WO1996020438A1 - Procede et systeme d'inference mettant en ÷uvre un modele hierarchique et procede et systeme de gestion associes - Google Patents

Procede et systeme d'inference mettant en ÷uvre un modele hierarchique et procede et systeme de gestion associes Download PDF

Info

Publication number
WO1996020438A1
WO1996020438A1 PCT/JP1995/002562 JP9502562W WO9620438A1 WO 1996020438 A1 WO1996020438 A1 WO 1996020438A1 JP 9502562 W JP9502562 W JP 9502562W WO 9620438 A1 WO9620438 A1 WO 9620438A1
Authority
WO
WIPO (PCT)
Prior art keywords
case
strategy
action
input
goal
Prior art date
Application number
PCT/JP1995/002562
Other languages
English (en)
French (fr)
Inventor
Satoru Isaka
Original Assignee
Omron Corporation
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Omron Corporation filed Critical Omron Corporation
Priority to KR1019960704169A priority Critical patent/KR970700877A/ko
Priority to DE69532237T priority patent/DE69532237T2/de
Priority to EP95940434A priority patent/EP0749057B1/en
Publication of WO1996020438A1 publication Critical patent/WO1996020438A1/ja

Links

Classifications

    • GPHYSICS
    • G05CONTROLLING; REGULATING
    • G05BCONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/0265Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
    • GPHYSICS
    • G05CONTROLLING; REGULATING
    • G05BCONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00Computing arrangements using knowledge-based models
    • G06N5/04Inference or reasoning models

Definitions

  • This invention is based on the use of historical data obtained when a human manipulates or controls an object to form a model, particularly in a chain of linked chains.
  • the present invention relates to a system and a method for automatically operating or controlling an object using a model formed by the method.
  • the system and method according to the invention is useful for factory automation, especially for real-time dynamic control operations in manufacturing and transport facilities. . Background art
  • factory automation is to process signals from sensors and to control the actuators.
  • the goal is to increase production efficiency by developing an algorithm.
  • Many device algorithms are boring, intense, repetitive, and so on. And freed humans from dangerous work. Nevertheless, in manufacturing and other industrial fields, humans are still involved in the jobs that require the power of automation. Automating to increase productivity, lower costs, reduce waste, improve quality, improve work safety, and reduce training costs. We need to go further.
  • AI artificial intelligence
  • This invention is a new lab that has been extended to both rule-based apps (f) (rule-basedappr 0 ach) and case-based approaches. It provides the approach.
  • the operation or control of a target by a human is modeled, and according to this model, the operation or control of a human by a machine is controlled. To reproduce, it is based on the following principle of action.
  • the source of this invention is a record of actual human manipulation or control.
  • the goal A model with a hierarchical structure consisting of functions and actions is constructed.
  • the inference method operates or controls an object. It consists of a set of input and output variable values that are obtained by performing the following:-Acquisition of a series of action data, and the above-mentioned action data Focusing on the input variable values of the data, the goal for the above operation or control is created for each input variable, and the above-mentioned action ' Focusing on the output variable values of the data, create a strategy to achieve the above-mentioned goal, and create a new strategy for the above-mentioned goal and its goals.
  • the strategy that has not been reached is chained, and the strategy described above is chained with the action data that realizes it.
  • a hierarchy consisting of linked chains, strategies, and functions. It forms a model of the key structure.
  • a goal was created from the actual action samples and a strategy for achieving the goal was achieved. Is created. And it is strongly tied to the goal, strategy and action. Therefore, the directionality is defined at a conceptual level for operation and control (goal), and for that goal, it is the goal. Decision making in accordance with the situation is made in a strategy that is linked to the rules. Finally, the keys that are linked to the strategy The decision-making (determination of the output value) is made by the case.
  • rule-based inference including fuzzy rules
  • an expert in order to create a rule, an expert can interact with the knowledge and use that knowledge. It is necessary to work on rules.
  • a goal and a strategy are automatically created from past operation records.
  • the inference method according to the present invention is specifically defined as follows.
  • the inference method consists of a set of input and output variable values obtained by manipulating or controlling an object. Take the sample, and pay attention to the input variable values of the above-mentioned action samples, the first value, the last value, and the maximum value of the input variable values. A goal case defined using the minimum and maximum values is created for each input variable, and the input output variable value of the above-mentioned action sample is created. First, a range of input variables whose output variable values are kept the same is extracted, and for each range, a strategy that connects the output variable value to the input variable range is used. ⁇ Create a case and store it with an input variable range that does not satisfy the above conditions of the goal case.
  • Chain the action case to its strategy case In this way, it is composed of a chain case, a strategy case and an action case, which are linked to each other. It forms a model of a hierarchical key structure.
  • the captured function sample is converted to binary data of a predetermined digit.
  • the steps for creating a goal case are the steps for creating a temporary goal case and for creating a goal case. Steps to output the completed goal case, and after the evaluation by the user, the goal and the final goal It consists of the steps that make up the case.
  • the user evaluation described above includes the addition, deletion, and modification of golf cases.
  • the above step of creating a strategy case is more specifically a sequential comparison of adjacent output variable values arranged in chronological order. And, when there is a change, storing the output variable values before and after the change to create a strategy sample.
  • Refer to the function sample above for the sample sample and store it in the format of d0 " ⁇ when-until ⁇ ⁇ min"-max. It consists of a step for converting to a lattice case, and a step for optimizing the above-mentioned strategy case after conversion.
  • An example of a step that optimizes the above strategy case is an example of two or more strategy cases with the same output variable values. It reduces the number.
  • the linkage between the strategy case and the goal case is concretely established in the memory and in each strategy case.
  • the linkage between the action case and the strategy case is, concretely, on a memory and each action case is To the storage case set up in the storage case and the function case set up in each storage case. It is realized by a pointer to
  • the action case consists of exactly the same set of input and output variable values, the resulting action sample, and one of the others. Created by deleting.
  • the model created in this way is validated using the above series of action samples as input / output data.
  • this verification is performed in such a way that the above-mentioned Action 'samples are fetched one by one in the order of time, and the fetched Action's samples are taken.
  • the storage case From the list of stored strategy cases, with the conditions that are satisfied by the above input variable values. Search the lattice case, and list the action case linked to the searched single-strand case. In the listed action case, the one with the input variable value that matches or is closest to the input variable value captured above. ⁇ This is done by outputting the output variable value of the case.
  • the above model is verified by the user narrowing down the output value and the output variable value of the function sample.
  • a method for manipulating or controlling an object using a model created by the above-mentioned five-volume method is also provided according to the invention. It has been.
  • the input data obtained from the target is sequentially acquired, and the goal is to satisfy a condition that is not satisfied by the acquired input data.
  • Search for the case and list the strategy case that is linked to the searched golf case. From the strategy case, the condition that is satisfied by the above-mentioned input data.
  • Searches for a financing case that has a case and lists the action cases that are linked to the searched stratagem case. And match the captured input data above in one of the listed action * cases above. Outputs the output variable value of the action case with the nearest input variable value, and repeats the above process each time the input data is taken.
  • the present invention also provides a system for executing the above-described inference method and control method.
  • the inference system is composed of a series of functions including input and output variable values obtained by manipulating or controlling an object.
  • Means to capture and sample the input variables, paying attention to the input variable values of the above-mentioned action * samples, and input variable values The method of creating a goal and case defined for each input variable using the first value, last value, maximum value, and minimum value of Focusing on the output variable values of the sample samples, extract the range of the input variables that keep the output variable values the same, and extract the output variable values for each range.
  • the systems for operating or controlling objects are:
  • Figure 1 is a block diagram showing the configuration of the entire system including the control target in the model formation mode.
  • Figure 2 includes the control target in the problem solving mode
  • FIG. 2 is a block diagram showing the configuration of the entire system.
  • FIG. 3 is a block diagram showing a configuration example of a system according to the present invention using a hierarchical key model.
  • FIG. 4 is a perspective view showing a clean control system as an example of a control target.
  • Fig. 5 shows another configuration example for detecting the deflection angle.
  • FIG. 6 is a block diagram showing a system in which a simulation of crane control is provided.
  • Figure 2 shows the variables in the simulation o
  • FIGS. 8a and 8b are flow charts showing the processing procedure in the model formation mode.
  • Figure 9 shows an example of the input and output data obtained in a simulation of crane control.
  • Fig. 11 shows the simulation data of the crane control, and the resources for creating the gore-sole case created for each input variable.
  • the cost is an example of the cost.
  • Fig. 12 shows an example of a golf case for crane control.
  • Fig. 13 is a schematic diagram of the strategy sun in crane control. Here is an example of a pull.
  • Figures 14 to 16 show a simple example to illustrate the conversion of a strategy sample into a storage case.
  • Fig. 14 is an example of an action sample
  • Fig. 15 is an example of a strategy sample
  • Fig. 16 is an action sample. Each of which represents a sample graph.
  • Fig. 17 shows a graph of the action sample in crane control.
  • Fig. 18a and Fig. B show examples of the strategy case in the crane control.
  • - Figure shows the binary string of the stratified case of Figures 18a and 18b
  • Figure 2G shows the linkage of each golf case to the strategy case. .
  • Fig. 21 shows the linkage of each strategy case to the goal case and the action case.
  • Figure 22 shows the chain of each action 'case' to the "staff"--case.
  • Fig. 23 shows a hierarchical key structure consisting of a goal, a strategy, and an action.
  • FIG. 24 is a flow chart showing a processing procedure in the problem solving mode.
  • Fig. 25 shows the input and output data in the problem solving mode.
  • the system according to the invention which uses a hierarchical model, has two modes. One is the model formation mode and the second is the problem solving mode.
  • control target is actually controlled according to the model that has already been formed.
  • Fig. 1 shows the configuration of the whole system including the control target in the model formation mode.
  • the system according to the present invention is embodied as a model formation system (or a learning system).
  • the amount of control in the controlled object (controlled system) 10 is generally detected by sensors 11, 12,... 1, o.
  • the controlled variable (input data X i, X 2 ,..., X
  • the operating device 20 uses the force to observe the status of the controlled object and, if necessary, the value of the displayed control amount (for example, the position of the needle of the instrument). (Confirmation of the operation), and operate (operation) so that the controlled Turn).
  • the manipulated variable y output data (not limited to one type) output from the operating device 20 is given to the actuator 21 and the actuator 21. In the evening 21, the controlled object 10 is actually driven or driven. All or some of the control amounts X ⁇ ,..., X k may not be provided to the operating device 2Q.
