JP2004178574A - Control parameter optimization method, control parameter optimization device, and control apparatus optimization program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a control parameter optimization method, a control parameter optimization device and a control parameter optimization program, which effectively perform optimization under an actual operation environment and dynamically changing environment, by providing a plurality of individuals corresponding to an environment in a condition where the environment discretely changes, and by including an environment preliminary assumption part for preliminarily assuming an environment, and an environment retroactive assumption part for assuming an actual environment after evaluation. <P>SOLUTION: A sailing control apparatus 3 includes a cruising control section 30 and a parameter optimization device 31. The parameter optimization device 31 includes an environment preliminary assumption part 31a, a genetic operation part 31b, an individual group set 31c, and an environment retroactive assumption part 31d. A genetic operation is applied to an individual of the individual group corresponding to the preliminary environment assumed by the environment preliminary assumption part 31a. A generation change of the individuals is made based on a comparison between the retroactive environment assumed by the environment retroactive assumption part 31d and the preliminary environment. <P>COPYRIGHT: (C)2004,JPO

Description

  The present invention relates to control of a control device, and more particularly, to a control parameter optimization method suitable for optimizing a control parameter of a control device in an environment where control conditions change dynamically.

  Conventionally, various adaptive control methods have been proposed as methods for controlling a control target whose characteristics are unknown in advance. Among them, a method of optimizing control parameters of a control device in an actual operating environment using an evolutionary calculation method. Does not require a teacher signal because only evaluation values need to be used, and because of its excellent global search capability, it can optimize unexpected control targets, and its flexible framework can be applied to various problems. Is possible. For this reason, research on evolutionary computation methods in recent years has shifted from basic research using conventional test functions to simulation of real systems and applied research using the same.

  However, when using online optimization of control parameters using an evolutionary calculation method in a real system, the observed values and evaluation values include noise, and the number of evaluations required for optimization is limited. There are problems such as things. In particular, if the landscape of fitness changes due to a change in the environment, the fitness of the entire population changes, and the problem that optimization is not successful makes application to real systems difficult. When the environment is considered to change discretely to some extent, a plurality of individual groups corresponding to the environment are prepared and each of them is evolved, so that optimization according to the change in the environment can be performed. However, there is also a problem that if it is not known in advance in which environment the individual will be evaluated before evaluating the individual, it is impossible to know which individual to select.

  Specifically, for example, when considering the constant speed cruise control of a small boat by an outboard motor, the hull (size, bottom shape, application, etc.) attached to the outboard motor differs greatly, and at the design stage Since it is almost impossible to predict what kind of control target, there is a problem that control parameters for effectively performing constant speed cruise control cannot be determined in advance. To solve this problem, online optimization of control parameters using a genetic algorithm is effective. However, even if the hull is uniquely determined, its characteristics greatly differ depending on the use environment. Specifically, the resistance received from the water surface of the hull changes due to natural factors such as changes in weather and sea conditions, and artificial factors such as steering operation and trim operation, and the controllability of speed control greatly differs. This phenomenon, that is, the change in the environment, occurs in a cycle shorter than the period for optimizing a certain hull under the actual operating environment, so the optimal solution and the landscape of the fitness dynamically change during the optimization. This causes optimization to fail.

  Therefore, the present invention has been made by focusing on the unsolved problem of the conventional technology, and prepares a plurality of individual groups corresponding to the environment under conditions where the environment changes discretely. , An environment pre-estimation unit that pre-estimates the environment and an environmental post-estimation unit that estimates the actual environment after the evaluation make it possible to effectively optimize the environment that changes dynamically in the actual operation It is an object of the present invention to provide a control parameter optimization method, a control parameter optimization device, and a control parameter optimization program that can be implemented.

In order to achieve the above object, a method for optimizing a control parameter according to claim 1 of the present invention provides a method for optimizing a control parameter of a control device used in an environment in which a control condition changes dynamically by using a genetic algorithm. An optimization method that uses and optimizes,
For each of the changing environment, to generate a population of the genetic algorithm according to the control parameters,
Based on predetermined information obtained under the current environment, the environment at the time of the next genetic operation by the genetic algorithm is estimated in advance from the environment that is in correspondence with the population,
From the population, select a population corresponding to the environment of the estimation result,
For the selected population, as the genetic manipulation,
Select a specific individual from the population corresponding to the environment estimated in advance,
A new individual is generated by performing a crossover process using a predetermined crossover method on the specific individual,
Perform evaluation of these individuals based on the evaluation value obtained from the control result using the new individual and the specific individual,
It is characterized in that the individual is updated after the evaluation based on the evaluation result.

  In other words, for each environment in which the control condition changes dynamically, a plurality of populations corresponding to each environment are prepared, and the environment to which the genetic algorithm will be applied next is estimated in advance, and then the estimated environment By selecting a population corresponding to the above, and performing a genetic operation on the individuals in this population, it becomes possible to optimize the population according to the environment, and to use the sampling data of the control results It becomes possible to use it.

  Here, the genetic algorithm (hereinafter referred to as GA) assigns random initial values to each individual, arranges them in a search space, applies genetic operations called crossover and mutation for each generation, and evaluates each individual. Propagation and selection of individuals according to the value yields a set of individuals for the next generation. An object is to asymptotically approach an optimal solution by repeating such generation alternation. Hereinafter, crossover, mutation, and selection, which are genetic operations, will be described.

  Crossover is an operation in which at least two individuals are set as parents, and one or more individual descendants are generated by replacing part of the individual information of the individual parent. By combining the good part of the individual information of a certain individual with the good part of the individual information of another individual, it is expected that an individual with a higher evaluation value will be obtained. For example, in a case where two individuals are generated as descendants using two individuals as parents, the individual information of one parent individual is set to “000110” and the individual information of the other parent individual is set to “110111”. Are obtained, the individual having the individual information of "000111" and the individual having the individual information of "110110" are obtained as the descendants.

Mutation is an operation of changing a specific portion of individual information of an individual with a predetermined probability, and increases diversity within an individual group. Specifically, this is an operation of inverting a specific bit of the individual information. For example, the individual information of a certain individual is set to “000111”, and a mutation is caused at the third position, whereby the individual information of “001111” is obtained. Obtain an individual with
The selection is an operation for leaving a better individual from the population to the next generation according to the evaluation value of the individual. In a selection method called roulette selection, each individual is selected with a probability proportional to the evaluation value. For example, in a certain generation, the evaluation values of individuals having individual information of “000000”, “111011”, “110111”, and “010111” are “8”, “4”, “2”, and “2”, respectively. I do. The probability that each individual is selected is “8/16”, “4/16”, “2/16”, “2/16”. Therefore, on average, in the next generation, the number of individuals having the individual information of “000000” increases to two, the number of individuals having the individual information of “111011” remains one, and the individual having the individual information of “110111” remains Alternatively, an individual group having any of the individuals having the individual information “010111” is obtained.

According to a second aspect of the present invention, there is provided a method for optimizing a control parameter, wherein the control parameter of a control device used in an environment where a control condition is dynamically changed is optimized using a genetic algorithm. An optimization method,
For each of the changing environment, to generate a population of the genetic algorithm according to the control parameters,
Based on predetermined information obtained under the current environment, the environment at the time of the next genetic operation by the genetic algorithm is estimated in advance from the environment that is in correspondence with the population,
From the population, select a population corresponding to the environment of the estimation result,
For the selected population, as the genetic manipulation,
Select a specific individual from the population estimated in advance,
A new individual is generated by performing a crossover process using a predetermined crossover method on the specific individual,
Perform evaluation of these individuals based on the evaluation value obtained from the control result using the new individual and the specific individual,
Based on the predetermined information obtained from the environment after the evaluation, the environment at this time is estimated from the environment corresponding to the population,
Compare the environment estimated after the evaluation and the environment estimated in advance,
An update process of the individual after the evaluation is performed based on the comparison result.

  In other words, for each environment in which the control condition changes dynamically, a plurality of populations corresponding to each environment are prepared, and the environment to which the genetic algorithm will be applied next is estimated in advance, and then the estimated environment Select a population corresponding to, perform a genetic operation on the individuals in this population using a genetic algorithm, and further estimate the environment after the individual evaluation in this genetic operation, and estimate this environment in advance. In the case of performing genetic manipulation on individuals in a population group corresponding to an environment different from the actual environment, appropriate Can be performed, and the accuracy of optimization can be improved.

According to a third aspect of the present invention, in the control parameter optimizing method according to the second aspect, in the comparison of the estimated environment, the environment estimated in advance and the environment estimated after the evaluation are different. When the environment is different, the individual after the evaluation is replaced with a predetermined individual of an individual group corresponding to the environment estimated after the evaluation.
In other words, for each environment in which the control condition changes dynamically, a plurality of populations corresponding to each environment are prepared, and the environment to which the genetic algorithm will be applied next is estimated in advance, and then the estimated environment Select a population corresponding to, perform a genetic operation on the individuals in this population using a genetic algorithm, and further estimate the environment after the individual evaluation in this genetic operation, and estimate this environment in advance. Compared to the environment, when the two were different environments according to the comparison result, the individual after the evaluation was replaced with a predetermined individual of the population corresponding to the environment also estimated after the evaluation, Even if genetic manipulation is performed on individuals in a population that corresponds to an environment different from the actual environment, the results are used without discarding, so the possibility of evolution adapted to multiple environments increases, And Also enabled the use of the sampling data of the control results.

