JPH10154002A - Synthetic control system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize a characteristic that all users satisfy by changing the control characteristic of a control system controlling a controlled system by adjusting it to the characteristics of a user and a use situation. SOLUTION: The evaluation system of evolution adaptability inputs external information and/or information on the characteristic of the user and/or information on the characteristic of the use situation, estimates the characteristic of the user using the control object and/or the use situation and evaluates an evolution processing in an evolution adaptation system based on the characteristics of the user and/or the use situation. The control module of the evolution adaptation system is hereditarily evolved based on the judgment of the evaluation system and at least one optimum control module is obtained at that time. A learning layer learns an input/output relation on the optimum control module where an evolution adaptation layer is evolved and the input/ output relation of a control system for execution on learning layer in the other control system while one control system executes control. The characteristic of the controlled system changes to a movement adapted to the characteristics of the user and use situation at every moment.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an overall control system for comprehensively controlling an object to be controlled.

[0002]

2. Description of the Related Art Conventionally, when controlling the characteristics of a product such as a vehicle or a home appliance, the characteristics of the product to be controlled are assumed at a development / design stage by assuming a user who will use the product. Taking into account the preferences and usage conditions of the virtual user, the virtual user is determined so as to adapt to a wide range of users.

[0003]

However, users who use the above-mentioned products have individual peculiar personalities,
Since the tastes are also different, as described above, all users are satisfied even if they develop and design products assuming the preferences of the users who are supposed to use the product. The problem is that it is impossible to provide properties. In order to solve the above-mentioned problems, at present, before purchasing a product, a user checks the characteristics of the product and determines whether or not the characteristics are satisfactory. The confirmation of the characteristics of the product before the purchase is troublesome for the user. Also, since basically the same product is often controlled with the same characteristics, for example, if you like the product design but do not like the characteristics, you have to give up purchasing the product, etc. Therefore, there is also a problem that the selection range of the product is limited. The present invention
It is an object of the present invention to solve the above-mentioned conventional problems and to provide a comprehensive control system capable of realizing characteristics that can be satisfied by all users.

[0004]

Means for Solving the Problems In order to solve the above-mentioned problems, an integrated control method according to the present invention judges the characteristics of a user and / or a use situation, and based on the judgment result,
The present invention is characterized in that the control characteristics of a control system for controlling a control target are changed in accordance with characteristics of a user and / or a use situation.

[0005]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing a general control system according to an embodiment of the present invention. FIG. 1 is a block diagram showing the basic concept of the integrated control system according to the present invention. As shown in the drawing, this comprehensive control system includes three control layers, a reflection layer, a learning layer, and an evolution adaptive layer, and inputs information (for example, information about an operation state) on a control target to be controlled from the outside. Then, based on the input information, the reflective layer determines the basic operation amount, the learning layer and the evolution adaptive layer determine the correction amount for the basic operation amount, and determines the final control output from the basic operation amount and the correction amount. I do. Hereinafter, the functions of the reflection layer, the learning layer, and the evolution adaptive layer in the comprehensive control method will be described. The reflection layer controls the relationship between information about the control target (hereinafter, referred to as external information) and a basic operation amount for the external information in the form of a mathematical expression, a map, a fuzzy rule, a neural network, or a subsumption architecture. This layer is provided in advance in the system, and when external world information is input, the basic operation amount for the external world information input from the control system is determined and output. Note that the subsumption architecture is known as behavioral artificial intelligence that performs parallel processing. The evolution adaptive layer is composed of two control systems, an evaluation system and an evolution adaptive system. The evaluation system inputs external world information and / or information on characteristics of the user (e.g., preference, skill, or state, etc.) and / or information on characteristics of a use situation (e.g., change in use environment), The user using the controlled object to be controlled and / or the characteristics of the use situation are estimated from the external information and the like, and the evaluation during the evolution processing in the evolution adaptive system is performed based on the characteristics of the user and / or the use situation. Do. The evolution adaptive system includes at least one control module for correcting the basic operation amount determined by the reflective layer so as to match the characteristics of the user and / or the use situation, and the control module is configured to determine the control module based on a determination of the evaluation system. Genetically evolving to obtain at least one optimal control module at that time. After acquiring the optimal control module, the evolution adaptive system fixes the control module to the optimal one, and outputs an evolution correction value for correcting the basic operation amount output from the reflection layer. The learning layer has two control systems that can be switched between learning and execution. While the control is being executed by one control system (for execution), the other adaptive control system (for learning) uses the evolution adaptive layer. The input / output relation of the optimal control module that has evolved and the input / output relation of the control system for executing the learning layer are learned together. When the learning in the learning control system is completed, the control system that is executing the control and the control system after learning are switched, and control using the control module obtained from the learning result is started in the learned control system, and the control is executed. The control system started to function for learning. The control system in the learning layer is set so as to output zero in the initial state. Therefore, in the initial state, the control by the reflection layer and the evolution adaptive layer is performed. After the learning layer learns information about the optimal control module,
The output is returned to zero, the operation is performed at regular time intervals, the evaluation of the user's preference and / or the use environment and the evolution of the control module are performed, and the evaluation when the output of the evolution adaptation layer is added is the evolution. If it is better than the evaluation when the output of the adaptive layer is not applied, the learning layer is again made to learn the information on the optimal control module. The information about the learned control modules in the learning layer can be stored and read out in an external storage means such as an IC card or a floppy disk. The basic correction amount can be read out from the storage means and output by the learning layer based on the information. Further, when the user reads the information on the past optimum control module from the external storage means and activates the learning layer, while the learning layer operates by the read control module, the output of the evolution adaptive layer is zero. To stop the evolution of the control module. By the operation of each layer described above, the control output from the integrated control system changes every moment in accordance with the characteristics of the user who uses the control target to be controlled, such as the preference of the user who uses the control target and changes in the use environment. Characteristic changes momentarily to an operation adapted to the characteristics of the user and / or the use situation. In the present specification, a state in which the characteristics of the controlled object evolve in accordance with the characteristics of the user and / or the use situation by this comprehensive control method is referred to as “training”.

FIG. 2 is a flow chart showing the above-mentioned comprehensive control method over time. In the initial state, the output of the learning layer is zero (step a). Therefore, immediately after the use of the control target is started, the control target is controlled only by the basic operation amount from the reflection layer. When the use of the controlled object starts,
The evolution adaptation layer evaluates the characteristics of the user and / or the use situation, and evolves the control module according to the evaluation value (step b). The evolution adaptation layer obtains at least one control module that is most desirable at that time by genetically evolving each control module (step c). The evolution adaptive layer fixes the control module to the most desirable control module obtained in step c, outputs an evolution correction value based on the control module, and corrects the basic operation amount output from the reflection layer. In the learning layer, the learning control system learns the input / output relationship of the evolution adaptive layer and the input / output relationship of the execution control system of the learning layer when the evolution adaptive layer is fixed to the optimal control module. In the initial state, the output of the control system for execution of the learning layer is zero, but after learning, the basic operation amount from the reflection layer is represented by the basic correction amount from the learning layer and the evolutionary correction amount from the evolution adaptive layer. Correction is performed (step d). The difference between the value obtained by adding the output of the learning control system in the learning layer to the basic operation amount from the reflection layer and the actual control output (basic operation amount + basic correction amount + evolution system correction amount) is smaller than the threshold value. At this point, the learning control system in the learning layer ends learning, the control system for learning and the control system for execution are switched, and the control system after learning functions as execution and executes control. The system functions for learning (step e), and control is performed by the reflection layer and the learning layer (step f). After making the learning layer learn the information about the optimal control module, the evolution adaptive layer operates at regular time intervals, and evaluates the time-dependent deviation of the control law of the learning layer (step g). Specifically, when the maximum fitness is not improved in the initial generation when the control module of the evolution adaptive layer is genetically evolved, it is determined that there is no shift in the control rule of the control layer and the process proceeds to step f. When the control by the reflection layer and the learning layer is continued and the maximum fitness is improved, the process proceeds to step b to acquire a new optimal control module in the evolution adaptive layer.