  • the controlled variable is a theoretical value (or a calculated or estimated value) that will be obtained from the controlled object 1Q.
  • Fig. 2 shows the configuration of the entire system including the control target ⁇ in the problem solving mode.
  • the system according to this invention is embodied as an automatic control system (automatic operation system) .1B.
  • Self-Dosei your S ystem 1 B was the model Dell that created the control amount ⁇ x k of the back down support 11 to control object 10 that will be detect in Tsu by the lk to capture. Destination By virtue of this, the operation amount y to be given to the controlled object 1Q is calculated.
  • the manipulated variable y drives the control target 10. Provided to the actuator 21 (and the actuator 21 so that the output of the actuator 21 becomes the calculated value). Is controlled). As a result, the controlled object 10 is automatically controlled in the same manner as the operator controls (operates) the controlled object using the operating device 2G. It is controlled (operated) by the control system 10B.
  • the key-key model according to the present invention is embodied as a model forming system or an automatic control system.
  • An example of the configuration of the used system is shown in FIG.
  • Computer 30 is what it is. In the evening, you can use other general-purpose computers, large or small.
  • the computer 3Q stores a program for executing a process in the model forming mode and the problem solving mode. And stores data (including data that represents the model) that is generated or used in these processes.
  • Memory is set up.
  • These memories 31 and ⁇ are realized by ROM, RAM, hard disk, floppy disk, optical disk, and the like. It may be realized by the hard disk as well as the memory 31, or the memory may be the ROM and the hard disk. The memory may be realized by the RAM and the hard disk, respectively, by the disk or the floppy disk. No. Here, memories 31 and 32 are conceptualized according to the evening species in which they are stored.
  • the input device 33, the display device 34 as an output device, and the printer 35 are also connected to the computer 3Q. Yes.
  • Input device 3 3 includes keyboard and mouse.
  • the display device 34 that displays the menu screen may be considered as a part of the input device 33.
  • the output device may be a display device 34 and / or at least one of the printers 35.
  • control amount in the control target 1 Q which is easily adjusted by the sensors 11 to I k (usually an AZD converter that converts the amount into digital data
  • the operation amount (in the case of analog amount) output from the operation device 20 is converted to digital data.
  • the amount of operation (usually digital) that is calculated in the AZD converter, and the problem solving mode and given to the actuator 21 Is connected to a D / A converter 38 which converts the amount of the AZD converter into an analog output as required.
  • D / A converter 38 which converts the amount of the AZD converter into an analog output as required.
  • an interface for input of control and manipulated variables and output of manipulated variables (differentiating circuit, etc., described below) is provided in the combo 3G as needed. Will be set up.
  • FIG. 4 shows a crane control system as an example of a control object. For simplicity, this clean system contains only one rail.
  • the book rail 50 is installed horizontally at an appropriate height.
  • a trolley (trolley) is supported by the wheels 50 by the wheels 41 on its own.
  • the wheels 41 are rotationally driven directly by a traveling motor 42 or via a reduction gear.
  • the trolley 40 is also equipped with a hoist.
  • the hoist includes a bridge 44, a hoist motor 46 for driving the pulley, and a wire 47 wound around the pulley 44.
  • the pulley 44 is fixed to a rotating shaft 45 received by the bracket fixed to the trolley 40, and the rotating shaft 45 is connected to the rotating shaft 45.
  • the motor 46 is driven to rotate directly or through a reduction gear.
  • At the lower end of the wire 47, a hook 48 is attached.
  • a suspended load 49 is applied to 48.
  • the suspended load 49 is connected to one point on the lane 50 and to another point. It is transported by running the rifle 40. At this time, the swing angle (the swing angle in the direction in which the trolley moves) of the wire 47 is set to be as small as possible (the predetermined angle).
  • the running control of the trolley 4Q is performed so that it is within the range of the maximum permissible vibration.
  • control amount As the control amount, the position of the trolley 40, the swing angle, and the differential value of the swing angle are employed.
  • the position of the exit 4Q is taken from the departure point A as a starting point and toward the end point B in a positive direction. This position is detected by a rotating transducer (not shown) provided on the rotating shaft of the motor 42 or the shaft of the wheel 41. .
  • the pulse train (or digitizing data indicating the position) output from the rotating transducer is stored in the computer 30 (A / D converter will not be necessary). If necessary, a source inspection device could be installed to detect when trolley 4Q is at departure point A.
  • the deflection angle of the wire 47 is attached to the trolley 40, and the television camera is arranged so as to capture an image of the wire 47.
  • La 51 and the camera power are detected by a processing device that performs image processing on the video signal.
  • the image processing device can be realized by a part of the computer 30. In this case, the video signal of the camera 51 is taken into the computer 30 through the AZD converter 36. Image processing for detecting the angle of a linear object with respect to the perpendicular is performed by a known method. 9
  • the derivative of the deflection angle can be calculated by the computer 30 o
  • the shaft A is fixed to the shaft 45 at its upper end.
  • the shaft 45 is rotated by the bracket and supported by itself o
  • the lower end of the shaft A is a hook 48.
  • the rotation angle of the shaft 45 is It is detected by the sensor 52. If the output signal of the angle sensor 52 is an analog signal, the signal is converted to a digital signal by the AZD converter 36.
  • the output signal of the angle sensor 52 is differentiated by a differentiating circuit, and the differentiated signal is supplied to the AZD converter 36.
  • the differential value of the deflection angle is taken into the computer 30 and the differential processing can be performed by the computer 30 as well as the differential processing. O Output of angle sensor 5 2 In the case of the amount of digital leakage, it is taken as it is at 30 pm.
  • the manipulated variable is determined by the degree of transfer of the mouth 40. 0
  • the transfer speed can also be expressed as a voltage or current applied to the running motor 42.
  • the system of the present invention can be applied to a sensor for detecting a control amount and to a control object without an operation unit to which an operation amount is given.
  • it is a forecast system for stock prices.
  • the input data is today's stock price (for each stock, ), The average stock price of the day, etc., which may be input from a keyboard, communication device, floppy disk drive, or the like.
  • the output data will be the sales decisions, buying decisions, and quantities. These would be output from a display or printer.
  • the input data is the day of the week, weather, stock, etc., and is input from a keyboard, communication device, floppy disk drive, etc.
  • the output data is forecast sales, purchase quantities, etc., and is output from display devices and printers.
  • FIG. 6 shows the minimum structure of the system according to the present invention (of course, the output device is the display device). Or one of the printers).
  • m is the mass of the suspended load
  • t is the length of the loop
  • g is the acceleration of gravity
  • V is the motor speed
  • V is the target speed of the night, which is the amount of operation given by the operator in the siren. That is, V d takes the following value
  • V Q ⁇ 40 (m / min) 0, 40 (m / min) ⁇
  • the model consists of three levels: gonorré (objective or target), strategy (strategy), and action (execution, action). It has a hierarchical key structure. This structure is based on the following principle.
  • Every action has at least one strategy associated with it, and every strategy has that strategy. There is at least one goal related to it o J
  • the operator is actually operating the controlled object 10 using the operating device 20 Operation (driving)
  • the control amount (input data) obtained from the sensors 11 to lk and the operation amount (output data) given to the actuator 21 are controlled.
  • Sampling is performed at regular time intervals, and is taken into the computer 30.
  • These input and output data are saved as data in teacher's memory 32.
  • Input and output data are actual operation records.
  • FIG. 1 An example of the input and output data obtained in the simulation of the above-mentioned claim control system is shown in FIG. .
  • Input and output data are sampled every 0.3 (sec).
  • the crane is operated by three types of operations: forward, reverse, and stop.
  • the operation output (travel speed of the trolley: output data)
  • u +40 (m / min)
  • u —40 (m min)
  • u +40 (m / min).
  • the input data consists of the trolley position X (m), the wire swing angle ⁇ (rad), and the swing angle change rate (differential value) d S (rad / sec). ) (1 ! 9 / / (11 is hereinafter referred to as 10).)
  • the time zone indicated by T1 is the time when the trolley is started, and the operation output u is set so that the swing angle is as small as possible. It has been adjusted.
  • the trolley is operating at a constant speed.
  • Time indicated by T 3 The operation output u of the belt is adjusted to stop the trolley at the target point B.
  • time zone T4 the trolley almost reaches the point B, and the swing angle ⁇ of the suspended load gradually decreases.
  • Input and output data is binarized and converted to form a binary string.
  • Each binary string is called an action sample.
  • Input data X, ⁇ , d ⁇ , and output data u are each composed of 8 bits.
  • the minimum value of each input and output variable is set to 0 0 0 0 0 0 0 0.
  • the maximum value is converted to 1 1 1 1 1 1 1 1 1 1 respectively.
  • the minimum value at position X is 1 (m), and the maximum value is ⁇ (m).
  • the minimum value of the deflection angle is 1 Q.1 (rad), and the maximum value is + 0 ⁇ 1 (rad).
  • the minimum value of the differential value d is 0.1 (rad / sec), and the maximum value is + G.1 (rad / sec).
  • the minimum value of the moving speed u is 1-40 (m / min), and the maximum value is +40 (mZmin).
  • it is represented by 0 0 0 0 0.
  • 'Initial coding has special significance when the processing program of the model formation mode is written in a high-level language.
  • the first reason for initiating the encoding is that the resolution of the action sample can be easily changed. is there . For example, by discarding the lower 8 bits of 16-bit data, it is possible to obtain 8-bit data.
  • the second reason is that the most suitable representation method of the genetic algorithm, which is one of the optimization algorithms, is binary. ⁇ It is a string.
  • the maximum and minimum values of input variables 0 and d0 are surrounded by solid rectangles.
  • the maximum and minimum values of the input variable X are the final and initial values enclosed by the dashed rectangle.
  • Input variables whose maximum value and minimum value are equal are those that do not change at all in the operation to be controlled, and have no effect on the operation or control. Because it is not given 0
  • Goal condition for that input variable is expressed as 0
  • the fact that the first value and the last value of the input variable X i are different in the actual operation to be controlled means that this operation sets the input variable X i to (the first value ) To (final value) from the intention that it was done (by the operator). is there . Then, the fluctuation range of the input variable at that time is (minimum value).
  • the goal (control purpose) for the input variable X i is to change the value of the input variable X j from the (initial value) to the final value.