According to a fourth aspect of the present invention, in the control parameter optimizing method according to the second aspect, in the comparison of the estimated environment, the environment estimated in advance and the environment estimated after the evaluation are different. When the environment was different, the individual was selected from the populations corresponding to these environments and the genetic operation was performed, and the individual after the genetic operation was changed to the environment estimated after the evaluation. It is characterized by being replaced with a predetermined individual of a corresponding individual group.
That is, at the transition of the environment, there is a high possibility that the environment estimated in advance is wrong, so that it is possible to obtain a stable evaluation by generating a population including intermediate individuals.

According to a fifth aspect of the present invention, in the control parameter optimizing method according to any one of the second to fourth aspects, the specific individual includes a best individual.
That is, since the best individual is included in the specific individual to be selected, it is possible to facilitate appropriate evolution and increase the speed of optimization.
According to a sixth aspect of the present invention, in the control parameter optimizing method according to any one of the first to fifth aspects, it is determined whether or not an individual of an individual population corresponding to the environment has been evaluated. When it is determined that the evaluation has been completed, the evaluation of the individual is omitted, the previously obtained evaluation value is held, and the updating process is performed using the held evaluation value.

That is, since the individual evaluated once is not re-evaluated, the number of evaluations can be reduced, and the speed of optimization can be increased.
According to a seventh aspect of the present invention, in the control parameter optimizing method according to the sixth aspect, a flag indicating the fact is set for an individual determined to have been evaluated, and an individual group corresponding to the environment is set. The number of times the individual has survived during the updating process is recorded, and when the recorded number exceeds a predetermined number, the set flag is released.

In other words, individuals who have survived a large number of times during the update process have a high possibility of having obtained an evaluation value higher than the actual evaluation value. By performing the re-evaluation, it is possible to more reliably select individuals to be selected.
Specifically, if the posterior estimation is incorrect without re-evaluation, returning the evaluated individual to a population corresponding to the wrong environment adversely affects the subsequent optimization. This is because the evaluation value obtained from each environment is different even for the same individual. For example, after obtaining a good evaluation value in environment 1, the posterior estimation is erroneously returned to the population corresponding to environment 2, In the case where the best individual having a remarkably good evaluation value is surpassed in the environment 2, there is a problem that the individual is not culled indefinitely because the reevaluation is not performed even if the individual should be originally culled in the environment 2. . Furthermore, when the best individual is always selected as a parent individual, in environment 2, a child individual is generated around an individual that should be originally selected, which causes a problem that search efficiency is reduced. Therefore, as described above, the individual that has survived a predetermined number of times or more is re-evaluated, and the correct evaluation is performed again.

According to an eighth aspect of the present invention, in the control parameter optimizing method according to the sixth or seventh aspect, a current evaluation value of each individual in the plurality of individuals and a past evaluation value in the plurality of individuals are used. The present invention is characterized in that the updating process is performed after comparing with the best evaluation value, re-evaluating the individual whose current evaluation value is superior to the past evaluation value, and then re-evaluating the individual.
In other words, an individual with an evaluation value better than the best individual up to the previous update process may have obtained an evaluation value higher than the actual evaluation value. By doing so, it is possible not to return the individuals that should be originally selected to the population.

According to a ninth aspect of the present invention, in the control parameter optimizing method according to any one of the first to eighth aspects, the control device is an outboard motor provided on a ship, and Is determined based on at least one of the following information: the traveling speed of the ship, the engine speed, the throttle operation amount, the steering operation amount, and the trim angle.
According to a tenth aspect of the present invention, in the method for optimizing the control parameters according to any one of the first to ninth aspects, the traveling speed, the engine speed, the throttle operation amount, and the steering operation amount of the ship are provided. And the trimming angle is estimated based on at least one of the trim angle and the trim angle.

According to an eleventh aspect of the present invention, in the method for optimizing a control parameter according to any one of the second to tenth aspects, the traveling speed, the engine speed, the throttle operation amount, and the steering operation amount of the ship are provided. And the environment after the evaluation is estimated on the basis of at least one of the information and the trim angle.
A control parameter optimizing device according to a twelfth aspect of the present invention optimizes a control parameter of a control device used in an environment where a control condition changes dynamically using a genetic algorithm. An optimization device,
Corresponding to each of the changing environment, a population generating means for generating a population of the genetic algorithm according to the control parameter,
Based on predetermined information obtained under the current environment, environment prior estimation means for preliminarily estimating an environment at the time of the next genetic operation by the genetic algorithm from among the environments in correspondence with the population,
Population selecting means for selecting a population corresponding to the environment of the estimation result from the population,
Genetic manipulation means for performing the genetic manipulation on the selected population,
The genetic manipulation means,
A specific individual selection unit for selecting a specific individual from a population corresponding to the environment estimated in advance,
A crossover processing unit that generates a new individual by performing a crossover process using a predetermined crossover method on the specific individual;
An individual evaluation unit that evaluates these new individuals and the individual based on an evaluation value obtained from a control result using the specific individual,
And an individual updating unit that updates the individual after the evaluation based on the evaluation result.

  With such a configuration, it is possible to generate a population of the genetic algorithm according to the control parameter, which corresponds to each environment changed by the population generation unit, and the current environment estimation unit estimates the current population. It is possible to preliminarily estimate the environment at the time of the next genetic operation by the genetic algorithm from the environment of the corresponding population based on the predetermined information obtained under the environment. It is possible to select a population corresponding to the environment of the estimation result from the population, and to perform the genetic operation on the population selected by the genetic operation means, The specific operation means can select a specific individual from a population corresponding to the environment estimated in advance by the specific individual selection unit as a genetic operation. Therefore, it is possible to generate a new individual by performing a crossover process using a predetermined crossover method on the selected specific individual by the crossover processing unit, and the new individual and the It is possible to evaluate these individuals based on the evaluation value obtained from the control result using the specific individual, and it is possible to perform the individual updating process after the evaluation based on the evaluation result by the individual updating unit. .

Here, the present invention is an apparatus for realizing the control parameter optimizing method according to the first aspect, and its effects are duplicated, so that the description is omitted.
A control parameter optimizing device according to a thirteenth aspect of the present invention optimizes a control parameter of a control device used in an environment where a control condition dynamically changes using a genetic algorithm. An optimization device,
Corresponding to each of the changing environment, a population generating means for generating a population of the genetic algorithm according to the control parameter,
Based on predetermined information obtained under the current environment, environment prior estimation means for preliminarily estimating an environment at the time of the next genetic operation by the genetic algorithm from among the environments in correspondence with the population,
Population selecting means for selecting a population corresponding to the environment of the estimation result from the population,
Genetic manipulation means for performing the genetic manipulation on the selected population,
The genetic manipulation means,
A specific individual selection unit for selecting a specific individual from a population corresponding to the environment estimated in advance,
A crossover processing unit that generates a new individual by performing a crossover process using a predetermined crossover method on the specific individual;
An individual evaluation unit that evaluates these new individuals and the individual based on an evaluation value obtained from a control result using the specific individual,
Based on predetermined information obtained from the environment after the evaluation, an environment posterior estimating unit for estimating the environment at this time from a certain environment of the corresponding population,
An environment comparison unit that compares the environment estimated after the evaluation with the environment estimated in advance,
And an individual updating unit that updates the individual after the evaluation based on the comparison result.

  With such a configuration, it is possible to generate a population of the genetic algorithm according to the control parameter, which corresponds to each environment changed by the population generation unit, and the current environment estimation unit estimates the current population. It is possible to preliminarily estimate the environment at the time of the next genetic operation by the genetic algorithm from the environment of the corresponding population based on the predetermined information obtained under the environment. It is possible to select a population corresponding to the environment of the estimation result from the population, and to perform the genetic operation on the population selected by the genetic operation means, The specific operation means can select a specific individual from a population corresponding to the environment estimated in advance by the specific individual selection unit as a genetic operation. Therefore, it is possible to generate a new individual by performing a crossover process using a predetermined crossover method on the selected specific individual by the crossover processing unit, and the new individual and the It is possible to evaluate these individuals based on the evaluation value obtained from the control result using the specific individual, based on predetermined information obtained from the environment after the evaluation by the environmental posterior estimating unit, the environment at this time Can be estimated from a certain environment of the corresponding population, and it is possible to compare the environment estimated after evaluation by the environment comparison unit with the environment estimated in advance, and the individual update unit Thus, it is possible to update the individual after the evaluation based on the comparison result.

Here, the present invention is an apparatus for realizing the control parameter optimizing method according to the second aspect, and its effects are duplicated, so that the description is omitted.
According to a fourteenth aspect of the present invention, in the control parameter optimizing device according to the thirteenth aspect, the comparison result in the environment comparison unit is such that the environment estimated in advance and the environment estimated after the evaluation are different. When the environment is different, the individual updating unit replaces the individual after the evaluation with a predetermined individual of an individual group corresponding to the environment estimated after the evaluation.

That is, when the comparison result in the environment comparison unit is an environment in which the environment estimated beforehand and the environment estimated after the evaluation are different, the individual updating unit estimates the individual after the evaluation after the evaluation. It is possible to replace a predetermined individual of the population corresponding to the environment.
Here, the present invention is an apparatus for realizing the control parameter optimizing method according to the third aspect, and the description thereof is omitted because the effects are duplicated.

  According to a fifteenth aspect of the present invention, in the control parameter optimizing device according to the thirteenth aspect, the comparison result in the environment comparing unit is such that the environment estimated in advance and the environment estimated after the evaluation are different. When the environment is different, the genetic manipulation means selects a specific individual from a population corresponding to these environments, performs the genetic manipulation, and sets the individual after the genetic manipulation to the It is characterized in that a predetermined individual of an individual group corresponding to the environment estimated after the evaluation is replaced.