Next, the above-described comprehensive control method will be described more specifically, taking a vehicle engine as an example of a control target. FIG. 3 is a schematic diagram illustrating a relationship between the engine 1 and a control device 10 that executes the comprehensive control method.
As shown in the drawing, the control device 10 controls an engine speed,
Engine control that inputs information such as intake negative pressure, throttle opening, throttle opening change rate, atmospheric pressure, intake air temperature, cooling water temperature, gear position, etc., and achieves both fuel economy performance and acceleration performance based on these input information. I do. FIG. 4 is a schematic block diagram of the control device 10. This control device 10
Consists of a reflective layer, a learning layer, and an evolution adaptive layer as described above.

(Reflection Layer) The reflection layer inputs the engine speed, intake negative pressure, throttle opening, throttle opening change rate, atmospheric pressure, intake air temperature, cooling water temperature, and the like. A basic value of the fuel injection amount (that is, a basic operation amount of the fuel injection device) is determined and output from an equation obtained by modeling a predetermined mathematical formula based on the calculated value.

(Evolution Adaptive Layer) The evolution adaptive layer is composed of an evaluation system and an evolution adaptive system. FIG. 5 is a flowchart of the basic operation of the evolution adaptive layer. Hereinafter, the basic operation of the evolution adaptive layer will be described with reference to this flowchart. The evaluation system learns the relationship between the distribution pattern of the maximum rotational speed at each gear position within a certain time (see FIGS. 6 and 7) and the running state index P (FIG. 8).
), A gear position signal and an engine speed signal are input (step 1), and a running state index P is determined by the neural network from the input information.
For example, a user who likes sporty running tends to turn the engine to a high rotation speed with a low gear, so that the distribution pattern is as shown in FIG. 6 (a). Are likely to change to high-speed gears early, and the distribution pattern is shown in FIG.
It becomes as shown in. Therefore, in the neural network shown in FIG. 8, the running state index P becomes large when the distribution pattern shown in FIG. 6A is used, and when the distribution pattern shown in FIG. If learned in advance so as to have a small value, the running state index P obtained from the neural network becomes larger as the user's preference for running is more sporty, and becomes smaller as the user's driving preference is milder. , Reflecting user preferences. In addition, as shown in FIG. 7A, when each of the first to sixth gears is used within a certain time, the traveling state index P
Becomes large, and when only low-speed gears are used as shown in FIG. 7 (b), the value of the traveling state index P becomes slightly large, and only high-speed gears are used as shown in FIG. 7 (c). If learning is performed by the neural network so that the value of the running state index P becomes small when used, each gear is used in a normal vehicle and a low speed is used in a congested road when the vehicle is running normally. Since only gears are used, and only high-speed gears are used when driving on a highway, for example, based on a running state index P obtained by a neural network, FIG.
In the case of FIG. 7A, it can be estimated that the vehicle is traveling normally, in the case of FIG. 7B, it is traveling on a congested road, and in the case of FIG. The evaluation system estimates the running condition at that time from the running state index P, and determines whether the user's preference is fuel efficiency performance or acceleration performance, and determines the acceleration importance ratio α. As shown in FIG. 9, the acceleration-oriented ratio α is obtained from a function of a predetermined acceleration-oriented ratio α and the traveling state index P. For example, when the vehicle is traveling on a highway (see FIG. 7C), When the index P is small, the acceleration-oriented ratio α is a small value that emphasizes fuel economy, and when the vehicle is running sporty (see FIG. 6B), the traveling state index P
Is large, the acceleration-oriented ratio α is a value that emphasizes the acceleration. The evolution adaptive system has a fuel efficiency module and an acceleration module, and obtains an adaptive change by cooperating and competing with each other. Each of the modules,
As shown in FIG. 10, each is composed of a hierarchical neural network having two inputs and one output. The fuel efficiency module aims at improving the fuel efficiency performance, and the acceleration module aims at the acceleration performance improvement. The input to each module is an engine speed signal and a throttle opening. Based on these inputs, each module outputs a correction amount of the fuel injection amount (that is, a correction amount of the basic operation amount from the reflection layer). It is configured. The evolutionary adaptation layer adjusts the degree of coupling of the neural network that constitutes the fuel economy module and the acceleration module in accordance with the evaluation in the evaluation system, that is, the user's Alternately evolve using a genetic algorithm according to taste and usage environment (Step 4), and after the evolution of both modules is completed, fix the connectivity of the neural network of both modules at the evolved connectivity do it,
The engine is controlled by the evolution correction value Y based on the outputs of both modules (step 5). Step 4 described above
The evolution of the module due to the genetic change in is alternated for each module, so when the fuel economy module is evolving, the connectivity of the neural network of the acceleration module is fixed, and the acceleration module is evolved. The degree of coupling of the neural network of the fuel economy module is fixed. The engine control based on the evolution correction value Y in the evolution adaptation layer is continued until the learning by the learning layer described later is completed (step 6). When the learning in the learning layer is completed, the fuel economy module and the acceleration module are reset. The output of the evolution adaptation layer is set to zero (step 7).