  • the fluctuation width of Xi falls between (the minimum value) and (the maximum value). That's what we do.
  • the first value and the last value are different for the input variable X. Then, the following goal condition is temporarily set for the input variable X.
  • the first value and the last value are the same, and both the first value and the last value lead to the goal at the O port. I can't give it out. In this case, it is presumed that the intention was to keep the input variables within a certain range during the operation process. This is the meaning of the goal condition described above.
  • the initial value and the final value are the same regardless of the deflection angle 0 and the derivative value d ((where: 0.000).
  • the following goal conditions are temporarily set for the input variables ⁇ and d S.
  • the results of the substeps 103-3, 103- and 10-5 described above are printed by the printer 35 or are displayed on the display device 34. Displayed 0
  • the goal condition described above is the binary array in the following array. It is encoded in the ring.
  • the hypothesis of the goal printed by the printer 35 or displayed on the display device 34 (the above-mentioned temporarily set goal 1) The user can evaluate them and add new or new ones as necessary, or delete any one of them. Or modify the goal conditions.
  • Goal 2 is unnecessary. Will be removed. Goal 2 is the input variable X
  • goal 1 and 2.3 of the first goal 1, goal 2 and goal 3 are "goal case numbers (goal numbers)", which are represented by 2 bits.
  • X, ,, d0 are the types of “input variables” that are also represented by 2 bits.
  • X is 0
  • S is 0
  • d 0 is 10 o
  • the third, “Evening Flag”, indicates the value of the “final state” or the width of the “transient state”. -This evening flag is 2 bits, and the value of "final state” is represented by 01, and the value of "transient state” is broadly represented by 10.
  • FIG 13 shows the finally obtained strategic sample.
  • the time and the action sample are also shown for reference.
  • the value at time 14.4 Q is different from the value at time H.70.
  • the output variable values at these times are stored.
  • the value at time 15.QQ and the value at time 15.3 15. are different, and the value at time 15.3Q and the value at time 15.6G are different.
  • the values at time 15.6Q are different from the values at time 15.90, so these values are stored.
  • the values at time 15.9 G and the time at time 16.2 Q are the same, so these values are not stored.
  • the strategy sample extracted in this way is referred to the action sample by referring to the strategy sample having the following format. - ⁇ Convert to a case.
  • Maker d 0... hen... until... min... mai... ”-do is followed by the value of one or more output variables.
  • whe 11 contains one or more input variable values. The value of the input variable follows after until. after min Is the minimum value for each input variable, and the maximum value for each input variable is arranged after maI.
  • Figure 14 shows a simple example of an action sample.
  • a strategy sample created based on the action sample of FIG. H is shown in FIG. In Fig. 14, focusing on the output variables, when the output variables change, the values of the output variables before and after the change are extracted.
  • the strategy sample shown in the figure is obtained as explained in step 1Q5.
  • Figure 16 shows a graph of the action sample in the input variable space.
  • the two orthogonal axes are the input variables ⁇ , X 2 .
  • the range in which the output variable u maintains a constant value is indicated by a rectangular box.
  • the range in which the output variables show a constant value is represented by a rectangular parallelepiped.
  • this range is represented by a hyperbox.
  • the input variable space is two-dimensional or three-dimensional, it includes rectangles and cuboids. For all cases, use the hyperbox.
  • the hyperbox Q is the output variable u
  • the range is 100.
  • the set of input and output variables that define the two hyperboxes is called the strategy case. Taking the case of the strat-one case Q as an example, this can be expressed as follows.
  • a graph of the action sample in the input space is shown in FIG.
  • a two-dimensional input variable space in which the position X and the deflection angle S are two axes orthogonal to each other is shown (of course, X, ⁇ , and d0 are It should be drawn in a three-dimensional input variable space, which is the axis orthogonal to).
  • the binarized code is represented by 8 bits for each variable value.
  • the order is X, ⁇ , the input variable value of d0 in the wh en ⁇ term, the input variable value of x, ⁇ , d0 in the until term, and the min term.
  • the storage cases 1, 2, 3, 4, 7, and 8 are not rectangles (boxes), but are represented by single points. (This point can also be included in the concept of hi-boxes).
  • two or more strategic cases For example, two or more strategic cases. -If the box has a completely overlapping box, remove one of them, leaving one of them.
  • the distances of two or more hyperboxes are calculated, and if the distance obtained is less than a predetermined threshold, these distances are calculated. Merge (or fuse) the strategy cases or delete one or the other.
  • the distance may be a point in the input variable space represented by the input variable values described in the when (or unU1) section, for example. Focus on the MIN point or the MAX point), and use the distance between these points in the two cases.
  • Substep 1 08-1 For the first strategy case (see Fig. 18a or Fig. 19), the state quantity of each case Compare with rule conditions.
  • the state quantity is the input variable value of the when term (or the input variable value of the until term) in the strategy case.
  • the first strategy case is associated with the goal case. . If the goal conditions are met, no association will be made. When the goal conditions are met, the strategy for the goal is no longer necessary.
  • Sub-step 1 (Repeat the processing of U-1 and 108-2 for all strategy cases.)
  • the value of the input variable d0 described in the when section of the strategy case 1 is O.QQQ.
  • Strategy cases 2, 3, 4, 5, 6, 6, 8, 9 and 10 are associated only with goal case 1.
  • Strategy case 7 is the same as golf case 1 and
  • the power of the goal of the golf case is as follows.
  • Goal case 1 is linked to strategy cases 1, 2, 3, 4, 5, 6, 7, 8, 9, and 10.
  • Goal Case 2 is not linked to any Strategy Case.
  • each strat- egy case has a link to all the goal cases linked to it. O (each of these pointers is stored in correspondence with the data representing each strategy case). The case has pointers to every strategy case linked to it.
  • FIG. 20 shows the pointers provided in each of the golf cases.
  • the PL indicates that it is a pointer to the strategy case, and the number indicated by the arrow following the PL is the number case. This is the number of the strategy case that is linked to The pointers PL are set for each gozo * case number.
  • No. 21 shows the pointers set up corresponding to each strategy case number.
  • the pointer PH is the pointer to the goal case
  • the pointer PL is the pointer to the function case (described below. ).
  • the number indicated by the arrow following PH is the number of the goal case linked to the strategy case. 0, the arrow following PL.
  • the number indicated by the mark is the number of the action case that is linked to that strategy case. is there .
  • Such a chain of strategy and gonorré cases is displayed or printed as needed. Users can add new storage cases, delete unnecessary storage cases, or delete existing storage cases as needed. Or modify the strategic case of the
  • a strategy case and an action case are related (or linked or chained) as follows.
  • An action case is a set of input and output variable values at each point in time, and is almost synonymous with an action 'sample. . However, the action samples with exactly the same values are removed from the action case.
  • the state quantity is calculated for each strategy case. Compare with the strategy condition.
  • the amount of deformation refers to the set of input variable values in the case.
  • a strategy condition is defined as the input variable value described in the until section from the set of input variable values described in the section of wh en in the strategy case.
  • the range up to the set of force variable values hereafter referred to as the range of wh ei! To until).
  • the first action case is associated with that strategy case. Guessing.
  • a pointer to the chain of action cases is also provided. This is what is indicated by PL in Figure 21.
  • the number indicated by the arrow of the pointer PL provided for each strategy case number is linked to that strategy case. This is the number of the action case to be chained.
  • the model consisting of the golf case, strategy case and action case force is shown in Fig. 23. It has a hierarchical structure, and it can be seen that these are related to each other. Gonoré (case) is at the highest level (conceptual decision-making level). The strategy (case) is at the middle level (decision-making level according to the situation). Action (Case) is the lowest level (level of empirical decision).
  • Step 1 1 0
  • Models created in this way will be validated, if necessary.
  • the above-mentioned action sample (teacher data; Fig. 9 or Fig. 10) is used.
  • the input data of the function sample follow the same procedure as in the problem solving mode described later (step 711-21). You will not necessarily need to capture and encode the input variable values), but the output will be calculated according to the model created earlier.
  • the output data obtained is compared with the output data in the function sample. If the two outputs are equal or close to each other (for example, the difference between the two is less than a predetermined threshold), the model is created correctly. Let 's see what was done 0
  • the input data may be generated at random, and the input data may be supplied to the system.
  • the system calculates and outputs the output data according to the model.
  • the power to determine whether the output data is valid is determined by the user.
  • the data memory representing the created model is transferred from the working area to the file area and saved. It is done.
  • the obtained input data is compared with the goal conditions of the goal 'case included in the model. If there is a goal case that has a goal condition that is not satisfied by the input data, it is linked to that goal case. List up the strategy case.
  • the —system attempts to make conceptual-level decisions by determining the right goal case. ing . This will determine the appropriate strategy case.
  • the input data is transferred to the drive selected in steps 1 and 2 above. Compare with the case. If there is a strategy case that has input data and a matching strat- egy condition (range of ff hei! To until), that strat- egy is considered. Actions that are linked to the stage case 'List up the case o
  • any of which have input variable values that exactly match the input data Selects the input variable value closest to the input data (the one with the shortest distance in the input variable space) as a candidate.
  • the output variable value of the candidate action case is defined as the output data. And output it. If two or more candidates exist, the best one is selected and the output variable of that action case is taken as the output data. And output it.
  • the hyperboxes of the two strategy cases partially overlap, and the input data is entered into this overlapping space. Data is included, there are two Different output variable values are derived from the strategy case. In this case, for example, a hyperbox that is closer to the distance between the input data and the end point (MIN or MAX point) of the hyperbox is used. The output value obtained from this is selected. Alternatively, the first output value obtained in the processing is unconditionally selected.
  • Fig. 25 shows the model of the crane control system, and the input data and model obtained during automatic operation according to the model.
  • Output data (used for crane control), and the number of the case selected in the process of obtaining the output value. Is shown.
  • the three numbers in the rightmost key (connected by arrows) are the previous key, the selected goal case number, and the stratum. Indicates the case number and the action case number c
  • These input data are The Gore condition of Goal case 1 is not satisfied, and the Gore condition of other Goal cases 2 and 3 is satisfied (see Fig. ⁇ ). Therefore, the ten strategy cases 1.2, 3, 3, 4, 5, 6, 7, 8, 9 and 10 linked to goal case 1 And 10 are selected (see Fig. 20) (step 1).