That is, when the comparison result in the environment comparison unit is an environment in which the environment estimated in advance and the environment estimated after the evaluation are different from each other, the , It is possible to perform the genetic operation by selecting a specific individual from the group, and replace the individual after the genetic operation with a predetermined individual of an individual group corresponding to the environment estimated after the evaluation.
Here, the present invention is an apparatus for realizing the control parameter optimizing method according to claim 4, and its effect is duplicated, so that the description is omitted.
According to a sixteenth aspect of the present invention, in the control parameter optimizing device according to any one of the thirteenth to fifteenth aspects, the specific individual includes a best individual.

Here, the present invention is an apparatus for realizing the control parameter optimizing method according to claim 5, and its effect is duplicated, so that the description is omitted.
According to a seventeenth aspect of the present invention, in the control parameter optimizing device according to any one of the thirteenth to sixteenth aspects, the genetic operation means has evaluated an individual of an individual group corresponding to the environment. It is determined whether or not the evaluation has been completed, and if it is determined that the evaluation has been completed, the evaluation value of the individual is omitted, the previously obtained evaluation value is retained, and the updating process is performed using the retained evaluation value. It is characterized by.

Here, the present invention is an apparatus for realizing the control parameter optimizing method according to claim 6, and its effect is duplicated, so that the description is omitted.
According to an eighteenth aspect of the present invention, in the control parameter optimizing device according to the seventeenth aspect, the genetic operation means sets a flag indicating that the individual is determined to have been evaluated, The number of surviving individuals in the individual group corresponding to the environment during the updating process is recorded, and when the recorded number exceeds a predetermined number, the set flag is released.

Here, the present invention is an apparatus for realizing the control parameter optimizing method according to claim 7, and its effect is duplicated, so that the description is omitted.
According to a nineteenth aspect of the present invention, in the control parameter optimizing device according to the seventeenth or eighteenth aspect, the genetic operation means includes: a current evaluation value of each individual in the plurality of individuals; Compared with the best evaluation value in the past in a plurality of populations, the current evaluation value is characterized by performing the update process after re-evaluating individuals that are better than the past evaluation value I have.

Here, the present invention is an apparatus for realizing the control parameter optimizing method according to claim 8, and its effect is duplicated, so that the description is omitted.
According to a twentieth aspect of the present invention, in the control parameter optimizing device according to any one of the twelfth to nineteenth aspects, the control device is an outboard motor provided in a ship, and Is determined based on at least one of the following information: the traveling speed of the ship, the engine speed, the throttle operation amount, the steering operation amount, and the trim angle.

That is, the control device is an outboard motor provided in the ship, and the changing environment is determined based on at least one of information on the running speed, the engine speed, the throttle operation amount, the steering operation amount, and the trim angle of the ship. Is done.
According to a twenty-first aspect of the present invention, in the control parameter optimizing device according to any one of the twelfth to twentieth aspects, the environment pre-estimating means includes a cruising speed of a ship, an engine speed, The prior environment is estimated based on at least one of a throttle operation amount, a steering operation amount, and a trim angle.

That is, the environment in advance can be estimated by the environment pre-estimating means based on at least one of the following information: the traveling speed of the ship, the engine speed, the throttle operation amount, the steering operation amount, and the trim angle. .
According to a twenty-second aspect of the present invention, in the control parameter optimizing device according to any one of the thirteenth to twenty-first aspects, the environmental posterior estimating unit includes: The environment after the evaluation is estimated based on at least one of a throttle operation amount, a steering operation amount, and a trim angle.

That is, the environment after the evaluation can be estimated by the environment posterior estimating unit based on at least one of the following information: the traveling speed of the ship, the engine speed, the throttle operation amount, the steering operation amount, and the trim angle. .
A control parameter optimization program according to claim 23 of the present invention optimizes a control parameter of a control device used in an environment in which control conditions change dynamically using a genetic algorithm. A computer executable program,
For each of the changing environment, to generate a population of the genetic algorithm according to the control parameters,
Based on predetermined information obtained under the current environment, the environment at the time of the next genetic operation by the genetic algorithm is estimated in advance from the environment that is in correspondence with the population,
From the population, select a population corresponding to the environment of the estimation result,
For the selected population, as the genetic manipulation,
Select a specific individual from the population corresponding to the environment estimated in advance,
A new individual is generated by performing a crossover process using a predetermined crossover method on the specific individual,
Perform evaluation of these individuals based on the evaluation value obtained from the control result using the new individual and the specific individual,
It is characterized in that the individual is updated after the evaluation based on the evaluation result.

Here, the present invention is a program for realizing the control parameter optimizing method according to the first aspect, and its effects are duplicated, so that the description is omitted.
A control parameter optimizing program according to claim 24 of the present invention optimizes a control parameter of a control device used in an environment in which control conditions change dynamically using a genetic algorithm. A computer executable program,
For each of the changing environment, to generate a population of the genetic algorithm according to the control parameters,
Based on predetermined information obtained under the current environment, the environment at the time of the next genetic operation by the genetic algorithm is estimated in advance from the environment that is in correspondence with the population,
From the population, select a population corresponding to the environment of the estimation result,
For the selected population, as the genetic manipulation,
Select a specific individual from the population corresponding to the environment estimated in advance,
A new individual is generated by performing a crossover process using a predetermined crossover method on the specific individual,
Perform evaluation of these individuals based on the evaluation value obtained from the control result using the new individual and the specific individual,
Based on the predetermined information obtained from the environment after the evaluation, the environment at this time is estimated from the environment of the corresponding population,
Compare the environment estimated after the evaluation and the environment estimated in advance,
An update process of the individual after the evaluation is performed based on the comparison result.
Here, the present invention is a program for realizing the control parameter optimizing method according to the second aspect, and its effect is duplicated, so that the description is omitted.

  As described above, according to the control parameter optimizing method according to claim 1 of the present invention, for each environment in which the control condition dynamically changes, a plurality of individual groups corresponding to each environment are prepared. Since, after estimating the environment to which the genetic algorithm is to be applied in advance, a population corresponding to the estimated environment is selected, and genetic operations are performed on individuals in this population. This makes it possible to optimize a group of individuals suitable for the environment, and to effectively use the sampling data of the control result.

  According to the control parameter optimizing method of the present invention, for each environment in which the control condition changes dynamically, a plurality of individuals corresponding to each environment are prepared, and then the genetic algorithm is used. After estimating the environment to be applied in advance, select a population corresponding to the estimated environment, perform a genetic operation on the individuals in this population by a genetic algorithm, Since the environment after the evaluation is estimated, the environment is compared with the environment estimated in advance, and the individual after the evaluation is updated based on the comparison result. When the genetic manipulation is performed on the individual, the appropriate treatment can be performed, and the precision of the optimization can be improved.

  According to the control parameter optimizing method of the present invention, for each environment in which the control condition changes dynamically, a plurality of individuals corresponding to each environment are prepared, and After estimating the environment to be applied in advance, select a population corresponding to the estimated environment, perform a genetic operation on the individuals in this population by a genetic algorithm, The environment after the evaluation is estimated, and this environment is compared with the environment estimated in advance. When the two environments are different according to the comparison result, the individual after the evaluation is replaced with the environment estimated after the evaluation. Replaced with a specific individual of the corresponding population, so that even if genetic operations were performed on individuals of the population corresponding to an environment different from the actual environment, the results were used without discarding In the possibility of evolution adapted to multiple environments increases, and, also the effective use of the sampling data of the control results.

Further, according to the control parameter optimizing method of the present invention, at the transition of the environment, there is a high possibility that the environment estimated in advance is wrong, so that the population of intermediate individuals is generated. Therefore, a stable evaluation can be obtained.
According to the control parameter optimizing method described in claim 5, the best individual is included in the specific individual to be selected, so that appropriate evolution can be facilitated and the speed of optimization can be increased. It becomes.
According to the control parameter optimizing method of the present invention, since the individual evaluated once is not re-evaluated, the number of evaluations can be reduced, and the speed of optimization can be increased. Become.

According to the control parameter optimizing method of the present invention, individuals that have survived a predetermined number of times or more during the update process are re-evaluated by invalidating the flag indicating that evaluation has been completed. Can be more reliably eliminated.
According to the control parameter optimizing method of the present invention, an individual having an evaluation value better than that of the best individual up to the previous update process is re-evaluated. It is possible not to return the power individual to the population.
In the control parameter optimizing method according to the ninth to eleventh aspects, the control device is an outboard motor provided in the marine vessel.

Here, the control parameter optimizing devices according to claims 12 to 22 are devices for realizing the control parameter optimizing methods according to claims 1 to 11, respectively, and their effects are duplicated. Is omitted.
Further, the control parameter optimizing programs according to claims 23 and 24 are programs for realizing the control parameter optimizing methods according to claim 1 and claim 2, respectively. Omitted.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. 1 to 15 are diagrams showing an embodiment of a navigation control system including a parameter optimizing device according to the present invention.
First, the configuration of a navigation control system according to the present invention will be described with reference to FIG. FIG. 1 is a block diagram showing a configuration of a cruise control system according to the present invention.
The cruise control system 1 has a configuration including an outboard motor 2 and a cruise control device 3.

The outboard motor 2 includes an electronic throttle valve 2a as a thrust adjusting device.
The cruise control device 3 has a configuration including a constant speed cruise control unit 30 and a parameter optimizing device 31.
The constant speed cruise control unit 30 includes a target engine speed calculation module 30a that calculates a target engine speed based on predetermined input information, and an electronic engine that calculates an electronic throttle valve opening based on the calculated target engine speed. And a throttle valve opening calculation module 30b.