Hereinafter, the evolution of the module by the genetic algorithm will be described with reference to the flowchart of FIG. First, FIG.
As shown in (1), a first generation consisting of a plurality of individuals an (in this embodiment, nine individuals) is generated by coding the coupling coefficient of a neural network constituting the fuel efficiency module as a gene (in the present embodiment, 9). Step 1). The initial value of the gene value of each individual (that is, the value of the coupling coefficient of the neural network) is randomly determined within a predetermined range (between approximately -10 and 10). At this time, if the learning layer has already learned and output, the number of individuals can be reduced by including one individual (individual a (1) in FIG. 12) that can make the output of the evolution adaptive layer zero. Even when there is a restriction on, diversity of the population during the evolution process can be maintained without impairing the performance at that time. Next, for one of the individuals an generated in step 1, for example, the individual a (1),
Using the neural network of the fuel economy module, the output x of the neural network with respect to the actual input information (engine speed and throttle opening) is determined (step 4), and this output is linearly transformed using equation (1). Then, the output yf of the fuel efficiency module is determined (step 5). The engine speed and the throttle opening of the input information are each normalized, and the output from the fuel consumption module is obtained by linearly converting the output of the neural network using the following equation (1). yf = 2 × Gx−G (1) where yf is the output of the fuel economy module, x is the output of the neural network in the fuel economy module, and G is the evolution adaptive layer output gain. As described above, by using the output x of the neural network in a linear transformation, the output yf from the fuel efficiency module does not become an extremely large value, and the evolution proceeds little by little as a whole, and the behavior of the engine is improved. Will not fluctuate extremely due to evaluation or evolution. After determining the output yf of the fuel economy module for the individual a (1), the output yf and the output ya of the acceleration module having a fixed coupling coefficient are averaged to obtain the output of the evolution adaptive layer (the provisional correction amount Yn (1) ) Is calculated (step 6). This weight is the acceleration-oriented ratio α obtained by the evaluation system.
Decide from. Assuming that the output of the fuel efficiency module is yf and the output of the acceleration control module is ya, the provisional correction amount Yn output from the evolution adaptation layer is represented by the following equation (2). Yn = αya + (1−α) yf (2) That is, when the acceleration-oriented ratio is 1, the provisional correction amount Yn is an output of only the acceleration module, and when the acceleration-oriented ratio is 0, the provisional correction amount Yn is only the fuel consumption module. Output.
After the output Yn (1) of the evolution adaptive layer for the individual a (1) is determined, the provisional correction value Yn (1) is actually output from the evolution adaptive layer and added to the basic operation amount from the reflection layer, The engine is operated with the control output corrected by the provisional correction value Yn (1) (step 7). The evaluation system in the evolution adaptation layer calculates the fuel efficiency by inputting the feedback information on the fuel efficiency from the engine operated by the control output corrected by the provisional correction amount Yn (1) obtained from the individual a (1) (step 8). Based on the result, the individual a (1) is evaluated, and the fitness for the individual a (1) is obtained (step 9). The fuel efficiency is calculated from the traveling distance and the fuel consumption.
The processing from step 4 to step 9 is performed until the fitness for the nine individuals a (1) to a (9) generated in step 1 is calculated, and the fitness for all the individuals is calculated. Proceed to the next process later (step 1
0). Step 4 from the second individual a (2)
In the processes from Step 9 to Step 9, in order to evaluate the fitness of each individual under the same conditions, the running condition is checked before the process is started (Step 2), and the running condition when the running condition is the first individual a (1) is checked. It is determined whether it is the same as the driving situation (step 3).
After calculating the fitness for all individuals, it is determined whether the generation to which the individual belongs is the last generation (step 11),
If it is not the last generation, a parent individual is selected (Step 1)
2). For this selection, a roulette-type selection method is used, and some parent individuals are selected stochastically with a probability proportional to the fitness of each individual. At this time, if the generational change is applied too strictly, individuals with high evaluations may be destroyed. Therefore, an elite preservation strategy that unconditionally leaves the elite (the individual with the highest evaluation) to the next generation is also combined. Used. In addition, a linear conversion of the fitness is performed so that the ratio between the maximum fitness and the average fitness in a group including a plurality of individuals is constant. When the selection of the parent individual is completed, crossover is performed using the selected individual as a parent individual to generate a second generation consisting of nine child individuals (step 13). For crossover between individuals, a technique such as one-point crossover, two-point crossover, or normal distribution crossover is used. The normal distribution crossover is a method of generating children according to a normal distribution that is rotationally symmetric with respect to an axis connecting parents with respect to a chromosome (individual) represented by a real value. The standard deviation of the normal distribution is proportional to the distance between the parents for the main axis component connecting the parents, and proportional to the distance between the straight line connecting the parents and the third parent sampled from the population for the other axis components. . This crossover method has the advantage that the characteristics of the parent are easily inherited by the child. In addition, the gene (coupling degree) value is randomly changed with a certain probability for the generated nine offspring individuals to cause gene mutation. After the second generation is generated by the above-described processing, the coupling coefficient of the neural network of the fuel economy module is fixed to any individual (elite) of the second generation, and the process proceeds to the evolution process of the acceleration module. After the first generation evolution processing is completed, the processing from step 1 is performed again to evaluate and select each individual of the second generation (see FIG. 14). At this time, in step 1, it is determined whether or not there is a coded individual, that is,
It is determined whether it is the second generation or later, and if there is a coded individual, the process proceeds to step 2 without performing the coding process. This process is repeatedly performed until the generation to be generated reaches a predetermined final generation. In this way, the offspring individuals that constitute each generation will follow the evaluation of the evaluation system,
That is, it evolves according to the user's preference. It is determined in step 11 whether or not the last generation has been reached.
When it is determined that the child is the last generation in step 1, an individual having the highest fitness (optimal individual), that is, one elite is selected from nine child individuals of the generation (step 14), and the neural network of the fuel economy module is selected. Is fixed with the gene constituting the above-mentioned optimal individual (step 15), and the evolution of the fuel economy module is terminated. In the acceleration module, the same processing as that of the fuel efficiency module described above is performed for each generation until the last generation is reached. Note that the evaluation of the individual corresponding to steps 8 and 9 in the acceleration module is performed using the acceleration evaluation index. The acceleration evaluation index is calculated by dividing the acceleration by the throttle opening change rate. FIG. 13 is a graph showing the relationship between the change in vehicle speed and the acceleration evaluation index at the same throttle opening.

In the above-described genetic algorithm, the following methods (1) to (3) are also considered. (1) Mutation of duplicate individuals Despite selecting different individuals as crossover parents,
If they are genetically identical, they are mutated with a higher than normal probability for both crossing parents. However, the mutation at this time changes the value of the selected gene based on a normal distribution. (2) Avoidance of crossover of the same individual There is a possibility that the parent of the crossover is selected and this is the same individual, but if this is left unchecked, it is expected that diversity as a group will be lost. For this reason, when the parent selected for crossover is the same individual, it is replaced with another selected individual, and crossover of the same individual is avoided as much as possible. (3) Reproduction method Instead of crossover, a reproduction method that is a method of generational replacement that replaces all individuals in a group at once is used.

When the evolution of the fuel economy module and the acceleration module is completed by the above-described genetic algorithm, and the neural networks of these modules are both fixed at the optimum degree of connection of the individual, the evolution adaptive layer becomes as described above. To
The fixed fuel consumption module and acceleration module output the evolution correction value Y. The evolution correction value Y determines the outputs of the neural networks of each module based on the input signals (engine speed and throttle opening), and linearly converts the outputs of these neural networks by the above equation (1). The value is set as the output yf, ya of each module, and furthermore, it is obtained by taking a weighted average of the outputs yf, ya of both modules using the above equation (2).

As described above, by generating and crossing a plurality of individuals for each module, a plurality of individuals compete between modules of the same type to evolve into a better module. By separately and alternately evolving the acceleration modules, the changing module evolves adaptively according to the output of the non-changing module, and cooperation between different modules is achieved. However, when the weight of one control module is small, even if a genetic change is applied to the control module, it is difficult to appear in the result. Therefore, it is conceivable to apply a genetic change only to a control module having a relatively large weight.