  • Strategy Case 1 Only Strategy Case 1 has a strategy condition that satisfies the input data (see Figure 18a). Then, the case that is linked to the strategy case 1 is selected (see Fig. 21). 1 2 3) o
  • steps 121 to 125 in FIG. 24 is repeated every time input data is taken from the control target (every fixed time).

Landscapes

  • Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Automation & Control Theory (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Feedback Control In General (AREA)

Description

明 mu 害 ヒ エ ラ ル キ ー · モ デ ル を 用 い る
推論 シ ス テ ム お よ び方法 な ら び に
制御 シ ス テ ム お よ び方法 技術分野
こ の 発明 は, 人間が対象を操作 ま た は 制 御 し た と き に 得 ら れ る 履歴デ ー タ を用 い て モ デル を 形成す る , 特 に 相 互 に 連 鎖 し た ゴ ー ル , ス ト ラ テ ジ ー お よ び ァ ク シ ョ ン カ、 ら 構成 さ れ る ヒ エ ラ ル キ ー 構造 を も つ モ デル を 形成す る 推論 シ ス テ ム お よ び方法, な ら び に 形成 し た モ デ ル を 用 い て対象を 自 動的 に操作 ま た は制御す る シ ス テ ム お よ び方法 に 関す る 。
こ の 発 明 に よ る シ ス テ ム お よ び 方 法 は 工 場 の 自 動 化, 特 に 製造設備お よ び輸送設備 に お け る 実時間動的 制御操作 の た め に有用 で あ る 。 背景技術
工場 の 自 動化の 目 的 は, セ ン サ か ら の 信号を 処理 し た り , ァ ク チ ユ エ 一 夕 を制御す る た め に , 電気 — 機械 装置や コ ン ピ ュ ー タ · ア ル ゴ リ ズ ム を 開発す る こ と に よ っ て生産効率を高 め る こ と に あ る 。 多 く の 装置ゃ ァ ル ゴ リ ズ ム が, 退屈 な , 激 し い, 繰返 し の 多 い , そ し て危険 な 仕事 か ら 人間を解放 し て き た。 そ れ で も , 製 造, そ の 他の 産業分野 に お い て, 自 動化力 望 ま れ る ベ き 仕事 に 依然 と し て 人間がかかわ つ て い る 。 生産性 を 高め , コ ス ト を 低下 さ せ, 無駄を省 き . 品質を 向上 さ せ, 労働 の 安全性を 改善 し , 訓練 に 要す る 費用 を 低減 さ せ る た め に 自 動化 を さ ら に お し 進 め る べ き で あ る 。
%Λ業社会 に お い て は , よ り 品質の 高 い 製品 を よ り 効 率 よ く 生産す る , よ り 洗練 さ れ た工場の 自 動化手段が ¾■ め り れ て い る 。 そ の よ う な手段の一つ は 人 間 の 操作 の 自 動化 モ デ ル を形成す る 装置で あ る 。
人 間の知識が関係す る 人工知脳 ( A I ) の-分野 に お い て, そ の モ デ ル を どの よ う に構築す る か と い う モ デ リ ン グ に つ い て多 く の ア プ ロ ー チ が あ る 。
最 も 一般的 な ア ブ ロ ー チ は エ キ ス パ ー ト · シ ス テ ム で あ る 。 そ れ は イ ン 夕 ビ ュ ー を通 し て 集 め ら れ た ェ キ ス パ 一 ト の 知 識 を - - 群 の ル ー ル で 表 現 す る も の で あ る 。 一群の ル ー ル は 論理的推論ェ ン ジ ン に よ つ て 処理 さ れ, そ れ に よ つ て説明, 予測 ま た は 示 唆が生成 さ れ o affi理的推論 メ 力 ニ ズ ム は フ ア ジ ィ · ロ ジ ッ ク を 含 む 種 々 の 方法 を も つ て い る 。 一般 に ル ー ル は知識 を 概 念 ま た は 状況の レ べ ル で表わ す。
ね つ と デ ー タ 志向性 の 強 い ア プ ロ ー チ は事例 ベ ー ス 推 a冊 、 c a s e — a s e d r e a s o n i n g ) と 呼 ば れ る も の で あ る 。 そ れ は事例 ( ケ ー ス ) を記憶 し , そ れ を 比較す る 学習戦略で あ る 。 す な わ ち , メ モ リ に 記憶 さ れ た 先の 同様な 状況 と 比較 し かつ 対比す る こ と に よ っ て新 し い 状況を解決す る 。 そ れ は新 し い状況を認識 し , 新 し い 推 論 を 行 い , 推論 が 実 行 で き な い と き は 質 問 を 発 す る 。 一般 に 事例 は知識 を操作 レ ベ ル で表 わ す。 明 の 開
こ の 発 明 は , ル一ル · ベ ー ス · ァ プ ロ 一 ■f ( r u l e— b a s e d a p p r 0 a c h ) と 事例 べ一 ス · ァ プ ロ 一 チ の 両方 に 広げ ら れ た 新 た な ァ ブ ロ ー チ を 提供す る も の で あ る 。
こ の 発明 で は, 人間 に よ る 対象の 操作 ま た は制 御 を モ デ ル化 し , こ の モ デ ル に し た が っ て, 機械 に よ っ て 人間 の 操作 ま た は制御 を再現す る た め に , 次の 行動原 理 に 基づ い て い る 。
「 ァ ク シ ョ ン に は 何 ら かの ス ト ラ テ ジ ー が存在 し , ス ト ラ テ ジ 一 に は何 ら か の ゴ ー ル が存在す る o 」
こ の 発 明 で は 人間 に よ る 実際の 操作 ま た は制 御 の 記 録を情報源 と す る 。 こ の 記録力、 ら ス ト ラ テ ジ ー を 抽 出 し, ス ト ラ テ ジ 一か ら.ゴ一ル を導 き 出 す こ と に よ り , ゴ一 ル , ス ト ラ テ ン 一 お よ び ァ ク シ ョ ン か ら な る ヒ ェ ラ ノレ キ一構造の モ デ ル が構築 さ れ る 。
こ の 発 明 に よ る 推 冊 方法 は, 最 も 抽象的 に 規 疋 " ¾ る と 次の よ う に な る 。
こ の 発明 に よ る 推論方法 は, 対象を 操作 ま た は 制御 す る こ と に よ り 得 ら れ る 入, 出 力変数値 の 組 か ら な る —連 の ア ク シ ョ ン · デ ー タ を 取込み , 取込 ん だ 上記 ァ ク シ ヨ ン · デ ー タ の う ち の 入力 変数値 に 着 目 し て , 上 記操作 ま た は 制御 が め ざ す ゴ ー ル を 入力 変数 ご と に 作 成 し , 取込 ん だ上記 ア ク シ ョ ン ' デ ー タ の う ち の 出 力 変数値 に 着 目 し て , 上記 ゴ ー ル を達成す る た め の ス ト ラ テ ジ一を 作成 し , 上記 ゴ ー ル と そ の ゴ ー ル に 未 だ 達 し て い な い ス ト ラ テ ジ ー と を 連鎖 さ せ, 上記 ス ト ラ テ ジ 一 と そ れ を 実現す る ァ ク シ ョ ン · デ ー タ と を 連鎖 さ せ , こ れ に よ つ て, 相 互 に 連鎖 し た ゴ ー ル , ス ト ラ テ ジ ー お よ び ァ ク シ ョ ン か ら 構成 さ れ る ヒ エ ラ ル キ ー 構 造 の モ デ ル を 形成す る も の で あ る 。
従来 の 事例 ベ ー ス 推論 は経験 的 な レ ベ ル , す な わ ち ア ク シ ョ ン , ケ ー ス に と ど ま る も の で あ り , そ れ以上 の 発展性 は な い 。 過 去 に 経験 し な か っ た ケ ー ス が 出 現 し た 場 合 に は 適切 な 解 は 得 ら れ な い 。
こ の 発 明 に よ る と 実 際 の ァ ク シ ョ ン · サ ン プ ル か ら ゴ ー ル が作成 さ れ, そ の ゴ ー ル を 達成す る た め の ス ト ラ テ ジ 一 が 作 成 さ れ る 。 そ し て , ゴ ー ル と ス ト ラ テ ジ 一 と ァ ク シ ョ ン と 力く 互 い に 結 び つ け ら れ る 。 し た が っ て , 操 作 や 制 御 の た め に 概念 的 な レ ベ ル で 方 向 性 が 定 め ら れ ( ゴ ー ル ) , そ の ゴ ー ル に 向 っ て , そ の ゴ ー ル に 連鎖 す る ス ト ラ テ ジ 一 で状況 に 応 じ た 意志決 定 が行 な わ れ る 。 最後 に , ス ト ラ テ ジ ー に 連鎖 す る ァ ク シ ヨ ン · ケ ー ス に よ っ て具体的 な 意志決定 ( 出 力値 の決定) が行 な わ れ る こ と に な る 。
フ ァ ジ ィ · ル ー ル を 含む ル ー ル · ベ ー ス 推論で は , ル ー ル を作成す る た め に エ キ ス パ ー ト に ィ ン タ ビ ュ ー し て そ の 知識 を ル ー ル化す る 作業が必要で あ る 。 こ の 発明 に よ る と , 過去 の 操作記録か ら 自 動 的 に ゴ ー ル と ス ト ラ テ ジ 一 が作成 さ れ る 。
こ の 発明 に よ る 推論方法 を具体的 に規定す る と 次 の よ う に な る 。