The target engine speed calculation module 30a is configured by a fuzzy inference system, and infers a target engine speed change amount by using a speed deviation (difference between the current speed and the target speed) and an acceleration as inputs. The present target engine speed is output by adding the target engine speed in the immediately preceding calculation process.
The electronic throttle valve opening calculation module 30b is configured by a fuzzy inference system similarly to the target engine speed calculation module 30a, and includes an engine speed deviation (a difference between the current engine speed and the target engine speed), The electronic throttle valve opening change amount is inferred by using the engine rotational speed change amount as an input, and the electronic throttle valve opening degree in the immediately preceding calculation process is added to the inference result, and the result is output as the current electronic throttle valve resolution. is there. Here, the fuzzy table of the fuzzy inference system is designed based on the knowledge of maneuvering by a skilled person, and uses the simplified inference method as the inference method, and the fuzzy rules in this method are represented by real values.

The parameter optimizing device 31 has a configuration including an environment pre-estimation unit 31a, a genetic operation unit 31b, a population set 31c, and an environment post-estimation unit 31d.
The pre-environment estimating unit 31a is configured to input predetermined information that dynamically changes during the voyage of a ship as an input, and to estimate and output the current marine running state (environment) of the ship in advance.
The genetic operation unit 31b employs a GA, encodes a normalization coefficient of a fuzzy membership function of the antecedent part or the consequent part in the electronic throttle valve opening calculation module 30b, generates an individual, and uses the GA. Thus, these normalization coefficients are optimized. Here, the normalization coefficient means a coefficient for adjusting the size of input / output information.

The population set 31c is a set of populations generated in accordance with the dynamically changing cruising state (environment) of the ship. The genetic operation unit 31b selects a population corresponding to the cruising state of the ship estimated by the environment pre-estimation unit 31a from the population set 31c, and applies GA to the individuals in the population. Optimize the normalization factor.
The environmental posterior estimating unit 31d estimates the cruising state (environment) of the vessel at that time by inputting predetermined information that dynamically changes during the cruising of the vessel after performing the genetic operation by the GA. is there.

  Hereinafter, an operation when the present system 1 is applied to constant-speed cruise control of a boat equipped with an outboard motor having an electronic throttle valve device will be described with reference to FIGS. 2 to 8, FIG. 11, and FIG. . FIG. 2 is an explanatory diagram of a fuzzy rule for calculating a target engine speed in the target engine speed calculation module 30a. FIG. 3 is a diagram for calculating an electronic throttle valve solvability in the electronic throttle valve opening calculation module 30b. FIG. 4 is a diagram showing a landscape of a fitness function when a ship makes a turn.

FIG. 5 is a diagram showing the duration of the previous cruising situation with respect to the steering operation, and FIG. 6 is a diagram showing the generation range when generating the initial value of the normalization coefficient as the initial individual of GA. 7 is a diagram showing a unimodal normal distribution crossover method, FIG. 8 is a diagram showing how a single individual is generated from a parent individual using a unimodal normal distribution crossover method, and FIG. FIG. 9 is a diagram showing generational change of individuals based on evaluation with a post-environment.
FIG. 15A is a diagram illustrating a data structure of an individual according to the first embodiment.

Hereinafter, a first embodiment of the present invention will be described.
First, the user sets a target speed. As the target speed, an arbitrary value determined by the user may be input, or the target speed may be selected from a plurality of values prepared at the shipping stage by the manufacturer. After the target speed is set, an initial value of the target engine speed is set based on this speed. For example, when the current speed of the ship is near the target speed, the current engine speed is set as the target engine speed, while when the current speed of the ship is not near the target speed, An initial value of a preset target engine speed is used. Here, the initial value of the target engine speed may be an arbitrary value determined by the user, or may be selected from a plurality of values prepared at the time of shipment by the manufacturer.

  After setting the initial value of the target engine speed, the target engine speed corresponding to the running speed at that time is calculated based on the fuzzy rule shown in FIG. 2 according to the actual running state of the ship. . That is, the inferred values of the speed deviation and the acceleration are obtained from the membership functions shown in FIG. 2A, and the inferred values are applied to the fuzzy rules of FIG. I do. The speed deviation (difference between the target speed and the actual speed) of the membership function is obtained from the detected value of the actual speed, and the acceleration is obtained by calculating the detected speed. As shown in FIG. 2 (a), four values corresponding to the magnitudes on the plus side (PL, PS) and the magnitudes on the minus side (NL, NS) are obtained from the membership function according to the speed deviation and the acceleration. These are weighted to the corresponding four values of the fuzzy rule shown in FIG. 2B to calculate an average value. As a result, the amount of change in the target engine speed is obtained. By adding this change amount to the current target engine speed, a new target engine speed is obtained.

  Then, the target engine speed calculation module 30a outputs the target engine speed to the electronic throttle valve opening calculation module 30b, and the electronic throttle valve opening calculation module 30b outputs the target engine speed to the input target engine speed. The throttle valve opening is calculated based on the throttle valve opening. Here, the electronic throttle valve opening calculation module 30b also calculates the electronic throttle valve opening by a fuzzy inference system, similarly to the target engine speed calculation module 30a. That is, when the target engine speed is obtained from the target engine speed calculation module 30a, an inference value of the engine speed deviation is obtained from the detected value of the engine speed based on the membership function shown in FIG. The amount of change in the engine speed is obtained by calculation from the detected value of the engine speed. That is, similarly to FIG. 2, the amount of change in the electronic throttle valve opening is obtained from the value of the membership function by performing arithmetic processing by weighting the fuzzy rule of FIG. 3B. By adding this change amount to the current electronic throttle valve opening, a new electronic throttle opening is obtained.

When the new electronic throttle valve solvability is calculated in this way, the constant speed cruise control unit 30 controls the electronic throttle valve device 2a so as to match the calculated new electronic throttle valve opening. .
In the present embodiment, the genetic operation unit 31b of the parameter optimizing device 31 encodes the normalization coefficient of the fuzzy membership function of the antecedent or the consequent part in the electronic throttle valve opening calculation module 30b to identify the individual. Generate and optimize these normalization coefficients using GA. Here, in the present embodiment, the fitness is evaluated as the deviation of the current engine speed from the target engine speed within a certain time approaching zero and the variation in the engine speed is smaller in this embodiment. This is performed according to a fitness function (formula (1) shown below) set to be better.

F = Σ (| N′−N | + ΔN) · Δt (1)
Here, in equation (1), N is the engine speed, N 'is the target engine speed within a certain time, ΔN is the number of changes in the engine speed, and Δt is the change time. According to the equation (1), the normalization coefficient of the electronic throttle valve opening calculation module is automatically optimized toward the target engine speed control characteristic, and even when the use environment and the hull change, the optimum coefficient is obtained. The engine speed control characteristic can be obtained.

The problem here is that the optimum normalization coefficient for the engine speed control characteristic varies depending on the environment. Specifically, the adaptability of the engine speed control is determined by the change in the resistance of the hull generated during turning. The point is that the landscape of the function changes, and the position of the optimum normalization coefficient changes accordingly.
As shown in FIG. 4, the landscape of the fitness function when the normalization coefficient of the membership function of the consequent part is changed for each magnitude of the turn in the equation (1) shows that the change in the fitness function is less than the followability in the straight traveling state. It can be seen that if the normalization coefficient is reduced to suppress the deterioration, the fitness improves, while if the turning radius decreases, the resistance received by the hull increases. It can be seen that the larger the normalization coefficient, the better the fitness. In addition, the case where two normalization coefficients of the membership function of the antecedent part are changed is similarly considered.

In the present embodiment, focusing on the above points, the turning state is divided into three environments 40a to 40c in the drawing (straight running state 40a, gentle turning 40b, and sharp turning 40c). respectively produces three populations P _1, P _2, populations set 31c consisting of P ₃ corresponding to the 40a to 40c.
Therefore, as shown in FIG. 15A, the data structure of each individual is composed of the normalization coefficients 1 to 3 respectively corresponding to the above three turning states and the evaluation value of the individual.

Here, as a parameter representing the environmental change, it is assumed that a certain time t velocity V _t and the engine speed N _t ratio propulsive efficiency S _t of, define the following equation (2).
_{S t = V t / N t} ········· (2)
That is, as the turning state transits to the straight traveling state 40a, the gentle turning state 40b, and the sharp turning state 40c, the water contact area of the hull increases, and the engine load changes, so that the engine speed increases even at the same speed. , And the propulsion efficiency changes depending on the turning state. Therefore, in the present embodiment, an optimum engine speed control characteristic is obtained even when the turning state dynamically changes by the environment pre-estimation unit 31a and the environment post-estimation unit 31d described below.

  The pre-environment estimating unit 31a predicts the next environment by inputting the steering operation amount that is a factor of turning. Basically, it is possible to determine the turning situation from the amount of steering operation, but in reality, there is a time delay before the steering operation is reflected in the turning operation. If the steering returns to straight ahead, the change in resistance caused by turning will continue when viewed in terms of propulsion efficiency.If the environment is simply distinguished by the amount of steering operation, the evaluation will be successful due to the sustained change in resistance. There is a problem of not going. Since the duration of the resistance change is related to the turn time, in the present embodiment, the next environment is predicted by estimating the duration of the environment change from the steering operation amount.

Specifically, the duration is calculated according to the following equations (3) to (6).
I ₂ = ΣA _t · Δt ₂ (3)
I ₃ = ΣA _t · Δt ₃ (4)
However, A _t is a steering operation amount, Delta] t ₂ is the time exceeds the threshold value A ₂ of the slow turning operation is A _t, I ₂ is the integrated value, Delta] t ₃ is greater than the threshold value A ₃ of A _t suddenly turning operation During this time, I ₃ is its integral value.