(Learning layer) The learning layer is composed of two neural networks A and B. When one of the two neural networks functions for learning, the other functions for execution. The neural network for learning is
When the evolution of each module is completed in the evolution adaptation layer, and the neural networks of the fuel economy module and the acceleration module are fixed at the optimum degree of connection of the individual individuals, the relationship between the input and output of the evolution adaptation layer is executed by the learning layer. The learning is performed together with the relationship between the input and output of the neural network functioning as a function.
During this time, the output of the evolution adaptation layer is performed by the fuel consumption module and the acceleration module that maximize the previous evaluation function, and the control law does not change over time. In the learning described above, inputs and outputs between the evolution adaptive layer and the neural network for execution of the learning layer are averaged with a certain step width, and this is used as input / output data for updating the teacher data set. For example, if the average engine speed per second is 5000 rpm and the average throttle opening is 20, the fuel injection correction amount of the execution neural network in the evolution adaptive layer and the learning layer at that time (that is, the evolution correction amount) The sum of the amount and the reference correction amount is used as input / output data (FIG. 1).
5). This input / output data is added to the previous teacher data to obtain a new teacher data set. At this time, old teacher data whose Euclidean distance to new data in the teacher data set is within a certain value is deleted. This situation is shown in FIG.
Shown in The initial value of the teacher data set is set to zero for all input data. At the learning layer,
Based on the updated teacher data set, the learning of the coupling coefficient of the neural network for learning is performed. The learning of the coupling coefficient is based on the output of the learning neural network during learning (ie,
This is performed until the error between the virtual control output obtained from the virtual correction value) and the basic operation amount from the reflection layer and the actual control output becomes equal to or less than the threshold value. The neural network is for execution, and the original neural network for control is for learning. After that, the learning layer determines the basic correction amount using the newly obtained neural network for execution and actually outputs it.At the same time, the output of the evolution adaptive layer becomes zero, and the control by the learning layer and the reflection layer is performed. Done.
The initial value of the neural network for executing the learning layer is
Set so that the output is always zero. By doing so, in the initial state, control can be performed only by the reflection layer and the evolution adaptive layer. The learned coupling coefficients of the execution neural network can be stored and read out in external storage means such as a floppy disk or an IC card.

As described above, after the evolution in the evolution adaptive layer is completed and the learning is learned in the learning layer, the evolution adaptive layer operates at fixed time intervals as described with reference to FIG. The deviation of the control law is checked, and if there is a deviation in the control law, the fuel control module and the acceleration control module are evolved again. If the user reads the coupling coefficient stored in the external storage means and operates the learning layer based on the coupling coefficient, the user does not check the control law deviation in the evolution adaptive layer, The evolution adaptation layer may be configured so that the process is stopped while the output of the evolution adaptation layer is fixed to zero, and the process of the evolution adaptation layer is restarted based on a processing start instruction from the user. As described above, the driver's preference is evaluated by the evolution adaptation layer, and the fuel efficiency control module and the acceleration control module are evolved in accordance with the evaluation, whereby the engine 1 emphasizes the fuel efficiency performance according to the user's preference. The training is conducted in a type or drivability performance-oriented type, and by making the evolutionary adaptation layer function at regular time intervals, the training follows changes in the user's preference and changes in the engine and the vehicle over time. .

As in the above-described embodiment, the following two advantages are obtained by using the technique in which the evolution adaptation layer adapts while limiting the correction amount and learns the results obtained therefrom. Can be・ Diversity of control modules in the evolution adaptation layer is secured,
Global search, which is a feature of the genetic algorithm, becomes possible. -It is possible to acquire faster and more intelligent information processing of the learning layer from trial and error information processing in the evolution adaptive layer. Although this is not remarkable in the above-described engine control, it is a great advantage for the path control of the mobile robot.

Next, another embodiment of the control device for executing the comprehensive control system according to the present invention with the air-fuel ratio of the vehicle engine being controlled will be briefly described. FIG. 17 is a schematic block diagram of the control device 20. This control device 20
Like the control device 10 of the second embodiment, the control device 10 includes a reflection layer, a learning layer, and an evolution adaptive layer. Except for the evolution adaptation layer and the learning layer, the second embodiment is the same as the second embodiment, and a detailed description is omitted here. The evolution adaptation layer is composed of an evaluation system and an evolution adaptation system. The evaluation system estimates the traveling state at that time from the traveling state index P, as in the evaluation system of the second embodiment, and sets the acceleration-oriented ratio relating to the user's preference. Determine α. The evolutionary adaptive system consists of a lower module consisting of a fuel consumption module that aims to obtain a fuel injection control law that maximizes fuel efficiency and a power module that aims to acquire a fuel injection control law that maximizes output. And a higher-level module including a control module that aims to obtain an output ratio of the fuel module and the power module according to the driving state according to the user's preference (see FIG. 17). Each of the modules is constituted by a hierarchical neural network having two inputs and one output, similarly to the control module of the second embodiment. The fuel efficiency module and the power module input the normalized throttle opening and the normalized engine speed, respectively. The control module outputs the primary output value of the evolution correction amount, and the control module inputs the normalized throttle opening and the normalized throttle opening change rate, and outputs the output ratio of the lower modules (ie, the fuel consumption module and the power module). Determines the final output value of the evolution correction amount. As shown in FIG. 18, the fuel efficiency module and the power module have an air-fuel ratio correction amount that can obtain an air-fuel ratio that maximizes fuel efficiency and an air-fuel ratio that optimizes output regardless of the driving state and the preference of the user. The coupling coefficient of each neural network is sequentially and autonomously evolved using a genetic algorithm so as to obtain. On the other hand, the control module adjusts the coupling coefficient of the neural network so as to obtain the output ratio of the fuel consumption module and the power module that is in accordance with the user's preference between the best fuel consumption air-fuel ratio and the best output air-fuel ratio. Evolve autonomously using a genetic algorithm. In addition, the evolution of each control module is based on the result of coding a degree of connectivity of the neural network that constitutes each control control module as a gene, generating multiple individuals, and actually operating the engine using all individuals. Each individual is evaluated by the evaluation system, several parent individuals including the individual with the highest evaluation result (elite individual) are selected, and these are crossed to generate a next-generation offspring individual, and each individual is evaluated. Is performed by repeating the above. The above-described evolution processing is repeatedly performed until a predetermined number of generations are completed, and after the evaluation of the final generation individual is completed, the coupling coefficient of the neural network is fixed by the final generation elite individual, and the learning layer learning control is performed. Train the module. When generating a population during the evolution process, if the learning layer has already learned and output,
One individual that can make the output of the evolutionary adaptive layer zero can be included, so that the diversity of the population during evolution can be maintained even if the number of individuals is limited. The above-described evolution processing is repeatedly performed on the same control module continuously while learning is performed by the learning layer for each predetermined number of generations until the evolution converges. During the evolution process, the control modules other than the control module to be evolved are fixed, and after the evolution of the control module under the evolution process ends, the evolution process of the next control module is performed. At the time of learning for each predetermined number of generations, the elite individual of that generation fixes the coupling coefficient of the neural network and makes the learning control module of the learning layer learn. When the learning is completed, a new population including the elite individuals used for the learning and the individuals whose output of the evolutionary adaptive layer becomes zero is generated. In this way, in the next evolution process, the diversity of the population under evolution can be maintained without impairing the performance of the previous evolution process.