こ の 発明 に よ る 推論方法 は, 対象 を操作 ま た は制御 す る こ と に よ り 得 ら れ る 入, 出 力変数値の 組か ら な る —連の ア ク シ ョ ン · サ ン プ ル を取込み , 取込ん だ上記 ア ク シ ョ ン · サ ン プ ル の う ち の 入 力 変数値 に 着 目 し て, 入力変数値の 最初の 値, 最後の 値, 最大値 お よ び 最小値 を用 い て規定 さ れ る ゴ ー ル · ケ ー ス を入力変数 ご と に作成 し , 取込ん だ上記 ア ク シ ョ ン · サ ン プ ル の う ち の 出 力変数値 に 着 目 し て, 出 力変数値が同一性 を 保つ 入力変数 の 範囲 を 抽 出 し , 各範囲 に つ い て, 出 力 変数値 に 入力変数 の 範囲 を結びつ け る ス ト ラ テ ジ ー · ケ ー ス を作成 し , 上記 ゴ ー ル · ケ ー ス の 条件を 満 た さ な い 入力変数範囲 を も つ ス ト ラ テ ジ ー · ケ ー ス を そ の ゴ ー ル · ケ ー ス に連鎖 さ せ, 上記 ス ト ラ テ ジ ー . ケ ー ス に お け る 入力変数範囲 に 入 る 入力変数値 を も つ ァ ク シ ョ ン · ケ ー ス を そ の ス ト ラ テ ジ ー · ケ ー ス に 連鎖 さ せ , こ れ に よ つ て, 相 互 に 連鎖 し た ゴ ー ル · ケ ー ス , ス ト ラ テ ジ ー · ケ ー ス お よ び ア ク シ ョ ン · ケ ー ス 力、 ら 構成 さ れ る ヒ エ ラ ル キ ー 構造 の モ デ ル を 形成す る も の で あ る 。
好 ま し い 実施態様 で は 取込 ん だ 上記 ァ ク シ ョ ン · サ ン プ ル が 所定桁 の バ イ ナ リ ィ · デ ー タ に 変換 さ れ る 。
他 の 好 ま し い 実施態 様 で は , ゴ ー ル · ケ ー ス を 作成 す る 上記 ス テ ッ プ は , 仮 の ゴ ー ル · ケ ー ス を 作成す る ス テ ッ プ と , 作 成 し た ゴ ー ノレ · ケ ー ス を 出 力 す る ス テ ツ プ と , ユ ー ザ に よ る 評 価 を 受 け た 後 の ゴ ー ル , ケ ー ス を 最終 的 な ゴ ー ル · ケ ー ス と す る ス テ ッ プ と か ら 構成 さ れ る 。
上 記 の ユ ー ザ に よ る 評 価 は , ゴ ー ル · ケ ー ス の 追 加 , 削 除 お よ び変更 を 含 む 。
ス ト ラ テ ジ ー · ケ ー ス を 作成す る 上記 ス テ ッ プ は , よ り 具体 的 に は , 時 間順序 で 配列 さ れ た 出 力 変数値 の 隣 り 合 う も の を 順次比較 し , 変化 が あ つ た と き に そ の 前 後 の 出 力変数値 を ス ト ァ し て ス ト ラ テ ジ ー · サ ン プ ル を 作成 す る ス テ ッ プ と . 上記 ス ト ラ テ ジ ー · サ ン プ ル を 上 記 ァ ク シ ョ ン · サ ン プ ル を 参 照 し て , d 0 "· w h e n -- u n t i l ··■ m i n "- m a x の 形 式 を も つ ス ト ラ テ ジ ー · ケ ー ス に 変換 す る ス テ ッ プ と , 変換後 の 上記 ス ト ラ テ ジ ー · ケ ー ス を 最適 化 す る ス テ ッ プ と か ら 構成 さ れ る 。 上記 ス ト ラ テ ジ ー · ケ ー ス を 最適化す る ス テ ッ プ の — 例 は , 同 じ 出 力 変 数 値 を も つ 2 つ 以 上 の ス ト ラ テ ジ ー · ケ ー ス の 数 を 減少 さ せ る も の で あ る 。
ス ト ラ テ ジ ー · ケ ー ス と ゴ ー ル · ケ ー ス と の連鎖 は 具 体 的 に は , メ モ リ に お い て , 各 ス ト ラ テ ジ ー · ケ ー ス に 設 け ら れ た ゴ ー ル · ケ ー ス へ の ポ イ ン タ と , 各 ゴ ー ル , ケ ー ス に 設 け ら れ た ス ト ラ テ ジ ー · ケ ー ス へ の ポ イ ン タ と に よ り 実現 さ れ る 。
ァ ク シ ョ ン · ケ ー ス と ス ト ラ テ ジ ー · ケ ー ス と の連 鎖 は 具 体 的 に は , メ モ リ 上 に お い て , 各 ア ク シ ョ ン · ケ ー ス に 設 け ら れ た ス ト ラ テ ジ ー · ケ ー ス へ の ボ イ ン 夕 と , 各 ス ト ラ テ ジ ー · ケ ー ス に 設 け ら れ た ァ ク シ ョ ン · ケ ー ス へ の ボ イ ン タ と に よ り 実 現 さ れ る 。
ア ク シ ョ ン · ケ ー ス は , 全 く 同 じ 入 , 出 力 変数値 の 組 力、 ら な る ア ク シ ョ ン · サ ン プ ル を , そ の 一つ を 残 し て 他 を 削 除 す る こ と に よ り 作成 さ れ る 。
こ の よ う に し て 作成 さ れ た モ デ ル は , 上記一連 の ァ ク シ ョ ン · サ ン プ ル を 入 出 力 デ ー タ と し て 用 い て 検証 さ れ る 。
す な わ ち , こ の 検証 は , 上記 ア ク シ ョ ン ' サ ン プ ル を そ の 時 間 の 順 序 で 一 つ ず つ 取 込 み , 取 込 ん だ ァ ク シ ョ ン · サ ン プ ル の 入 力 変数値 に よ っ て 満 た さ れ な い 条 件 を も つ ゴ ー ル · ケ ー ス を 捜 し 出 し , 捜 し 出 し た ゴ ー ル . ケ ー ス に 連鎖す る ス ト ラ テ ジ ー · ケ ー ス を リ ス ト し , 上記の リ ス ト さ れ た ス ト ラ テ ジ ー · ケ ー ス の 中 か ら . 上記取込ん だ 入 力 変数値 に よ っ て 満 た さ れ る 条件 を も つ ス ト ラ テ ジ ー · ケ ー ス を 捜 し 出 し , 捜 し 出 し た ス ト ラ テ ン 一 · ケ ー ス に 連 鎖 す る ア ク シ ョ ン · ケ ー ス を リ ス ト し , 上記 の リ ス ト さ れ た ア ク シ ョ ン · ケ ー ス の う ち で 上記 の 取込ん だ 入力 変数値 と 一致す る ま た は 最 も 近 い 入力変数値 を も つ ァ ク シ ョ ン · ケ ー ス の 出 力変数値 を 出 力 す る こ と に よ り 行 な わ れ る 。 こ の 出 力 値 と ァ ク シ ヨ ン • サ ン プ ル の 出 力変数値 と を ユ ー ザ が比絞 す る こ と に と よ っ て , 上記 の モ デ ル が検証 さ れ る 。
上述 し た 推 5冊 法 に よ っ て作成 さ れ た モ デ ル を 用 い て 対象 を 操 作 ま た は 制 御 す る た め の 方 法 も ま た こ の 発 明 に よ っ て 提供 さ れ て い る 。
こ の 方 法 は , 対象 か ら 得 ら れ る 入 力 デ ー タ を 順次取 込 み , 取込 ん だ入力 デ ー タ に よ っ て 満 た さ れ な い 条件 を も つ ゴ ー ル · ケ 一 - ス を 捜 し 出 し , 捜 し 出 し た ゴ ー ル · ケ ー ス に 連鎖す る ス ト ラ テ ジ ー · ケ ー ス を リ ス ト し . 上記 の リ ス ト さ れ た ス ト ラ テ ジ ー · ケ ー ス の 中 か ら , 上記 の 取込 ん だ 入 力 デ ー タ に よ っ て 満 た さ れ る 条
h — ― 、、*
件 を も つ ス フ ァ ン 一 · ケ ー ス を 捜 し 出 し , 捜 し 出 し た ス ト ラ テ ジ ー · ケ ー ス に 連鎖す る ァ ク シ ョ ン · ケ ー ス を リ ス ト し , 上記 の リ ス ト さ れ た ア ク シ ョ ン * ケ 一 ス の う ち で 上記 の 取込ん だ 入力 デ ー タ と 一致 す る ま た は最 も 近 い 入力変数値を も つ ァ ク シ ョ ン · ケ 一 ス の 出 力変数値 を 出 力 し , 上記処理を入力 デ ー タ を 取込む ご と に繰返す も の で あ る 。
こ の 発明 は上述 し た推論方法お よ び制御方法を 実行 す る シ ス テ ム も 提供 し て い る 。
こ の 発明 に よ る 推論 シ ス テ ム は, 対象 を 操作 ま た は 制御す る こ と に よ り 得 ら れ る 入, 出 力変数値 の 組 力、 ら な る 一連 の ァ ク シ ョ ン · サ ン プ ル を 取込 ん で g己 tkす る 手段, 取込 ん だ上記 ァ ク シ ョ ン * サ ン プ ル の う ち の 入 力変数値 に 着 目 し て, 入力変数値の 最初 の 値 , 最 後 の 値, 最大値 お よ び最小値を用 い て規定 さ れ る ゴ ー ル , ケ ー ス を 入力変数 ご と に作成す る 手段, 取込ん だ上記 ァ ク シ ョ ン · サ ン プ ル の う ち の 出 力 変 数値 に 着 目 し. て, 出 力変数値が同一性を 保つ 入力変数の 範囲 を 抽 出 し , 各範囲 に つ い て, 出力変数値 に 入力変数の 範囲 を 結 びつ け る ス ト ラ テ ジ ー · ケ ー ス を作成す る 手段, 上 言己 ゴ一 ル • ケ ー ス の 条件を満 た さ な い 入力変数範囲 を も つ ス ト ラ テ ジ 一 · ケ ー ス を そ の ゴ ー ル · ケ ー ス に 連' 鎖 さ せ る 手段, な ら び に 上記 ス ト ラ テ ジ ー · ケ ー ス に お け る 入力変数範囲 に 入 る 入力変数値 を も つ ァ ク シ ョ ン · ケ 一 ス を そ の ス ト ラ テ ジ ー · ケ ー ス に 連 鎖 さ せ る 手段 を 備 え , こ れ に よ つ て . 相互 に 連鎖 し た ゴ ー ル · ケ ー ス , ス ト ラ テ ジ ー · ケ ー ス お よ び ァ ク シ ヨ ン · ケ ー ス か ら 構成 さ れ る ヒ エ ラ ル キ ー 構造 の モ デ ル を形 成す る も の で あ る