In other words, the integral value I can be calculated by adding between the product of the steering operation amount A _t and the sampling time interval Delta] t A _t exceeds the threshold value A.
Next, the duration of the environmental change can be calculated from the following equations (5) and (6) from the linear approximation equation obtained from the preliminary experiment.
T ₂ = α ₂ · I ₂ + β ₂ (5)
T ₃ = α ₃ · I ₃ + β ₃ (6)
Here, α ₂ , α ₃ and β ₂ , β ₃ are coefficients in the gentle turning operation and the sharp turning operation, respectively, and T ₂ and T ₃ are the durations in the gentle turning operation and the sudden turning operation, respectively.

That is, the duration T can be calculated by adding the coefficient β to the product of the coefficient α and the integral value I obtained by the above equations (3) and (4).
By using the above equations (3) to (6), it is possible to estimate the duration of the environmental change from the steering operation amount. For example, as shown in FIG. Even if it is returned to straight ahead, if it is within the duration of the environmental change, the environment is continuously evaluated as a gentle turning state, and similarly, although not shown, the steering is returned to straight after performing a sharp turning operation for a certain period of time. Also, the evaluation is continuously performed as the environment of the sharp turning state within the duration of the environmental change.

The environmental posteriori estimation section 31d, after the end of evaluation in the genetic manipulation unit 31b, calculates the thrust efficiency S _t being evaluated, the environment of the actual rating compared with a threshold S _2, S ₃ of the turning state presume. Specifically, the environment is a "straight-running state" if "S _t> S _2" environment if it is "S ₃ <S _t ≦ S _2" is set to "slow turning condition", "S _t ≦ S If “ ₃ ”, the environment is estimated as “sudden turning state”.

Further, the optimization process in the real operation environment by the GA in the genetic operation unit 31b of the parameter optimization device 31 will be specifically described.
In the genetic operation unit 31b, as shown in FIG. 6, first, an initial value of the normalization coefficient is determined by a random number within a range that does not become an extreme value, and a process of setting the initial value as an initial individual is performed. Then, as shown in FIG. 6, an individual group (first generation) composed of the plurality of initial individuals is generated in a number corresponding to the environment. That is, in the present embodiment, individual groups corresponding to three environments of a straight traveling state, a gentle turning state, and a sharp turning state are generated.

Next, the environment pre-estimation unit 31a estimates the environment for performing the next genetic operation by estimating the duration of the above-described environment change from the amount of steering operation performed by the user, and determines the pre-estimated environment number E _pri . .
Then, select the assigned population P_E _pri environmentally E _pri, this subjected to genetic manipulation. In this genetic operation, as shown in FIG. 8, first, a parent individual is selected from the population P_E _pri . In this embodiment, the parent individual is the best evaluation value from the population P_E _pri. In this case, a total of two, one of the best individual having and the one selected at random.

Next, crossover processing is performed on the two selected parent individuals to generate two child individuals corresponding to each parent individual, and as shown in FIG. get the family individual F_E _pri consisting of child individual two. In the present embodiment, since the value of each individual is a fixed-point number, a unimodal normal distribution crossover (UNDX) shown in FIG. 7 which is a representative crossover method of the real-valued GA is used. . Here, in the unimodal normal distribution crossover, offspring individuals can be generated along the dependencies between design variables in the target problem.

When unimodal normal distribution offspring individuals by crossover is generated, further, the evaluation of each individual constituted family individual f_e _pri these.
Further, the evaluated family individuals are replaced with the original parent individuals in the population in descending order of the evaluation value.
Then, when the evaluation of the individual is completed, the environmental posterior estimating unit 31d confirms the propulsion efficiency during the individual evaluation, estimates the posterior environment based on the efficiency, and determines the posterior environment estimation number E _post . When E _post is determined, the environment of E _{pri and} the environment of E _post are compared, and if E _pri = E _post , that is, if the environment estimated in advance is correct, the population corresponding to E _post is used as it is. An evaluation individual is returned to a certain F_E _post , and generation change is performed.

On the other hand, if E _pri ≠ E _post , that is, if the environment estimated in advance is incorrect, substitute it into an unevaluated individual of a family individual in the population corresponding to the environment estimated by the environmental posterior estimating unit 31d. I do.
Here, when the environment estimated in advance is different from the environment estimated afterwards, the processing is performed according to the flow illustrated in FIG. First, the pre-environment estimation unit 31a estimates that the environment will change to environment 1 next, and the genetic operation unit 31b selects a family individual of the population 1 from the population set 31c as the individual to be evaluated (( 1)). Next, in order to perform evaluation, the parameters of the constant-speed cruise control unit 30 are rewritten to parameters coded as individual genes ((2) in the figure). Then, the cruise control unit 30 is actually operated to evaluate the performance of the cruise control unit 30 ((3) in the figure). Further, after the evaluation, the environmental posterior estimating unit estimates that the environment under evaluation is Environment 2, and selects a family individual of the population 2 ((4) in the figure). Finally, the evaluated individual is replaced with a family individual waiting for evaluation ((5) in the figure).

  In other words, if the environment estimated in advance and the environment estimated after are different, the evaluated individual cannot be returned to the individual group corresponding to the original environment. This is because, even for the same individual, evaluation values obtained from the respective environments are different. For example, when an individual excellent in environment 1 obtains a bad evaluation value when evaluated in environment 2, it corresponds to environment 1. When returned to the population, they are eliminated. In addition, individuals that should be originally eliminated in environment 2 obtain good evaluation values when evaluated in environment 1 and then returned to the population corresponding to environment 2. This is because there is a problem that it is not culled forever. Therefore, such individuals are usually considered to be discarded, but a problem arises in that the search does not proceed if sampling opportunities are wasted in the early stage of the search. Therefore, by using the method as shown in FIG. 10A, the chance of sampling at the initial stage of the search is not wasted, and the individual evaluated in the wrong environment is also a candidate for the generation change. The possibility that an individual having an evaluation value is selected increases.

In this way, by repeating the optimization, an individual with good fitness, that is, a normalization coefficient with good fitness is gradually selected, and finally, the engine speed control characteristics corresponding to a plurality of environments can be obtained. It becomes possible.
Further, as shown in FIG. 10B, there is a method of coping with a mistake in the pre-estimation in advance by generating an intermediate individual whose parents are individuals before and after the environment at the transition of the environment. The flow of the process will be described. First, it is assumed that the environmental posterior estimating unit 31d has estimated that the environment under evaluation was the environment 2 ((1) in the figure). It is assumed that the post-environment estimation unit 31d estimates that the environment will change to the environment 1 next. Here, the genetic operation unit 31b determines that this is a change from the previous environment 2, and selects the best individual among the family individuals of the individual groups 1 and 2 ((2) in the figure). Then, in order to evaluate the individual in which the best individuals cross each other, the parameters of the constant-speed cruise control unit 30 are rewritten to parameters coded as individual genes ((3) in the figure). Thereafter, the constant speed cruise control unit 30 is actually operated to evaluate the performance of the constant speed cruise control unit 30 ((4) in the figure). Further, after the evaluation, the environmental posterior estimating unit estimates that the environment under evaluation was Environment 1, and selects a family individual of the individual group 1 ((5) in the figure). Finally, the evaluated individual is replaced with a family individual waiting for evaluation ((6) in the figure).

  That is, since the transition of the environment is a region having the characteristics of the preceding and following environments, there is a high possibility that a mistake in the pre-estimation occurs, and the problem described in the description of FIG. 10A frequently occurs. Therefore, when a change in environment is detected, it is considered that an excellent evaluation value can be obtained in either environment by generating an intermediate child individual whose parent is an individual having an excellent evaluation value in each environment. . Therefore, by using the method described in the description of FIG. 10B, it is possible to deal with a mistake in the preliminary estimation in advance.

Furthermore, the flow of the operation processing in the constant-speed cruise control unit 30 will be described with reference to FIG. FIG. 9 is a flowchart showing an operation process in the constant-speed cruise control unit 30.
As shown in FIG. 9, the process first proceeds to step S100, sets the target speed, and proceeds to step S102. Here, the target speed may be an arbitrary value determined by the user as described above, or may be selected from a plurality of values prepared by the manufacturer at the time of shipment.

In step S102, an initial value of the target engine speed is set, and the process proceeds to step S104. Here, the initial value of the target engine speed may be an arbitrary value determined by the user as described above, or may be selected from a plurality of values prepared by the manufacturer at the shipping stage. Is also good.
In step S104, the target engine speed is calculated by the target engine speed calculation module 30a using the fuzzy inference system as described above, and the calculation result is input to the electronic throttle valve opening calculation module 30b. Move to

In step S106, the electronic throttle valve opening calculation module 30b calculates the electronic throttle valve opening corresponding to the target engine speed using the fuzzy inference system based on the input target engine speed as described above. Then, control proceeds to a step S108.
In step S108, the constant-speed cruise control unit 30 controls the electronic throttle valve device 2a of the outboard motor 2 based on the calculated electronic throttle valve opening, and then proceeds to step S110.

In step S110, it is determined whether or not the cruise control is OFF. If it is determined that the cruise control is OFF (Yes), the process proceeds to step S112, and if not (No), the process proceeds to step S104. .
If the process has proceeded to step S112, the throttle opening is detected and the process ends.
Further, the flow of the operation processing of the parameter optimizing device 31 according to the first embodiment will be described with reference to FIG. FIG. 11 is a flowchart showing the operation processing of the parameter optimization device 31.

As shown in FIG. 11, first, the process proceeds to step S200, the initial value of the normalization coefficient is determined by a random number within a range that does not become an extreme value as described above, and the process proceeds to step S202.
In step S202, in advance, the environment (E _pri ) is estimated by estimating the duration of the environmental change by inputting the steering operation amount, which is the factor of the turning, as described above, in the environment preliminary estimating unit 31a. The process moves to S204.

