target −T _cab ) ² /
2)｝ ⁻¹ , wherein T _target is a target temperature in the passenger compartment set by the control program, and T _cab is T _cab = (T _head +
T _foot) a room representative temperature of the vehicle interior required as / 2, T _head is a head near the temperature of the vehicle interior, wherein the T _foot is legs near the temperature of the vehicle interior, wherein In the equation for calculating the fitness, the part of the sum of the deviation between the target temperature and the indoor representative temperature is a sum of the deviation between the target temperature and the indoor representative temperature obtained at predetermined time intervals during a predetermined time. The head vicinity temperature and the leg vicinity temperature for obtaining the room representative temperature are executed for a predetermined time by the simulation means by executing a control program obtained by decoding a chromosome to be evaluated. The simulation means obtains an estimated temperature obtained at a predetermined time interval, and the simulation means has first and second neural networks, and the control program Is applied to the first and second neural networks to perform a simulation. Each of the first and second neural networks includes a blower rotation speed, an evaporator temperature, and a mix in a vehicle air conditioner. A door opening, a heater core temperature, a room representative temperature, and an outside air temperature as input data, the first neural network outputs an evaporator temperature estimated value and a heater core temperature estimated value, and the second neural network outputs The estimated value of the temperature near the head and the estimated value of the temperature near the leg are output. The estimated value of the evaporator temperature and the estimated value of the heater core temperature are returned to the input as input data of the first and second neural networks. Of the estimated value of the temperature near the head and the estimated value of the temperature near the legs. Value, automatic program generating apparatus according to claim 9, wherein the fed back to the input of the first and second neural networks as an indoor representative temperature.