Hereinafter, the evolution of the module by the above-described genetic algorithm will be described in more detail with reference to the flowchart of FIG. First, a first generation consisting of a plurality of individuals an (in this embodiment, nine individuals) is generated by coding the coupling coefficient of the neural network constituting the fuel economy module as a gene (step 1). At this time, by including one individual that can make the output of the evolutionary adaptation layer zero, the diversity of the population during the evolution process is maintained even if the number of individuals is limited while maintaining the performance before evolution. Can be kept. Next, for one of the individuals an generated in step 1, for example, the individual a (1), the actual input information (engine speed and throttle opening) using the neural network of the fuel economy module. Is determined (step 2), and the output is linearly transformed using equation (1) to determine the output yf (1) of the fuel efficiency module for the individual a (1). (Step 3). yf (n) = 2 × Gx (n) -G (1) where yf (n) is the output of the fuel economy module, x (n)
Is the output of the neural network in the fuel economy module, G
Is an evolution adaptive layer output gain, and n indicates an individual. Output yf of fuel economy module for individual a (1)
After determining (1), using the output ya of the power module and the output OR of the control module, the following equation (2) is used.
Of the evolution adaptation layer (evaluation correction amount Ya
(N)) is determined (step 4). Ya (n) = OR × ya + (1−OR) × yf
(N) During the evolution processing of the fuel efficiency module, the coupling coefficient of the power module and the control module is fixed. Output OR of control module as described above
Is the ratio between the output yf of the fuel economy module and the output ya of the power module. Therefore, when the output OR of the control module is "1", the output of the evolution adaptive layer becomes the output of the power module, and the output OR But"
In the case of 0 ", the output of the evolution adaptation layer is the output of the fuel economy module. The output Ya of the evolution adaptation layer for the individual a (1)
After (1) is determined, the provisional correction value Ya (1) is actually output from the evolution adaptive layer, and the provisional correction value Ya (1), the basic correction value Yb from the learning layer, and the basic operation amount from the reflection layer are obtained. And the control based on the operation amount corrected by the correction value Ya (1) + Yb is performed for a predetermined time (step 5).
While actually performing control using the individual a (1), the evaluation system of the evolutionary adaptive layer feeds back information on the result of this control to determine an evaluation value (for example, fuel consumption) for the individual a (1). (Step 6). In the processing from step 1 to step 6 described above, the calculation of all evaluation values for the nine individuals a (1) to a (9) generated in step 1 is performed as one cycle of preliminary evaluation processing. The process is repeated until a predetermined cycle is performed (step 7). After calculating evaluation values for all the individuals for the predetermined cycle, the process proceeds to the comprehensive evaluation process for each individual (step 8). In the comprehensive evaluation process, for each individual, a total evaluation value is calculated by dividing the total travel distance in the above-described evaluation value calculation cycle by the sum of all the evaluation values (in this case, the total fuel consumption). The fitness of each individual is evaluated based on the value (step 8). As mentioned above,
The control using each individual is sequentially performed a predetermined number of times, and the individual is operated in a pseudo-parallel manner by time division (see FIG. 20). It is possible to perform fairly under the same conditions, and it is possible to perform online evaluation, that is, evaluation while the vehicle is running. After the above-described comprehensive evaluation process is completed, it is determined whether or not the generation to which the individual belongs is the last generation (step 9). If not, the parent individual is selected (step 10). For this selection, a roulette-type selection method is used, and some parent individuals are selected stochastically with a probability proportional to the fitness of each individual.
At this time, the elite individual is left unconditionally as a parent individual. When the selection of the parent individual is completed, crossover is performed using the selected individual as a parent individual to generate a second generation of nine child individuals again (step 11). In addition, the gene (coupling degree) value is randomly changed with a certain probability for the generated nine offspring individuals to cause gene mutation. Note that these nine offspring individuals include one that can make the output of the evolutionary adaptive layer zero. After the second generation is generated by the above processing, the preliminary evaluation processing from step 2 is repeated again. The above-described evolution processing is repeatedly performed until a predetermined number of generations has elapsed. As a result, the offspring constituting each generation evolves in accordance with the evaluation of the evaluation system, that is, in the case of the fuel efficiency module, to acquire a fuel injection control law that maximizes the fuel efficiency. Step 9 determines whether a predetermined number of generations has elapsed.
If it is determined in step 9 that the child is the last generation, an individual having the highest fitness (optimal individual), that is, one elite is selected from the nine child individuals of that generation (step 12), and the fuel efficiency is determined. The coupling coefficient of the neural network of the module is fixed with the gene constituting the above-mentioned optimal individual (step 13), and the process proceeds to the learning process for the learning control module of the learning layer. The evolution processing for one control module (in this case, the fuel consumption module) is continuously repeated while the evaluation of the control by the learning layer and the reflection layer after the learning is higher than the evaluation of the control before the evolution. If the evaluation of the control no longer improves than the evaluation of the control before the evolution processing, it is determined that the evolution of the control module has converged, and the process proceeds to the evolution processing of the next control module (power module in this embodiment). I do.

The evolution process of the power module and the control module is performed in the same manner as the above-described evolution process of the fuel efficiency module. However, the total evaluation value for each individual in the power module is obtained by dividing the average engine speed during the evaluation value calculation cycle by the average throttle. It is calculated by dividing by the opening.
As a result, the child individuals constituting each generation during the evolution process of the power module evolve in accordance with the evaluation of the evaluation system, that is, to acquire a fuel injection control law that maximizes the output. The overall evaluation value for each individual in the control module is a fuel efficiency evaluation value and a response evaluation value during an evaluation value calculation cycle, an acceleration importance ratio α,
The determination is made using the reference fuel efficiency evaluation value and response evaluation value. That is, the fuel efficiency evaluation value during the value calculation cycle is FC, the response evaluation value is RP, and the reference fuel efficiency evaluation value is FC.
Assuming that the base and the reference response evaluation value are RPbase, the total evaluation value CNT in the control module is given by the following equation. CNT = (1−α) × Scale × (FC−FCbase) / F
Cbase + α × (RP−RPbase) / RPbase Here, Scale is a coefficient for balancing the fuel efficiency evaluation value and the response evaluation value. The fuel efficiency evaluation value is calculated in the same manner as the overall evaluation value in the fuel efficiency module described above, and the response evaluation value RP is regarded as acceleration if a change in throttle opening larger than a predetermined threshold value continues for a certain period of time. The change in speed is divided by the rate of change in throttle opening, and this is taken as the average of the values calculated for each acceleration during the time of the predetermined evaluation value calculation cycle. As described above, by evaluating the fuel efficiency evaluation value and the response evaluation value in accordance with the ratio of the acceleration-oriented ratio α determined based on the user's preference, the control module can adjust the driving state according to the user's preference. It will evolve to obtain the output ratio of the fuel efficiency module and the power module according to As described above, the fuel economy module and the power module that constitute the lower control module group have evolved to give priority to fuel economy and output, respectively, and the control module that constitutes the upper control module has the output ratio of the lower control module according to the user's preference. , Control can be performed using lower-level modules capable of optimal output for each function (ie, fuel efficiency and engine output) at an output ratio that matches the user's preference.