In step S204, to determine whether there unevaluated individuals in family population f_e _pri for environments, if unrated individual does not exist (Yes), the process moves to step S206, there unevaluated individuals If yes (No), the process moves to S210.
When the process proceeds to step S206, in the genetic manipulation section 31b, from a population P_E _pri for environments among the generated population set 31c, the parent individual, the best individual and random in the population Two of the selected individuals are selected, and the process proceeds to step S208.

In step S208, the parent individuals selected, performs a crossover process with monomodal normal distribution crossover technique as described above, the step to form a product to families populations f_e _pri two individuals S210 Move to
In step S210, unrated individual family population f_e _pri (including parent individuals) was transferred to the constant-speed cruising control unit 30 proceeds to step S212.

In step S212, as described above, the genetic operation unit 31b determines whether or not the evaluation value calculated by performing the electronic throttle valve control on the unevaluated individual has been obtained. If it is determined that the evaluation value has been obtained ( (Yes) moves on to step S214, otherwise (No) stands by until acquisition.
When the process proceeds to step S214, when the evaluation of the individual is completed, the environment (E _post ) at that time is estimated by the environment ex post estimation unit 31d using the propulsion efficiency as an input, and the process proceeds to step S216.

In step S216, it is determined whether or not the pre-environment (E _pri ) and the post-environment (E _post ) are equal. If it is determined that they are equal (Yes), the process proceeds to step S218; Move to step S220.
When the process proceeds to step S218, an operation of returning the evaluated individual to the family population F_E _post is performed as described above, and the process proceeds to step S222.

On the other hand, when the process proceeds to step S220, as described above, among the family population F_E _post corresponding to the environment estimated by the environmental posterior estimating unit 31d, the unevaluated individual and the evaluated individual are replaced, and the process proceeds to step S222. Transition.
In step S222, it is determined by the genetic operation unit 31b whether or not an unevaluated individual exists in the family population F_E _{post. If} there is no unevaluated individual (Yes), the process proceeds to step S224, and the unevaluated individual is determined. If an individual exists (No), the process proceeds to S202.

When the process proceeds to step S224, the individuals of the family population F_E _post are rearranged in ascending order of the evaluation value, and the process proceeds to step S226.
In step S226, the genetic operation unit 31b returns the top two individuals with good evaluation values of the family population F_E _{post to} the place where the parent individuals of the population P_E _post existed, and proceeds to step S228.

In step S228, the genetic operation unit 31b determines whether or not the optimization has been completed. If it is determined that the optimization has been completed (Yes), the process ends. If not (No), the process proceeds to step S202. I do.
Further, an operation process of the parameter optimizing device 31 according to the second embodiment of the present invention will be described based on FIG. 12 and FIG. FIG. 12 is a flowchart illustrating an operation process of the parameter optimizing device 31 according to the second embodiment, and FIG. 15B is a diagram illustrating a data structure of an individual according to the second embodiment.

In the present embodiment, the parameter optimization device 31 enables the evaluated flag for the individual evaluated once (sets the value of the evaluation flag to 1), and in the subsequent optimization processing, the evaluated flag is used. Does not evaluate the individual in the effective state, retains the previous evaluation value as it is, and performs optimization processing using the retained evaluation value.
Therefore, in the present embodiment, as shown in FIG. 15B, the data structure of the individual includes three normalization coefficients 1 to 3 respectively corresponding to the above-described turning state, an evaluation value, and an evaluated flag value. And is included.

The configuration and operation other than those described above are the same as those in the above-described first embodiment, and a description thereof will not be repeated.
Further, a flow of operation processing of the parameter optimizing device 31 according to the second embodiment will be described with reference to FIG.
As shown in FIG. 12, the process first proceeds to step S400, where the initial value of the normalization coefficient is determined by a random number within a range that does not become an extreme value as described above, and the evaluated flag of each individual is invalidated. (The value of the evaluated flag is set to 0), and the process proceeds to step S402.

In step S402, the pre-environment (E _pri ) is estimated in the pre-environment estimating unit 31a by estimating the duration of the environmental change by inputting the steering operation amount that is the factor of the turn as described above. Move to S404.
In step S404, to determine whether there unevaluated individuals in family population f_e _pri for environments, if unrated individual does not exist (Yes), the process moves to step S406, there unevaluated individuals If yes (No), the process moves to S410.

When the process proceeds to step S406, in the genetic manipulation section 31b, from a population P_E _pri for environments among the generated population set 31c, the parent individual, the best individual and random in the population Two of the selected individuals are selected, and the routine goes to Step S408.
In step S408, the selected parent individual is subjected to a crossover process using the unimodal normal distribution crossover method as described above, to generate two individuals whose evaluated flag is in an invalid state, and to generate a family individual A group F_E _pri is formed, and the routine goes to Step S410.

In step S410, the evaluation flag family population f_e _pri is disabled individuals by transmitting the constant-speed cruising control unit 30 proceeds to step S412.
In step S412, as described above, it is determined whether or not the genetic operation unit 31b has obtained an evaluation value calculated by performing electronic throttle valve control on an individual whose family individual evaluated flag is in an invalid state. If it is determined that the information has been acquired (Yes), the evaluation-completed flag is made valid and the process proceeds to step S414. If not (No), the process waits until the information is acquired.

Here, the processing from step S416 to step S428 shown in FIG. 12 is the same as the processing from step S216 to step S228 in FIG. 11 in the first embodiment, and a description thereof will be omitted.
Further, an operation process of the parameter optimizing device 31 according to the third embodiment of the present invention will be described with reference to FIGS. FIG. 13 is a flowchart illustrating an operation process of the parameter optimizing device 31 according to the third embodiment, and FIG. 15C is a diagram illustrating a data structure of an individual according to the third embodiment.

  In the present embodiment, as in the second embodiment, the parameter optimization device 31 enables the evaluated flag for the individual evaluated once (sets the value of the evaluation flag to 1), In the subsequent optimization process, the individual with the evaluated flag in the valid state is not evaluated, the previous evaluation value is held as it is, and the optimization process is performed using the held evaluation value. ing.

Furthermore, in the present embodiment, an unupdated counter for counting the number of times each individual has not been updated in the optimization process is provided, and the value of the unupdated counter is set to a predetermined value a which is a preset threshold value. When the value exceeds, the evaluated flag is invalidated.
Therefore, in the present embodiment, as shown in FIG. 15C, the data structure of the individual includes three normalization coefficients 1 to 3 corresponding to the above-described turning states, an evaluation value, and an evaluated flag value. And an unupdated counter value.

The configuration and operation other than those described above are the same as those in the above-described first embodiment, and a description thereof will not be repeated.
Further, a flow of operation processing of the parameter optimizing device 31 according to the third embodiment will be described with reference to FIG.
As shown in FIG. 13, the process first proceeds to step S500, where the initial value of the normalization coefficient is determined by a random number within a range that does not become an extreme value as described above, and the evaluated flag of each individual is invalidated. (The value of the evaluated flag is set to 0), and the process proceeds to step S502.

In step S502, the pre-environment (E _pri ) is estimated in the pre-environment estimating unit 31a by estimating the duration of the environmental change by inputting the steering operation amount that is the factor of the turn as described above. The process moves to S504.
In step S504, to determine whether there unevaluated individuals in family population f_e _pri for environments, if unrated individual does not exist (Yes), the process moves to step S506, there unevaluated individuals If yes (No), the process moves to S510.

When the process proceeds to step S506, in the genetic manipulation section 31b, from a population P_E _pri for environments among the generated population set 31c, the parent individual, the best individual and random in the population Two of the selected individuals are selected, and the process proceeds to step S508.
In step S508, the selected parent individual is subjected to the crossover process using the unimodal normal distribution crossover method as described above, and two individuals whose evaluated flag is in an invalid state are generated. to form a group f_e _pri, the process proceeds to step S510.

At step S510, the evaluation flag family population f_e _pri is disabled individuals by transmitting the constant-speed cruising control unit 30 proceeds to step S512.
In step S512, as described above, it is determined whether or not the genetic operation unit 31b has obtained an evaluation value calculated by performing electronic throttle valve control on an individual whose family individual evaluated flag is in an invalid state. If it is determined that the information has been acquired (Yes), the evaluation-completed flag is enabled and the process proceeds to step S514; otherwise (No), the process waits until the information is acquired.

Here, the processing from step S514 to step S522 shown in FIG. 13 is the same as the processing from step S214 to step S222 in FIG. 11 in the first embodiment, and a description thereof will be omitted.
In step S524, the individuals in the family individual group F_E _post are rearranged in ascending order of the evaluation value, and the process proceeds to step S550.

In step S550, the non-updated counter values of the top two individuals with good evaluation values of the family individual group F_E _post are incremented by one, and the flow shifts to step S552.
In step S552, it is determined whether or not the value of the non-update counter has exceeded a preset threshold value a. If it is determined that the value has been exceeded (Yes), the process proceeds to step S554, and if not (No), the process proceeds to step S526. Move to

In step S554, the evaluated flag is invalidated, the unupdated counter value is set to 0, and the flow shifts to step S526.
In step S526, the genetic operation unit 31b returns the top two individuals with good evaluation values of the family population F_E _{post to} the place where the parent individuals of the population P_E _post existed, and proceeds to step S528.

In step S528, the genetic operation unit 31b determines whether or not the optimization has been completed. If it is determined that the optimization has been completed (Yes), the process ends. If not (No), the process proceeds to step S502. I do.
Furthermore, an operation process of the parameter optimizing device 31 according to the fourth embodiment of the present invention will be described with reference to FIGS. 14 and 15D. FIG. 14 is a flowchart illustrating an operation process of the parameter optimizing device 31 according to the fourth embodiment, and FIG. 15D is a diagram illustrating a data structure of an individual according to the fourth embodiment.