上記の 推 ス テ ム に よ っ て作成 さ れ た モ デ ル を 用
§一
い て対象 を 操作 ま た は制御す る た め の シ ス テ ム は , 対 シ
象か ら 得 ら れ る 入力 デー タ を順次取込む手段, 取込ん だ 入 力 デ ー タ に よ っ て 満 た さ れ な い 条件 を も つ ゴ ー ル · ケ ー ス を 捜 し 出 し , 捜 し 出 し た ゴ ー ル · ケ ー ス に 連鎖す る ス 卜 ラ テ ジ ー * ケ ー ス を メ モ リ 上で リ ス 卜 す る 手段, 上記の リ ス ト さ れ た ス ト ラ テ ジ ー · ケ ー ス の 中 か ら , 上記の 取込ん だ入力 デー タ に よ っ て満 た さ れ る 条件 を も つ ス ト ラ テ ジ ー · ケ ー ス を捜 し 出 し , 捜 し 出 し た ス ト ジ ー · ケ ー ス に連鎖す る ァ ク シ ヨ ン · ケ ー ス を メ モ リ 上で リ ス ト す る 手段, 上記 の リ ス ト さ れ た ァ ク シ ョ ン • ケ ー ス の う ち で上記の 取込ん だ入力 デ ー タ と一致す る ま た は最 も 近 い 入力変数値を も つ ァ ク シ ヨ ン · ケ 一 ス の 出 力変数値を 出力す る 手段, な ら び に 上記処理 を入力 デ ー タ を取込む ご と に 繰返す よ う 制御す る 手段 を 備 え て い る 。
こ の 発明 の 他 の 特徴 お よ び利点 は . 図面を 参照 し た 実施例 の 説明 に お い て 明 ら 力、 に な る で あ ろ う
図面 の 簡単 な 説明
第 1 図 は モ デ ル形成 モ ー ド に お け る , 制 御対象 を 含 む シ ス テ ム 全体 の 構成 を 示す ブ ロ ッ ク 図 で あ る 。 第 2 図 は 問題解決 モ ー ド に お け る , 制 御対象 を 含む シ ス テ ム 全体 の 構成を 示す ブ ロ ッ ク 図で あ る 。
第 3 図 は ヒ エ ラ ル キ ー • モ デル を用 い る こ の 発明 に よ る シ ス テ ム の 構成例 を示す ブ ロ ッ ク 図 で あ る 。
第 4 図 は制御対象の 例 と し て の ク レ ー ン 制御 シ ス テ ム を 示す斜視図で あ .る 。
第 5 図 は 振れ角 検 出 の た め の他の 構成例 を 示 し て い る o
第 6 図 は ク レ ー ン 制御の シ ミ ユ レ ー シ ョ ン を 汀 つ シ ス テ ム を 示す ブ ロ ッ ク 図で あ る 。
第 Ί 図 は シ ミ ュ レ ー シ ョ ン に お け る 変数を 示す も の で あ る o
第 8 a 図 お よ び第 8 b 図 は モ デル形成 モ ー ド に お け る 処理手順 を 示す フ ロ ー • チ ャ ー ト で あ る 。
第 9 図 は ク レ ー ン 制御の シ ミ ュ レ ー シ ョ ン に お い て 得 ら れ た 入, 出 力 デ ー タ の一例 を示す も の で あ る 。
第 1 ( 図 は 第 9 図 に 示 す 入 , 出 力 デ ー タ を バ イ ナ U ィ * ェ ン コ 一デ ィ ン グす る こ と に よ り 得 ら れ た バ イ ナ リ イ • ス ト リ ン グ を 示す。
第 1 1図 は ク レ ー ン 制御の シ ミ ュ レ ー シ ョ ン · デ ー タ に お い て , 入力変数 ご と に 作成 さ れ た ゴ ー ソレ · ケ ー ス 作成の た め の リ ス ト の 例 を示す。
第 1 2図 は ク レ ー ン 制御 に お け る ゴ ー ル · ケ ー ス の 例 を 示す
第 1 3図 は ク レ ー ン 制御 に お け る ス ト ラ テ ジ ー · サ ン プ ル の 例 を 示す。
第 1 4図か ら 第 1 6図 は ス ト ラ テ ジ一 · サ ン プ ノレ の ス ト ジ ー · ケ ー ス へ の 変換を 説明す る た め の 簡単 な 例 を 示す も の で あ り , 第 1 4図 は ア ク シ ョ ン • サ ン プ ル の 例, 第 1 5図 は ス ト ラ テ ジ ー • サ ン プ ル の 例, 第 1 6図 は ァ ク シ ョ ン · サ ン プ ル の グ ラ フ を そ れぞ れ示す も の で あ る
第 1 7図 は ク レ ー ン 制御 に お け る ア ク シ ョ ン · サ ン プ ルの グ ラ フ を 示す。
第 1 8 a 図 お よ び第 b 図 は ク レ ー ン 制御 に お け る ス ト ラ テ ジ ー · ケ ー ス の 例 を 示す。 - 第 図 は 第 1 8 a 図 お よ び第 1 8 b 図の ス 卜 ラ テ シ 一 ケ ー ス の バ イ ナ リ ィ · ス ト リ ン グを 示す
第 2 G図 は 各 ゴ ー ル · ケ ー ス の ス ト ラ テ ジ ー · ケ ー ス への 連鎖 を 示す。 .
第 2 1図 は 各 ス ト ラ テ ジ ー • ケ ー ス の ゴ一 ル · ケ ー ス お よ び ア ク シ ョ ン · ケ ー ス へ の 連鎖を 示す。
第 2 2図 は 各 ァ ク シ ョ ン ' ケ ー ス の ス ト フ " ン、、 — · ケ ー ス へ の 連鎖 を 示す。
第 2 3図 は ゴ ー ル , ス ト ラ テ ジ 一 お よ び ア ク シ ョ ン か ら な る ヒ エ ラ ル キ ー 構造を示す。
第 2 4図 は 問題解決 モ ー ド に お け る 処理手順を 示す フ ロ ー · チ ヤ 一 ト で あ る 。
第 2 5図 は 問題解決 モ ー ド に お け る 入, 出 力 デ ー タ の 例を示す 発明 を実施す る た め の最良の形態
( 1 ) シ ス テ ム に お け る 2 つ の モ ー ド
ヒ エ ラ ノレ キ ー · モ デ ル を用 い る こ の 発明 に よ る シ ス テ ム は 2 つ の モ ー ド を も つ 。 そ の 一 つ は モ デ ル 形成 モ ー ド で あ り , 二つ 目 は 問題解決モ ー ド で あ る o
モ デ ル形成 モ ー ド で は 制御対象 (披制御 シ ス テ ム ) を人間 (オ ペ レ ー タ ) が実際 に操作 し た と き に 得 ら れ る デー タ (操作量 お よ び制御量) を入力 し て, 後述す る ヒ エ ラ ノレ キ ー 構造 を も つ モ デルを形成す る 。
問題解決 モ ー ド で は , 既 に 形成 さ れ た モ デル に し た が っ て制御対象を 現実 に 制御す る 。
第 1 図 は モ デ ル形成 モ ー ド に お け る , 制御対象 を 含 む シ ス テ ム 全体の 構成 を 示 し て い る 。 こ の発明 に よ る シ ス テ ム は モ デ ル 形 成 シ ス テ ム ( ま た は 学習 シ ス テ ム ) と し て体現す る 。
制御対象 (被制 御 シ ス テ ム ) 1 0に お け る 制御量 は一 般 に セ ン サ 1 1, 1 2 , …, 1 に よ っ て検 出 さ れ る o .
さ れ た制御量 (入力 デ ー タ X i , X 2 , …, X
k ) は 操作装置 2 0に 与え ら れ る 。 オ ペ レ ー タ は制御対象 の 状 況を 目 視 し な 力《 ら , 必要 に 応 じ て, 表示 さ れ た 制 御量 の 値を見 な 力 ら ( た と え ば計器の 針の 位置 を 確認 し な が ら ) , 制御対象が所望 の 状態 と な る よ う に 操作 (運 転) す る 。 操作装置 20か ら 出 力 さ れ る 操作量 y ( 出 力 デ ー タ ) ( 1 種類 と は 限 ら な い) は ァ ク チ ユ エ 一 夕 21 に与え ら れ, ァ ク チ ユ エ 一 夕 21に よ っ て制御対象 10が 実 際 に 駆 動 ま た は 運 転 さ れ る 。 制 御 量 X ゥ , … , X k の す ベ て ま た は 一部が操作装置 2 Qに 与え ら れ な い こ と も あ ろ う 。 制御量 は制御対象 1 Qか ら 得 ら れ る で あ ろ う 理 論 値 ( ま た は 計 算 値 も し く は 推 定値 ) で あ つ "" ね よ い
モ デ ル形成 シ ス テ ム 1 A に は セ ン サ 11〜 l kに よ っ て 検 出 さ れ た 制 御量 X } 〜 X k , お よ び操作装置 2 Qか ら 出力 さ れ る 操作量 y が入力す る 。 ァ ク チ ユ エ ー タ Πの 出力, ま た は そ の 測定値が操作量 と し て モ デ ル形成 シ ス テ ム 1 A に 入力す る よ う に し て も よ い 。 モ デル形成 シ ス テ ム 1 A は こ れ ら の 入力 デ ー タ ェ 〜 x k お よ び 出 力 デ ー タ y を 用 い て , 制御 ( ま た は操作) に 関 す る モ デ ル を 作成す る 。
第 2 図 は 問題解決 モ ー ド に お け る . 制御対象 ^ 含 む シ ス テ ム 全体 の 構成 を 示 し て い る 。 こ の 発明 に よ る シ ス テ ム は 自 動制 御 シ ス テ ム ( 自 動操作 シ ス テ ム ) . 1 B と し て体現す る 。
自 動制 御 シ ス テ ム 1 B は セ ン サ 11〜 l kに よ っ て 検 出 さ れ る 制御対象 10の 制 御量 〜 x k を取込み . 先 に 作成 し た モ デ ル に し た 力く つ て, 制御対象 1 Qに 与え る ベ き 操作量 y を 算 出 す る 。 操作量 y は制御対象 10を駆動 す る ァ ク チ ユ エ 一 夕 21に 与 え ら れ る ( ま た ァ ク チ ュ エ ー タ 21の 出 力 が 算 出 し た 値 と な る よ う に ァ ク チ ュ エ ー タ 21が制 御 さ れ る ) 。 こ れ に よ つ て , 制 御対象 10 は , オ ペ レ ー タ が操作装 置 2 Gを 用 い て制 御対象 を 制 御 ( 操作) し た の と 同 じ よ う に , 自 動制 御 シ ス テ ム 10 B に よ っ て 制 御 (操作) さ れ る こ と に な る 。