(Learning Layer) Next, the learning layer will be briefly described. As shown in FIG. 17, the learning layer includes a group of control modules corresponding to the control modules of the evolution adaptive layer, and each control module group includes two neural networks A and B. When one of these two neural networks is functioning for learning, the other functions for execution. The control modules in this learning layer are:
Every time the control module corresponding to the evolution layer evolves by a predetermined number of generations, a neural network for learning the neural network that functions to execute the learning layer by using the relationship between the input and output of the evolution adaptation layer. And learn the relationship between the input and output. The above-mentioned learning ends when the error between the output of the learning neural network during learning and the control output becomes smaller than a threshold value, and thereafter, the neural network for learning becomes an execution one. Is used for learning. This learning is performed every time a predetermined number of generations of the evolution processing of the control module in the evolution adaptive layer are performed as described above. Therefore, from the start of the evolution process to one control module until the convergence of the evolution, learning is performed every time a predetermined number of generation evolution processes are performed. As described above, when the evolution and the learning are alternately repeated for each control module, the evaluation value is improved faster and the evolution proceeds faster than when the evolution is performed only by the evolution processing without performing the learning (see FIG. 21). ).

As described above, in the evolution adaptation layer, the fuel consumption module and the power module constituting the lower module group evolve with priority on fuel consumption and engine output, respectively, and the control module constituting the upper module is provided by the user. By evolving to obtain the output ratio of the lower module according to the user's preference, it is possible to obtain the optimum target air-fuel ratio pattern according to the user's preference, and the engine 1 is adapted to the user's preference. The trainees will be trained to focus on fuel efficiency, drivability, or balance. In addition, by making the evolution adaptive layer function at regular time intervals, the training follows changes in the user's preference and changes in the engine and the vehicle over time.

(Others) In the above-described embodiment, the control module is divided for each function of acceleration and fuel consumption. However, this may be performed for control output such as fuel injection amount and ignition timing. In this case, for example, when the control of the intake pipe length is newly enabled, there is no need to change the existing control module, and there is an effect that comprehensive control can be realized by adding the intake pipe length control module. The control output for controlling the engine may be, for example, the electronic throttle opening, intake / exhaust valve timing, valve lift, intake / exhaust control valve timing, etc. in addition to the above (see FIG. 3). here,
The intake control valve is a valve provided on the intake pipe for controlling tumble and swirl, and the exhaust control valve is a valve provided on the exhaust pipe for controlling exhaust pulsation. Further, in the present embodiment, the fuel economy module and the acceleration module are alternately evolved for each generation by the evolution processing in the genetic algorithm.
This evolution processing is not limited to the present embodiment. For example, first, the acceleration module is fixed, the evolution processing of the fuel economy module is performed to the last generation, the optimal control module is acquired, and then the fuel economy module is optimized. And processing may be performed to evolve the acceleration module. Further, in the present embodiment, the control module is divided into a fuel consumption module and an acceleration module, and the evolution processing for the genetic algorithm is performed for each module. Without being limited to the embodiment, one type may be used,
Also, three or more types of control modules may be used. Further, in the present embodiment, in the genetic algorithm, the evolution is terminated when the evolution reaches a predetermined final generation, but the timing of terminating the evolution is not limited to the present embodiment. Alternatively, the process may be terminated when the evaluation value of the evolving individual changes by a predetermined change width or more.
Furthermore, in the present embodiment, the driving state index P for estimating the driver's preference and driving state is estimated using a neural network based on the distribution pattern of the gear position and the maximum rotation speed. Means for estimating the traveling state index P may be any method without being limited to the present embodiment.
For example, the estimation may be performed using fuzzy inference. When the running state index P is determined using the fuzzy inference, for example, "If the maximum speed of the first speed is large and the maximum speed of the second speed is large, the running state index P is very large." Such IF-THEN rules are described, and fuzzy inference is performed (see FIG. 22). In the case of a neural network, the coupling coefficient is determined by learning from a teacher signal, and thus becomes a black box. On the other hand, the method using fuzzy inference enables a knowledge-based approach at the time of design. Further, in the present embodiment, the preference of the user is evaluated from the distribution pattern of the gear position and the maximum number of revolutions. However, the parameter for evaluating the preference of the user is not limited to the present embodiment, and any parameter may be used. In, the physiological indicators of the user, for example, a means for detecting pulse, blood pressure, body temperature, brain waves, etc., provided in the user's equipment, for example, in the case of a motorcycle, a helmet, gloves, or boots, etc. Alternatively, a physiological index may be detected and evaluated based on this. further,
These physiological indices can also be used to evaluate the condition of the user (driving condition of the driver). Further, in the present embodiment, the user's preference is evaluated and training is performed in accordance with the evaluation. However, parameters for determining the training policy are not limited to the present embodiment. May be evaluated, and training may be performed in accordance with the skill of the user. In this case, as the parameters for evaluating the skill of the user, for example, in the case of a vehicle, the inclination angle of the vehicle, the vertical acceleration of the vehicle, the operation amount of the brake, the use ratio of the front and rear brakes, and the like can be considered. Furthermore, in this embodiment,
Although the learning layer is configured by a hierarchical neural network, the configuration of the control system of the learning layer is not limited to this embodiment,
For example, CMAC (Cerebellar Model Arithmetic Compu
ter) may be used. The advantages of using CMAC are:
Compared to the hierarchical neural network, the additional learning capability is excellent, and the learning speed is high. Furthermore, in the third embodiment of the present invention, the comprehensive evaluation process for each individual during the evolution process is time-divided, control using each individual is continuously performed for a predetermined number of cycles, and each individual is pseudo-parallel. By actuating each individual individually, it is possible to evaluate fairly without being affected by changes in running conditions etc.,
The evaluation method for each individual is not limited to this embodiment. For example, an evaluation area is subdivided to calculate a local evaluation value, and a comprehensive evaluation is performed using a local evaluation value common to all the individuals. Alternatively, weighting may be performed for each of the subdivided local evaluation areas, and the overall evaluation may be performed in consideration of the weighting. Specifically, for example, in the case of fuel efficiency evaluation, as shown in FIG. 23, the relationship between the engine torque and the engine speed during the evaluation process and the fuel consumption per unit time is determined for all individuals (in the case of FIG. 23, 23) (see FIGS. 23 (a) and (b)), and using this fuel consumption as a local evaluation value, a local evaluation value obtained under common conditions (FIG. 23 (c)). Can be evaluated for each individual with the average value of Further, as a method different from FIG. 23, for example, as shown in FIG. 24, a weight map in which each local evaluation portion is weighted in advance is prepared (FIG. 24).
(C), the relationship between the engine torque and the engine speed during the evaluation process and the fuel consumption per unit time is examined for all individuals (two in the case of FIG. 24) (FIG. 24).
(See (a) and (b)), the evaluation may be performed by determining a comprehensive evaluation value in consideration of a weight map for each individual.