  In the present embodiment, similarly to the above-described second and third embodiments, the parameter optimization device 31 enables the evaluated flag for the individual evaluated once (the value of the evaluation flag is set to 1). Do), in the subsequent optimization processing, the evaluation flag is not evaluated for the individual in the valid state, the previous evaluation value is held as it is, and the optimization processing is performed using the held evaluation value. It has become.

Further, in this embodiment, the good individuals most evaluation value in the family population f_e _post, when the best individuals has been updated in up to the previous evaluation process, good most evaluation value in the family population f_e _post The priority evaluation flag is made valid for the individual (the value of the priority evaluation flag is set to 1), and the evaluation of the individual whose priority evaluation flag is in the valid state is performed with priority.

Therefore, in the present embodiment, as shown in FIG. 15D, the data structure of the individual includes three normalization coefficients 1 to 3 respectively corresponding to the above-described turning state, an evaluation value, and an evaluated flag value. And a priority evaluation flag value.
The configuration and operation other than those described above are the same as those in the above-described first embodiment, and a description thereof will not be repeated.

Furthermore, the flow of the operation processing of the parameter optimizing device 31 according to the fourth embodiment will be described with reference to FIG.
As shown in FIG. 14, the process first proceeds to step S600, where the initial value of the normalization coefficient is determined by a random number within a range that does not become an extreme value as described above, and the evaluated flag of each individual is invalidated. (The value of the evaluated flag is set to 0), and the process proceeds to step S602.

In step S602, the pre-environment estimating unit 31a estimates the pre-environment (E _pri ) and proceeds to step S604.
At step S604, the family population F_E priority evaluation flag _pri it is determined whether there is an individual in the available state, when it is determined that there (Yes), the process moves to step S606, the otherwise (No) The process moves to S620.

When the process proceeds to step S606, the new best individual previously obtained in step S638 is transmitted to the constant-speed cruise control unit 30, and the process proceeds to step S608.
In step S608, as described above, the genetic operation unit 31b determines whether or not the evaluation value calculated by performing the electronic throttle valve control on the new best individual has been obtained. If it is determined that the evaluation value has been obtained ( (Yes) moves on to step S610, otherwise (No) stands by until acquisition.

In step S610, the individuals in the family individual group F_E _post are rearranged in ascending order of the evaluation value, and the process proceeds to step S612.
In step S612, the priority evaluation flag is invalidated, and the flow shifts to step S614.
In step S614, when the evaluation of the individual is completed, the environment (E _post ) at that time is estimated by the environment posterior estimating unit 31d, and the process proceeds to step S616.

In step S616, it is determined whether or not the pre-environment (E _pri ) and the post-environment (E _post ) are equal. If it is determined that the pre-environment (E _post ) is equal (Yes), the process proceeds to step S644, and if not (No), It moves to step S618.
On the other hand, when the process proceeds to step S618, as described above, among the family population F_E _post corresponding to the environment estimated by the environmental posterior estimating unit 31d, the unevaluated individual and the evaluated individual are replaced, and the process proceeds to step S636. Transition.

Further, in step S604, the family population f_e _pri priority evaluation flag is not a valid state individuals, or when the process proceeds to step S620, unrated individuals present in the family population f_e _pri for environments not If no unevaluated individual exists (Yes), the process proceeds to step S626, and if an unevaluated individual exists (No), the process proceeds to S622.
When the process proceeds to step S622, in the genetic manipulation section 31b, from a population P_E _pri for environments among the generated population set 31c, the parent individual, the best individual and random in the population Two of the selected individuals are selected, and the routine goes to Step S624.

In step S624, the selected parent individual is subjected to a crossover process using the unimodal normal distribution crossover method as described above, to generate two individuals whose evaluated flag is in an invalid state, and to generate a family individual to form a group f_e _pri, the process proceeds to step S626.
In step S626, an individual evaluation completion flag family population f_e _pri is invalid state by transmitting the constant-speed cruising control unit 30 proceeds to step S628.

  In step S628, as described above, it is determined whether or not the genetic operation unit 31b has obtained an evaluation value calculated by performing electronic throttle valve control on an individual whose family individual evaluated flag is in an invalid state. If it is determined that the information has been acquired (Yes), the evaluated flag is enabled and the process proceeds to step S630. If not (No), the process waits until the information is acquired.

Here, the processing from step S630 to step S636 shown in FIG. 14 is the same as the processing from step S414 to step S422 in FIG. 12 in the above-described second embodiment, and a description thereof will be omitted.
In step S638, the individuals in the family individual group F_E _post are rearranged in ascending order of the evaluation value, and the flow advances to step S640.

In step S640, it is determined whether or not the best individual so far has been updated by the individual with the highest evaluation value of the family population F_E _{post. If} it is determined that the individual has been updated (Yes), the process proceeds to step S642, and so on. If not (No), the flow shifts to S644.
When the process proceeds to step S642, the priority evaluation flag of the new best individual is made valid, and the process proceeds to step S644.

In step S644, the genetic operation unit 31b returns the top two individuals with good evaluation values of the family population F_E _{post to} the place where the parent individuals of the population P_E _post existed, and proceeds to step S646.
In step S646, the genetic operation unit 31b determines whether or not the optimization has been completed. If it is determined that the optimization has been completed (Yes), the process ends. If not (No), the process proceeds to step S602. I do.

In the fourth embodiment, the process of monitoring the number of unupdated times of the individual using the unupdated counter described in the third embodiment is added to further increase the robustness against the influence of an error in post estimation. You may do it.
Here, the generation processing of the individual group in the genetic operation unit 31b shown in FIG. 1 corresponds to the individual group generation means according to claim 12 and claim 13, and the selection processing of the individual group corresponding to the prior environment is performed. The processing for selecting a specific individual from the population corresponding to the environment, which corresponds to the individual group selecting means according to claim 12 and claim 13, is performed by the specific individual selection unit according to claim 12 and claim 13. Correspondingly, the crossover process of the specific individual selected by the unimodal normal distribution crossover method corresponds to the crossover processing unit according to claim 12 and claim 13, and the individual evaluation process based on the evaluation value is described in claim 12 and claim The pre-environment estimating section 31a corresponds to the individual evaluation section described in claim 13, the pre-environment estimating section 31d corresponds to the pre-environment estimating means described in claim 12, 13, and 21, and the post-environment estimating section 31d corresponds to claim 13 and 22. Corresponding to the environmental .

Note that, in the above embodiment, a method of selecting an individual having an excellent evaluation value and a random individual has been described as an example of a method of selecting an individual. However, the present invention is not limited thereto, and a roulette selection method, a tournament selection method, and the like. Other selection methods may be used.
Further, in the above embodiment, an example in which the unimodal normal distribution crossover method is used as the crossover method is shown. However, the present invention is not limited to this, and when a gene is represented by a bit string, a one-point crossover method is used. Alternatively, another crossover method may be used.

FIG. 1 is a block diagram illustrating a configuration of a navigation control system according to the present invention. FIG. 4 is an explanatory diagram of a fuzzy rule for calculating a target engine speed in a target engine speed calculation module 30a. FIG. 9 is an explanatory diagram of a fuzzy rule for calculating an electronic throttle valve degree of solution in an electronic throttle valve opening calculation module 30b. It is a figure which shows the landscape of the fitness function at the time of a ship turning. It is a figure which shows the duration of the previous running situation with respect to a steering operation. FIG. 9 is a diagram illustrating a generation range when an initial value of a normalization coefficient is generated as an initial individual of a GA. It is a figure which shows the unimodal normal distribution crossover technique. It is a figure showing signs that a child individual is generated from a parent individual using the unimodal normal distribution crossover method. 4 is a flowchart showing an operation process in a constant speed cruise control unit 30. It is a figure showing generation change of an individual based on evaluation with a posterior environment. 5 is a flowchart illustrating an operation process of the parameter optimization device 31 according to the first embodiment. 11 is a flowchart illustrating an operation process of a parameter optimizing device 31 according to the second embodiment. 13 is a flowchart illustrating an operation process of a parameter optimizing device 31 according to the third embodiment. 15 is a flowchart illustrating an operation process of a parameter optimizing device 31 according to a fourth embodiment. (A) is a diagram showing a data structure of an individual in the first embodiment, (b) is a diagram showing a data structure of an individual in the second embodiment, and (c) is a diagram showing the data structure of the individual in the second embodiment. It is a figure which shows the data structure of the individual in 3rd Embodiment, (d) is a figure which shows the data structure of the individual in 4th Embodiment.