(2) シ ス テ ム 構成 と 応用 例
モ ー ド に 応 じ て モ デ ル形成 シ ス テ ム ま た は 自 動制御 シ ス テ ム と し て 体 現 す る こ の 発 明 に よ る ヒ ェ ラ ル キ ー · モ デ ル を 用 い た シ ス テ ム の 構 成 例 が 第 3 図 に 示 さ れ て い る 。
こ の シ ス テ ム は 基 本 的 に は コ ン ピ ュ ー タ 30に よ り 実 現 さ れ る 。 コ ン ピ ュ ー タ 30は い わ ゆ る ノ、。 一 ソ ナ ノレ ' コ ン ピ ュ ー 夕 , 大 型 , 小 型 を 問 わ ず そ の 他 の 汎 用 コ ン ピ ュ ー タ を 利 用 で き る 。
コ ン ピ ュ ー タ 3 Qに は , モ デ ル形成 モ ー ド お よ び 問 題 解決 モ ー ド に お け る 処理 を 実行す る た め の プ ロ グ ラ ム を 格納 し た メ モ リ 31, な ら び に こ れ ら の 処理 に お い て 生成 さ れ る , ま た は 用 い ら れ る デ ー タ ( モ デ ル を 表 わ す デ ー タ を 含 む ) を 格納 す る メ モ リ Πが 設 け ら れ て い る 。 こ れ ら の メ モ リ 31と Πは R O M , R A M , ハ ー ド デ ィ ス ク , フ ロ ッ ピ ー · デ ィ ス ク , 光 デ ィ ス ク 等 に よ り 実現 さ れ る 。 メ モ リ 31と Π力く と も に ハ ー ド デ ィ ス ク に よ り 実現 さ れ て も よ い し , メ モ リ Πが R O M と ハ ー ド デ イ ス ク ま た は フ ロ ッ ピ ー · デ ィ ス ク に よ り , メ モ リ 3 1が R A M と ハ ー ド デ ィ ス ク に よ り そ れ ぞ れ 実現 さ れ て も よ い 。 こ こ で は , メ モ リ 3 1 と 3 2は 格 納 さ れ る 夕 の 種 類 に 応 じ て 分 類 さ れ た 概 念 的 な も の で あ る
Ώ ン ピ ュ ー タ 3 Qに は ま た , 入力装 置 3 3 , な ら び に 出 力 装 置 と し て の 表 示装 置 34お よ び プ リ ン 夕 3 5が接続 さ れ て い る 。 入 力 装 置 3 3は キ ー ボ ー ド お よ び マ ウ ス を 含 む。 メ ニ ュ 一 画 面 を 表示す る 表示装 置 34を 入力 装 置 3 3 の 一部 と 考 え て も よ い 。 出 力装 置 と し て は 表示 装 置 34 お よ び プ リ ン タ 3 5の う ち の 少 な く と も い ず れ か 一 方
(好 ま し く は 表 示装 置 3 4 ) が あ れ ば よ い 。
Π ン ピ ュ ー タ 3 0に は さ ら に , セ ン サ 1 1 〜 I kに よ っ て 快 UJ さ れ た 制 御対象 1 Qに お け る 制 御量 (通常 は ア ナ 口 グ量 ) を デ ィ ジ タ ル · デ ー タ に 変換す る A Z D 変換器
3 6 , モ デ ル 形成 モ ー ド に お い て操作装 置 2 0か ら 出 力 さ れ る 操作量 ( ア ナ ロ グ量 の 場 合) を デ ィ ジ タ ル ' デ ー 夕 に 変換す る A Z D 変換器 Π , お よ び 問 題解決 モ ー ド に お い て 算 出 さ れ ァ ク チ ユ エ 一 夕 2 1に 与 え ら れ る 操 作 量 (通常 は デ ィ ジ タ ル 量 で あ る ) を , 必要 に 応 じ て ァ ナ 口 グ 量 に 変 換 す る D / A 変 換 器 3 8が 接 铳 さ れ て い る o こ れ ら の A Z D 変 換 器 , D / A 変 換 器 は コ ン ュ 一 夕 に 内 蔵 さ れ た も の を 利 用 す る こ と も で き る 。 ま た , こ れ ら の A Z D 変換器, D / A 変 換 器 は 不要 で あ れば設 け る 必要 は な い 。 た と え ば ァ ク チ ユ エ 一 夕 が デ ィ ジ タ ル量 を 受入 れ る も の で あ る 場合 に は D Z A 変 換器 3 8は 不要で あ る 。
コ ン ビ ュ 一 夕 3 Gに は必要 に 応 じ て, 制御量や操作量 の 入力, 操作量の 出 力 の た め の イ ン タ ー フ ユ イ ス (後 述す る 微分回路等) が設 け ら れ よ う 。
第 4 図 は制御対象 の 例 と し て の ク レ ー ン 制御 シ ス テ ム を 示 し て い る 。 簡単の た め に こ の ク レ ー ン · シ ス テ ム は 一本 の レ ー ル の み を 含む も の で あ る 。
—本の レ ー ル 5 0が適当 な 高 さ 位置 に水平 に架設 さ れ て い る 。 こ の レ ー ノレ 5 0に ト ロ リ (台車) が車輪 4 1に よ っ て移動 自 在 に 支持 さ れて い る 。 車輪 4 1は走行モ 一 夕 4 2に よ つ て 直接 に ま た は 減速機を介 し て回転駆動 さ れ る 。
ト ロ リ 4 0に は ま た 巻上機が設 け ら れて い る 。 こ の 巻 上機は ブ ー リ 4 4と , プ ー リ を駆動す る 巻上モ ー タ 4 6 と , プー リ 4 4に 巻回 さ れ た ワ イ ヤ 4 7と を 含む。 プ ー リ 4 4は ト ロ リ 4 0に 固定 さ れ た ブ ラ ケ ッ ト に 回転 自 在 に 受 け ら れ た 回転軸 4 5に 固定 さ れて お り , こ の 回転軸 4 5 がモ ー タ 4 6に よ り 直接 に ま た は 減速機を介 し て 回転駆 ϋ¾ さ れ ο ワ イ ヤ 4 7の 下端 に は フ ッ ク 4 8が取付 け ら れ, こ の フ ッ ク 4 8に 吊 り 荷 4 9が掛 け ら れ る 。
次の よ う な ク レ ー ン 制御が仮定 さ れ る 。 す な わ ち , 吊 り 荷 4 9が レ ー ン 5 0上の 一点 Α 力、 ら 他の 点 Β ま で ト 口 リ 40を 走行 さ せ る こ と に よ っ て運 ば れ る 。 こ の と き , ワ イ ヤ 47の 振 れ角 ( ト ロ リ が移動 す る 方 向 に お け る 振 れ 角 ) が で き る だ け 小 さ ぐ な る よ う に ( 所定 の 角 度, こ れ を 最 大 許 容 振 れ 度 と い う , 以 内 に 収 ま る よ う に ) , ト ロ リ 4 Qの 走行制御 を 行 う 。
制 御量 と し て は , ト ロ リ 40の 位 置, 振 れ 角 お よ び振 れ 角 の 微分値 が採用 さ れ る 。
ト 口 リ 4 Qの 位 置 は 出 発 点 A を 原 点 と し て 終 点 B に 向 っ て 正方 向 に と ら れ る 。 こ の 位 置 は モ ー タ 42の 回 転 軸 ま た は 車輪 41の 軸 に 設 け ら れ た 回 転 卜 ラ ン ス デ ュ ー サ ( 図 示 略 ) に よ っ て 検 出 さ れ る 。 回 転 ト ラ ン ス デ ュ ー ザ か ら 出 力 さ れ る パ ル ス 列 ( ま た は 位 置 を 表 わ す デ ィ ジ タ ノレ · デ ー タ ) は , コ ン ピ ュ ー タ 30に 入 力 す る ( A / D 変 換 器 は 不 要 で あ ろ う ) 。 必 要 に 応 じ て , ト ロ リ 4 Qが 出 発点 A に あ る こ どを 検知 す る 原 点検 出 器 が設 け ら れ よ う 。
ワ イ ヤ 47の 振 れ 角 は , ト ロ リ 40に 取 付 け ら れ, か つ ワ イ ヤ 47の 像 を 撮像 す る よ う に 配 置 さ れ た テ レ ビ ジ ョ ン · カ メ ラ 51お よ び こ の カメ ラ 力、 ら め 映像信号 を 画 像処理 す る 処理装 置 に よ っ て 検 出 さ れ る 。 画 像処 理装 置 は コ ン ピ ュ ー タ 30の 一部 に よ っ て 実現 で き る 。 こ の 場 合 に は カ メ ラ 51の 映 像信号 が A Z D 変 換 器 36を 通 し て コ ン ピ ュ ー タ 30に 取込 ま れ る 。 線 状物 体 の 垂線 に 対 す る 角 度検 出 の た め の 画 像処理 は 公 知 の 手法 に よ り 行 9
つ こ と がで さ る 。
れ角 の 微分値 は コ ン ピ ュ ー タ 3 0に よ つ て演算す る こ と がで き る o
第 5 図 に 示す よ う に , ワ イ ヤ 4 7に 代え て, シ ャ フ 卜
4 1 A を 用 い て 吊 り 荷を 吊 る 場合 に は, 振れ角 は よ り 簡 単 に 検 J C" れ る 。 シ ャ フ ト A は そ の 上端で軸 4 5に 固 定 さ れ る 。 軸 4 5は ブ ラ ケ ッ ト に 回転 自 在 に 支持 さ れ る o シ ャ フ ト A の 下端が フ ッ ク 4 8と な っ て い る。 軸 4 5の 回転角 が角 度 セ ン サ 5 2に よ っ て検 出 さ れ る 。 角 度 セ ン サ 5 2の 出 力信号が ア ナ ロ グ量の 場 合 に は A Z D 変 換器 3 6で デ ィ ジ タ ル量 に 変換 さ れて コ ン ピ ュ ー 夕 3 0に 取込 ま れ る 。 ま た 角 度セ ン サ 5 2の 出力信号 は 微分回路 で微分 さ れ, こ の 微分信号が A Z D 変換器 3 6を通 し て. 振れ角 の 微分値 と し て コ ン ピ ュ ー タ 3 0に 取込 ま l O o も ち ろ ん 微分処理 は コ ン ピ ュ ー タ 3 0で行 う こ と も で き る o 角 度 セ ン サ 5 2の 出 力力くデ ィ ジ タ ノレ量の 場 合 に は そ の ま ま ン ピ ュ ー 夕 3 0に 取込 ま れ る 。