[0023]

According to the integrated control system of the present invention described above, the control characteristics of the control system for controlling the control target can be changed in accordance with the characteristics of the user and / or the situation of use. The subject is "trained" to the characteristics that are appropriate for the user who uses it and / or the context of use, making it easier to use, and also training the user to control the characteristics of the controlled object to their own unique characteristics. This has the effect of giving consciousness fun. In addition, by automatically changing the control law in response to a change in the use environment or deterioration over time, an advantageous effect is obtained in which the optimum operation state can be realized in any case. Further, setting for obtaining the control parameters of the control object is not required, and the cost can be reduced. In addition, when an engine mounted on a vehicle is applied as a control target, by performing control in accordance with the driver's preference, it becomes possible to adjust the operating characteristics of the engine to the user's preference. By performing the control according to the skill of the driver, it is possible to obtain the traveling performance according to the skill level, and it is possible to provide the driver with a comfortable traveling. Furthermore, when an engine mounted on a vehicle is applied as a control target, the operating characteristics of the engine are trained in accordance with the purchaser's preferences after purchasing the purchaser. Does not restrict the selection range. Also,
By applying the auxiliary power of a bicycle or wheelchair with auxiliary power as the control target, the assist characteristics of the auxiliary power can be adjusted to the characteristics according to the user's preference, so the optimal assist characteristics for each user Can be obtained. Furthermore, when a robot is applied as a control target, it is possible to adjust the operation characteristics of the robot to characteristics according to the user's preference, so that the robot performs an optimal operation for each user. . Further, when a suspension or a seat is applied as a control target, it is possible to adjust a damper characteristic of the suspension or the seat to a characteristic according to a user's preference, thereby obtaining an optimal damper characteristic for each user. Becomes possible. Furthermore, if the steering system of the vehicle is applied as the control target, the steering control characteristics of the steering system can be adjusted to the characteristics according to the user's preference, so that the optimal steering control characteristics can be obtained for each user. It becomes possible.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a basic concept of an integrated control system according to the present invention.

FIG. 2 is a flowchart showing the overall control method shown in FIG. 1 over time.

FIG. 3 is a schematic diagram showing a relationship between an engine 1 and a control device 10 that executes the comprehensive control method.

FIG. 4 is a schematic block diagram of the control device 10.

FIG. 5 is a flowchart of a basic operation of the evolution adaptive layer in FIG.

FIGS. 6 (a) and 6 (b) are graphs each showing a distribution pattern of the maximum number of revolutions at each gear position within a predetermined time, and FIG. 6 (a) is a graph showing a sporty traveling;
(B) is a graph at the time of mild running.

7 (a) to 7 (c) are graphs showing distribution patterns of the maximum number of rotations at each gear position within a certain time, FIG. 7 (a) is a graph during normal running, and FIG. FIG. 4C is a graph when traveling on a road, and FIG. 4C is a graph when traveling on a highway.

FIG. 8 is a schematic diagram of a neural network for estimating a driving state index.

9 (a) to 9 (c) are graphs each showing a relationship between a running state index and an evaluation function.
(B) shows a state in which the time required to reach the maximum rotation speed is short, and (c) shows a state in which the time required to reach the maximum rotation speed is long.

FIG. 10 is a schematic diagram of a neural network constituting the fuel consumption module and the acceleration module.

FIG. 11 is a flowchart of the evolution of a fuel economy module by a genetic algorithm.

FIG. 12 is a diagram conceptually showing coding of a neural network.

FIGS. 13A to 13C are graphs each showing a relationship between a change in vehicle speed and an acceleration evaluation index at the same throttle opening.

FIG. 14 is a diagram conceptually showing the relationship between the evolution of a fuel economy module and the evolution of an acceleration module.

FIG. 15 is a diagram conceptually showing a state in which a teacher data set acquires new teacher data.

FIG. 16 is a diagram conceptually showing updating of a teacher data set.

FIG. 17 is a schematic block diagram of another embodiment of the control device that executes the comprehensive control method according to the present invention.

FIG. 18 is a diagram showing an evolution tendency of each control module.

FIG. 19 is a flowchart of the evolution of a fuel economy module by a genetic algorithm.

FIG. 20 is a diagram showing a state of a preliminary evaluation process of each individual by a time division method.

FIG. 21 is a diagram showing a graph comparing the evolution performance of evolution with only evolution processing and evolution in which evolution processing and learning processing are alternately repeated.

FIG. 22 is a diagram conceptually showing a state in which a running state index is estimated by fuzzy inference.

FIG. 23 is a diagram conceptually showing another method of fuel efficiency evaluation in the fuel efficiency module.

FIG. 24 is a view conceptually showing still another method of fuel efficiency evaluation in the fuel efficiency module.

[Explanation of symbols]

 Reference Signs List 1 engine 10 control device (second embodiment) 20 control device (third embodiment)

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁶ Identification code FI F02D 45/00 340 F02D 45/00 340H 340Z G06F 15/18 550 G06F 15/18 550E 550C

Claims (51)