Explanation of reference numerals

Reference Signs List 1 cruise control system 2 outboard motor 2a electronic throttle valve device 3 cruise control device 30 constant speed cruise control unit 30a target engine speed calculation module 30b electronic throttle valve opening calculation module 31 parameter optimization device 31a environment pre-estimation Unit 31b genetic operation unit 31c population set 31d environmental ex-post estimation unit

Claims (24)

An optimization method for optimizing a control parameter of a control device used in an environment where a control condition dynamically changes, using a genetic algorithm,
For each of the changing environment, to generate a population of the genetic algorithm according to the control parameters,
Based on predetermined information obtained under the current environment, the environment at the time of the next genetic operation by the genetic algorithm is estimated in advance from the environment that is in correspondence with the population,
From the population, select a population corresponding to the environment of the estimation result,
For the selected population, as the genetic manipulation,
Select a specific individual from the population corresponding to the environment estimated in advance,
A new individual is generated by performing a crossover process using a predetermined crossover method on the specific individual,
Perform evaluation of these individuals based on the evaluation value obtained from the control result using the new individual and the specific individual,
A method of optimizing a control parameter, comprising performing an updating process on an individual after evaluation based on the evaluation result. An optimization method for optimizing a control parameter of a control device used in an environment where a control condition dynamically changes, using a genetic algorithm,
For each of the changing environment, to generate a population of the genetic algorithm according to the control parameters,
Based on the predetermined information obtained under the current environment, the environment at the time of the next genetic operation by the genetic algorithm is estimated in advance from the environment that is in correspondence with the population,
From the population, select a population corresponding to the environment of the estimation result,
For the selected population, as the genetic manipulation,
Select a specific individual from the population estimated in advance,
A new individual is generated by performing a crossover process using a predetermined crossover method on the specific individual,
Perform evaluation of these individuals based on the evaluation value obtained from the control result using the new individual and the specific individual,
Based on the predetermined information obtained from the environment after the evaluation, the environment at this time is estimated from the environment corresponding to the population,
Compare the environment estimated after the evaluation and the environment estimated in advance,
A control parameter optimizing method, characterized in that an update process of the individual after the evaluation is performed based on the comparison result.   In the comparison of the estimated environment, when the environment estimated beforehand and the environment estimated after the evaluation are different environments, the individual after the evaluation corresponds to the environment estimated after the evaluation. 3. The control parameter optimizing method according to claim 2, wherein a predetermined individual of the population is replaced.   In the comparison of the estimated environment, when the environment estimated in advance and the environment estimated after the evaluation are different environments, specific individuals are respectively selected from the populations corresponding to these environments. The genetic parameter is selected and the genetic operation is performed, and the individual after the genetic operation is replaced with a predetermined individual of a population corresponding to the environment estimated after the evaluation. Optimization method.   The control parameter optimization method according to any one of claims 2 to 4, wherein the specific individual includes a best individual.   It is determined whether or not the individual of the population corresponding to the environment has been evaluated.If it is determined that the individual has been evaluated, the evaluation value of the individual is omitted and the previously acquired evaluation value is retained. The control parameter optimization method according to any one of claims 1 to 5, wherein the updating process is performed using the obtained evaluation value.   A flag indicating that fact is set for an individual determined to be evaluated, the number of individuals of the individual group corresponding to the environment that has survived during the update process is recorded, and the recorded number exceeds a predetermined number. 7. The control parameter optimizing method according to claim 6, wherein the flag that has been set is released when the error occurs.   Compare the current evaluation value of each individual in the plurality of populations with the best evaluation value in the past in the plurality of populations, and compare the current evaluation value to an individual whose current evaluation value is superior to the past evaluation value. 8. The control parameter optimizing method according to claim 6, wherein the updating process is performed after re-evaluation is performed.   The control device is an outboard motor provided in the ship, and the changing environment is determined based on at least one of information on a running speed of the ship, an engine speed, a throttle operation amount, a steering operation amount, and a trim angle. The control parameter optimization method according to any one of claims 1 to 8, wherein:   10. The environment according to claim 1, wherein the preliminary environment is estimated based on at least one of information of a cruising speed of the ship, an engine speed, a throttle operation amount, a steering operation amount, and a trim angle. The control parameter optimizing method according to any one of the preceding claims.   The environment after the evaluation is estimated on the basis of at least one of information on a running speed of a ship, an engine speed, a throttle operation amount, a steering operation amount, and a trim angle. The control parameter optimizing method according to any one of the preceding claims. An optimization device for optimizing a control parameter of a control device used in an environment in which a control condition changes dynamically, using a genetic algorithm,
Corresponding to each of the changing environment, a population generating means for generating a population of the genetic algorithm according to the control parameter,
Based on predetermined information obtained under the current environment, environment prior estimation means for preliminarily estimating an environment at the time of the next genetic operation by the genetic algorithm from among the environments in correspondence with the population,
Population selecting means for selecting a population corresponding to the environment of the estimation result from the population,
Genetic manipulation means for performing the genetic manipulation on the selected population,
The genetic manipulation means,
A specific individual selection unit for selecting a specific individual from a population corresponding to the environment estimated in advance,
A crossover processing unit that generates a new individual by performing a crossover process using a predetermined crossover method on the specific individual;
An individual evaluation unit that evaluates these new individuals and the individual based on an evaluation value obtained from a control result using the specific individual,
A control unit for optimizing the control parameter, comprising: an individual updating unit for performing an updating process of the individual after the evaluation based on the evaluation result. An optimization device for optimizing a control parameter of a control device used in an environment in which a control condition changes dynamically, using a genetic algorithm,
Corresponding to each of the changing environment, a population generating means for generating a population of the genetic algorithm according to the control parameter,
Based on predetermined information obtained under the current environment, environment prior estimation means for preliminarily estimating an environment at the time of the next genetic operation by the genetic algorithm from among the environments in correspondence with the population,
Population selecting means for selecting a population corresponding to the environment of the estimation result from the population,
Genetic manipulation means for performing the genetic manipulation on the selected population,
The genetic manipulation means,
A specific individual selection unit for selecting a specific individual from a population corresponding to the environment estimated in advance,
A crossover processing unit that generates a new individual by performing a crossover process using a predetermined crossover method on the specific individual;
An individual evaluation unit that evaluates these new individuals and the individual based on an evaluation value obtained from a control result using the specific individual,
Based on predetermined information obtained from the environment after the evaluation, an environment posterior estimating unit for estimating the environment at this time from a certain environment of the corresponding population,
An environment comparison unit that compares the environment estimated after the evaluation with the environment estimated in advance,
A control unit for optimizing the control parameters, comprising: an individual updating unit that performs an updating process on the individual after the evaluation based on the comparison result.   When the comparison result in the environment comparison unit is different from the environment estimated beforehand and the environment estimated after the evaluation, the individual updating unit converts the individual after the evaluation after the evaluation. 14. The control parameter optimizing device according to claim 13, wherein a predetermined individual of an individual group corresponding to the estimated environment is replaced.   When the comparison result in the environment comparison unit is an environment in which the environment estimated in advance and the environment estimated after the evaluation are different, the genetic manipulation means sets a population corresponding to these environments And performing the genetic operation by selecting a specific individual from among the individuals, and replacing the individual after the genetic operation with a predetermined individual of an individual group corresponding to the environment estimated after the evaluation. The control parameter optimizing device according to claim 13, characterized in that:   The control parameter optimizing device according to claim 13, wherein the specific individual includes a best individual.   The genetic manipulation means determines whether or not the individual of the population corresponding to the environment has been evaluated, and when it is determined that the individual has been evaluated, the evaluation value obtained by omitting the evaluation of the individual is omitted. 17. The control parameter optimizing device according to claim 13, wherein the update processing is performed using the held evaluation value.   The genetic operation means sets a flag indicating that the individual is judged to be evaluated, records the number of times the individual of the individual group corresponding to the environment has survived the update process, and 18. The control parameter optimizing device according to claim 17, wherein the set flag is released when the number of times exceeds a predetermined number.   The genetic operation means compares a current evaluation value of each individual in the plurality of populations with a past best evaluation value in the plurality of populations, and the current evaluation value is a past evaluation value. 19. The control parameter optimizing apparatus according to claim 17, wherein the updating process is performed after re-evaluating a better individual.   The control device is an outboard motor provided in the ship, and the changing environment is determined based on at least one of information on a running speed of the ship, an engine speed, a throttle operation amount, a steering operation amount, and a trim angle. 20. The control parameter optimizing device according to claim 12, wherein the control parameter is optimized.   The environment pre-estimating means estimates the pre-environment based on at least one of the following information: a traveling speed of a ship, an engine speed, a throttle operation amount, a steering operation amount, and a trim angle. Item 12. The control parameter optimizing device according to claim 20.   The post-environment estimating unit estimates the post-evaluation environment based on at least one of information among a cruising speed of a ship, an engine speed, a throttle operation amount, a steering operation amount, and a trim angle. The control parameter optimizing device according to any one of claims 13 to 21. A computer-executable program for optimizing a control parameter of a control device used in an environment in which a control condition dynamically changes using a genetic algorithm,
For each of the changing environment, to generate a population of the genetic algorithm according to the control parameters,
Based on predetermined information obtained under the current environment, the environment at the time of the next genetic operation by the genetic algorithm is estimated in advance from the environment that is in correspondence with the population,
From the population, select a population corresponding to the environment of the estimation result,
For the selected population, as the genetic manipulation,
Select a specific individual from the population corresponding to the environment estimated in advance,
A new individual is generated by performing a crossover process using a predetermined crossover method on the specific individual,
Perform evaluation of these individuals based on the evaluation value obtained from the control result using the new individual and the specific individual,
A control parameter optimizing program for performing an updating process of an individual after an evaluation based on the evaluation result. A computer-executable program for optimizing a control parameter of a control device used in an environment in which a control condition dynamically changes using a genetic algorithm,
For each of the changing environment, to generate a population of the genetic algorithm according to the control parameters,
Based on predetermined information obtained under the current environment, the environment at the time of the next genetic operation by the genetic algorithm is estimated in advance from the environment that is in correspondence with the population,
From the population, select a population corresponding to the environment of the estimation result,
For the selected population, as the genetic manipulation,
Select a specific individual from the population corresponding to the environment estimated in advance,
A new individual is generated by performing a crossover process using a predetermined crossover method on the specific individual,
Perform evaluation of these individuals based on the evaluation value obtained from the control result using the new individual and the specific individual,
Based on the predetermined information obtained from the environment after the evaluation, the environment at this time is estimated from the environment of the corresponding population,
Compare the environment estimated after the evaluation and the environment estimated in advance,
A control parameter optimizing program for performing an updating process of the individual after the evaluation based on the comparison result.

Images