[Claims] 1. A characteristic of a user and / or a use condition is determined, and a control characteristic of a control system for controlling a control target is changed in accordance with a characteristic of the user and / or a use condition based on the result of the determination. A comprehensive control method characterized by the following. 2. The comprehensive control method according to claim 1, wherein the characteristics of the user and / or the use situation are estimated using a neural network or a fuzzy rule. 3. The comprehensive control method according to claim 1, wherein the user's characteristics are at least one of user's preference, skill, operation pattern, and state. 4. The control characteristic according to claim 1, wherein the control characteristic is adaptively changed in accordance with a change in a use environment and / or a temporal deterioration of a control target. Comprehensive control method. 5. The control system includes a reflection layer as a lowermost layer,
The comprehensive control method according to any one of claims 1 to 4, wherein the integrated control method has a hierarchical structure having a learning layer on a layer above it and an evolution adaptive layer on the top layer. 6. The general control method according to claim 5, wherein a basic amount of control output is output from the reflection layer, and outputs of the learning layer and the evolution adaptive layer are correction amounts for the basic amount. . 7. The method according to claim 1, wherein the evolution adaptation layer includes at least one control module that behaves autonomously, and the control module is adaptively evolved by competing and / or cooperating with the control module. Item 7. The comprehensive control method according to item 5 or 6. 8. The total control system according to claim 7, wherein there are a plurality of said control modules, and said plurality of control modules cooperate and compete with each other to adaptively change said control system. . 9. A lower module group in which the control module determines a primary output based on a characteristic related to a control target, and a higher module determining a secondary output from the primary output of the lower module group based on a user and / or a use situation. The integrated control system according to claim 8, further comprising a module group. 10. The system according to claim 1, wherein the lower module group is evolved while being evaluated based on characteristics related to a control target, and the upper module group is evolved while being evaluated based on a user and / or a use situation. 9. The comprehensive control method described in 9. 11. The integrated control method according to claim 8, wherein a ratio of outputs of a plurality of control modules of the evolution adaptive layer is changed according to characteristics of a user and / or a use situation. 12. The control module according to claim 7, wherein the adaptive evolution of the control module is performed using a genetic algorithm and / or a multi-agent technique.
The comprehensive control method according to any one of the above. 13. The control module of the evolution adaptation layer is evolved by a predetermined number of generations by a genetic algorithm, and each time the learning result is learned by the learning layer, the control module is most evaluated among a plurality of individuals of the generation trained by the learning layer. 13. The comprehensive control method according to claim 12, wherein a new population including a population having a high number of individuals and a population having an output of the evolution adaptive layer of zero is generated. 14. The comprehensive control method according to claim 12, wherein an evaluation function in the genetic algorithm is automatically changed according to characteristics of a user and / or a use situation. 15. The comprehensive control method according to claim 14, wherein a relationship between the evaluation function and characteristics of a user and / or a use situation can be changed by a user's instruction. 16. The comprehensive control method according to claim 5, wherein a limit is provided for an output gain of the evolution adaptive layer. 17. The learning layer according to claim 5, wherein said learning layer has two neural networks for execution and learning.
17. The comprehensive control method according to any one of items 16 to 16. 18. The comprehensive control method according to claim 17, wherein the control characteristic obtained by the evolution of the control module in the evolution adaptive layer is learned by a learning neural network in the learning layer. 19. The learning layer has a learning data set for learning, and outputs only the sum of the outputs of the evolutionary adaptive layer and the neural network for execution of the learning layer that has been output within a certain period in the past. 19. The comprehensive control method according to claim 18, wherein learning is performed by a learning neural network in a learning layer by using as new teacher data, and teacher data in other areas is the same as before. 20. When the learning of the learning neural network is completed, the learning neural network functions as an execution, and the original execution neural network functions as a learning. The integrated control system of Kishia according to any one of claims 15 to 19. 21. The information according to claim 15, wherein information on the learned neural network in the learning layer is stored in an external storage medium such as a floppy disk or an IC card, and can be stored and read. The comprehensive control method according to any one of claims. 22. The reflection layer according to claim 5, wherein the reflection layer performs processing using any one of a mathematical model, a fuzzy rule, a neural network, a map, and a subsumption architecture. The comprehensive control method according to any one of claims. 23. The comprehensive control method according to claim 3, wherein the skill of the user is estimated from an external state quantity. 24. The comprehensive control method according to claim 2, wherein the state of the user is estimated using a physiological index. 25. The comprehensive control method according to claim 24, wherein the physiological index is at least one of a user's pulse, blood pressure, body temperature, and brain wave. 26. The comprehensive control method according to claim 1, wherein the control target operates actively. 27. The comprehensive control method according to claim 26, wherein the control target comprises an engine. 28. The engine for a vehicle, wherein the characteristics of the user are determined based on the driver's preference, skill, and / or state, and the characteristics of the usage are determined based on the running condition of the vehicle. 28. The comprehensive control method according to claim 27, wherein the operation characteristics of the engine are changed based on the control information. 29. Means for detecting a driving state of a vehicle are provided, and based on at least a part of the detection result, the driver's preference, skill, and / or characteristics of one or both of a state and a driving situation. 29. The comprehensive control method according to claim 28, wherein a state index suitable for the condition is estimated, and the operating characteristics of the engine are changed based on the state index. 30. The method of claim 30, wherein the state index is a neural network,
30. The comprehensive control method according to claim 29, wherein the estimation is performed using one or both of the fuzzy rules. 31. The comprehensive control method according to claim 29, wherein an evaluation function in a genetic algorithm for coordinating and / or competing control modules of the evolutionary adaptive layer is changed based on the state index. 32. A relationship between the evaluation function and the state index based on a time required to reach a maximum rotational speed in each gear position, a rate of change of an engine rotational speed, or a driver's input using an instruction input button. 32. The general control method according to claim 31, wherein 33. The comprehensive control method according to claim 29, wherein the means for detecting the driving state of the vehicle is means for detecting an engine speed and a gear position. 34. Control outputs for changing the operating characteristics of the engine include a fuel injection amount, an ignition timing, an electronic throttle opening, an intake / exhaust valve timing, a valve lift amount,
The comprehensive control method according to any one of claims 27 to 33, wherein intake and exhaust control valve timing and the like are used. 35. A technique for estimating a driver's skill from at least one of a driver's clutch operation speed, a vehicle inclination angle, a vertical acceleration of the vehicle, a brake operation amount, and a use ratio of front and rear brakes. The comprehensive control method according to any one of claims 28 to 34. 36. Means for detecting a physiological index of a driver is provided in an accessory worn by the driver during driving, and the state of the driver is estimated based on the physiological index obtained by the means. The comprehensive control method according to any one of claims 28 to 35, wherein: 37. The integrated control system according to claim 36, wherein the accessory is at least one of a helmet, a glove, and a boot. 38. The control object is an auxiliary power of a bicycle or a wheelchair using an electric motor or an engine as an auxiliary power,
27. The comprehensive control method according to claim 26, wherein a control characteristic of the control system is a characteristic relating to an assist characteristic of the auxiliary power. 39. The comprehensive control method according to claim 26, wherein the control target is a robot, and the control characteristics of the control system are characteristics relating to the operation characteristics of the robot. 40. The motion characteristic according to claim 39, wherein the motion characteristic is at least one of a robot path selection, an arm moving method, a moving speed, and a talking method.
General control method described in 41. The comprehensive control method according to claim 39, wherein the robot is a personal robot. 42. The comprehensive control method according to claim 1, wherein the control target operates passively. 43. The control target is a vehicle steering system,
43. The comprehensive control method according to claim 42, wherein the control characteristic of the control system is a characteristic relating to a steering control characteristic of the steering system. 44. The comprehensive control system according to claim 42, wherein the control object is a suspension or a seat of a vehicle body, and a control characteristic of a control system is a characteristic relating to a damper characteristic of the suspension or the seat. 45. The characteristics of the user are determined by the driver's preference, skill,
And / or changing the control characteristic of the control system based on a result of the determination based on a result of the determination based on a result of the determination.
4. The comprehensive control method described in 4. 46. A means for detecting a driving state of the vehicle, and based on at least a part of the detection result, the driver's preference, skill, and / or characteristics of one or both of the state and the driving situation. 46. The comprehensive control method according to claim 45, wherein a state index suitable for is estimated, and operating characteristics of the engine are changed based on the state index. 47. The method of claim 27, wherein the state index is a neural network,
47. The comprehensive control method according to claim 46, wherein the estimation is performed using one or both of the fuzzy rules. 48. The comprehensive control method according to claim 46, wherein an evaluation function in a genetic algorithm for coordinating and / or competing control modules of the evolution adaptive layer is changed based on the state index. 49. A technique of estimating a driver's skill based on at least one of a driver's clutch operation speed, a vehicle inclination angle, a vertical acceleration of the vehicle, a brake operation amount, and a use ratio of front and rear brakes. The comprehensive control method according to any one of claims 45 to 48. 50. Means for detecting a physiological index of a driver is provided in an accessory worn by the driver during driving, and the state of the driver is estimated based on the physiological index obtained by the means. The comprehensive control method according to any one of claims 45 to 48, wherein: 51. The comprehensive control method according to claim 50, wherein the accessory is at least one of a helmet, a glove, and a boot.