操作量 は ト 口 リ 4 0の 移 is度で め る 0 移勦速度 は 走' 行 モ 一 夕 4 2に 与 え る 電圧 ま た は電流 と し て も 表現 で き る 0
の 発 明 の シ ス テ ム は制御量を検出 す る セ ン サ や 操 作量が与え ら れ る ァ ク チ ユ エ 一 夕 の な い 制 御対象 に も 応用 で き る 。 た と え ば株価の予測 シ ス テ ム で あ る 。 こ の 場 合 , 入 力 デ ー タ は 本 日 の 株 価 ( 各 銘 柄 に つ い て の ) , 本 日 の 平均株 価等 で あ り , キ ー ボ ー ド, 通信装 置, フ ロ ッ ピ ー · デ ィ ス ク · ド ラ イ バ等 か ら 入力 さ れ よ う 。 出 力 デ ー タ は 売 り 決定, 買 い 決定 お よ び そ の 量 と な ろ う 。 こ れ ら は 表示装置 や プ リ ン タ か ら 出 力 さ れ よ う 。
他 の 例 と し て は , 店舗 の 売上予 測 シ ス テ ム で あ る 。 入力 デ ー タ は 曜 日 , 天候, 在庫量等 で , キ ー ボ ー ド , 通信装 置, フ ロ ッ ピ ー · デ ィ ス ク · ド ラ イ バ 等 か ら 入 力 さ れ る 。 出 力 デ ー タ は 予 測売上, 仕入量等 で, 表 示 装 置 や プ リ ン タ カ、 ら 出 力 さ れ る 。
以 下 に 詳 述 す る モ デ ル 形 成 モ ー ド お よ び 問 題 解 決 モ ー ド の 説 明 に お い て は , 上述 し た ク レ ー ン 制 御 シ ス テ ム が例 に 挙 げ ら れ る 。 モ デ ル形成 モ ー ド で用 い ら れ る 実 際 の 運転 デ ー タ は シ ミ ユ レ ー シ ョ ン に よ っ て 得 ら れ た デ ー タ で あ る 。 シ ミ ユ レ ー シ ョ ン · シ ス テ ム の 例 が 第 6 図 に 示 さ れ て い る 。 こ の シ ス テ ム は , コ ン ピ ュ ー 夕 3 0 , プ ロ グ ラ ム · メ モ リ 3 1 , デ ー タ ' メ モ リ 3 2, 入 力 装 置 3 3, 表 示装 置 3 4お よ び プ リ ン タ 3 5か ら 構 成 さ れ る 。 こ の シ ス テ ム に お い て 問題解 決 モ ー ド の 処 理 も 行 な わ れ る 。 第 6 図 は こ の 発 明 に よ る シ ス テ ム の 最 小限 の 構造 を 示 し て い る と い え よ う ( も ち ろ ん , 出 力 装 置 と し て は 表示装 置 ま た は プ リ ン タ の い ず れ か 一 方 が あ れ ば よ い ) 。
シ ミ ュ レ ー シ ョ ン に お い て , ト ロ リ の 位 置 X お よ び 振れ角 0 を 求 め る た め に 用 い ら れた 計算式 は 次の 通 り で あ る (図 7 参照) 。
Figure imgf000023_0001
で,
m , は ト ロ リ の 質
m は 吊 り 荷の 質量
t は ロ ー プの 長 さ
g は 重力 の 加速度
F は モ ー タ に よ る 推進力 で次式 で求 め ら れ る
F = K + Κ ω + Κ V た だ し ω = V d - V ω 2 = J ω ι て ) d
こ こ で , Κ 2 お よ び Κ 3 は 定数で あ る V は モ ー タ 速度, V は モ 一 夕 の 目 標 速 度 で , シ ュ レ ー ン ン に お い てオ ペ レ ー タ に よ っ て与え ら れ る 操作量で あ る 。 す な わ ち V d は 次の 値を と る
V Q = { 4 0 ( m / m i n ) 0 , 4 0 ( m / m i n ) }
( 3 ) モ デ ル形成 モ 一 ド に お け る 処理
こ の 項 お よ び次の 「 ) 問題解決モ ー ド に お け る 処 理」 は主 に コ ン ビ ユ ー タ 3 0に よ っ て実行 さ れ る 。 し た が っ て , 特 に 言及 し な い 限 り , 以下 に記述す る 具体的 処理 は コ ン ビ ュ タ 3 0に よ っ て実行 さ れ る 処理 と 理解 さ れ た い
モ デ ル は , ゴ ー ノレ ( 目 的 ま た は 目 標 ) , ス ト ラ テ ジ 一 (戦略) お よ び ア ク シ ョ ン (実行, 行動) の 3 つ の レ べ ル か ら な る ヒ エ ラ ル キ ー構造 を も つ 。 こ の 構造 は次の 原理 に 基づ い て レヽ る 。
「すベ て の ァ ク シ ヨ ン に は, そ れ に 関連す る 少 な く と も 一つ の ス ト ラ テ ジ 一 が存在 し , す べ て の ス ト ラ テ ジ 一 に は そ れ に 関連す る 少 な く と も 一つ の ゴ ー ル が存 在す る o J
モ デ ル形成 モ ー ド に お け る 具体的処理 の 手順 に つ い て第 8 a 図 お よ び第 8 b 図を参照 し て説明 す る 。
ス テ ツ プ 1 0 1
制御対象 1 0を操作装置 2 0を用 い て オ ペ レ ー タ が実際 に操作 (運転) す る 。 こ の と き , セ ン サ 11〜 l kか ら 得 ら れ る 制御量 (入力 デ ー タ ) お よ び ァ ク チ ユ エ 一 夕 21 に与え ら れ る 操作量 ( 出 力 デー タ ) を, 一定時間 間隔 で サ ン プ リ ン グ し , コ ン ピ ュ ー タ 30に取込む。 こ れ ら の 入, 出 力 デ ー タ は 教 師 デ ー タ と し て デ ー タ ' メ モ リ 32に セ ー ブ さ れ る 。 入, 出力 デ ー タ は実際 の 操作記録 で あ る 。
上述 し た ク レ ー ム 制 御 シ ス テ ム の シ ミ ユ レ 一 シ ョ ン に お い て得 ら れ た 入, 出 力 デ ー タ の 一例 が第 9 図 に 示 さ れて い る 。
入, 出 力 デ ー タ は 0. 3 (s e c) ご と に サ ン プ リ ン グ さ れ て い る 。 上 述 し た よ う に , ク レ ー ン は 前進, 後退 お よ び 停 止 の 3 種 類 の 操 作 で 運 転 さ れ る 。 前 進 で は, 操作出 力 ( ト ロ リ の 移動速度 : 出 力 デ ー タ ) u = + 40 ( m / m i n ) , 後退で は u = — 40 ( mノ m i n ) , 停 止 で は u = 0 ( m / m i n ) に 設 定 さ れ る 。 入 力 デ ー タ は , ト ロ リ の 位 置 X ( m ) , ワ イ ヤ の 振 れ 角 Θ ( r a d ) お よ び 振 れ 角 の 変 化 率 ( 微 分 値 ) d S ( r a d / s e c ) で あ る ( 1 !9 // (1 1 を 以'下 1 0 と.表現 す る ) 。
第 9 図 に お い て, T 1 で示す 時間帯 は ト ロ リ の 始動 時で あ り 振 れ角 が で き る だ け 小 さ い 範囲 に お さ ま る よ う に操作 出 力 u が調整 さ れて い る 。 T 2 で示す 時間 帯で は ト ロ リ は 定速運転 さ れて い る 。 T 3 で示す 時間 帯 は ト ロ リ を 目 的地点 B に丁度停止 さ せ る た め に , 操 作 出 力 u が調整 さ れて い る 。 時間帯 T 4 は ト ロ リ が地 点 B に ほ ぼ達 し , 吊 り 荷の 振れ角 ø が徐 々 に減衰 し て い く 過程で あ る 。
ス テ ッ プ 102
入 出 力 デ ー タ が 2 値化 さ れ, バ イ ナ リ ィ · ス ト リ ン グ を 形成す る よ う に 変換 さ れ る 。 各バ イ ナ リ ィ · ス ト リ ン グ を ア ク シ ョ ン · サ ン プル と 呼ぶ。
第 9 図 に 示す入 出 力 デ ー 夕 の バ イ ナ リ ィ · ス ト リ ン グ 力く 第 1 G図 に 示 さ れ て い る 。 入 力 デ ー タ X , Θ , d θ , 出 力 デ ー タ u は そ れ ぞれ 8 ビ ッ ト で構成 さ れて い る
こ の ェ ン コ 一 デ ィ ン グ 処 理 に お い て , い ず れ の 入 , 出 力 変 数 に お い て も , そ の 最 小 値 力く 0 0 0 0 0 0 0 0 に , 最大値が 1 1 1 1 1 1 1 1 に そ れ ぞ れ 変換 さ れ る 。 位置 X に つ い て の 最小値 は 一 1 ( m ) , 最大値 は Π ( m ) で あ る 。 振 れ角 の 最小値 は 一 Q . 1 ( r a d ) , 最 大 値 は + 0 · 1 ( r a d ) で あ る 。 そ の 微 分 値 d の 最 小 値 は 一 0. 1 ( r a d / s e c ) . , 大値 は + G . 1 ( r a d / s e c ) で あ る 。 移動速度 u の 最 小値 は 一 40 ( m / m i n ) , 最大値 は + 40 ( m Z m i n ) で あ る 。 第 10図 を 理 解 し や す く す る た め に , 位 置 X = 0 ( m ) , 振 れ 角 ( 0 = 0 ( r a d ) , 振 れ 角 の 微 分 値 d 0 = 0 ( r a d / s e c ) お よ び 移 動 速 度 u = 0 ( m / m i n ) は そ れ ぞ れ , バ イ ナ リ ィ 値 0 0 0 0 1 0 1 1 , 1 0 0 0 0 0 0 0 , 1 0 0 0 0 0 0 0 , 1 0 0 0 0 0 0 0 で 表 わ さ れ て い る こ と を 付 け 加 え て お く 。
' イ ナ リ イ · コ ー デ ィ ン グ は , モ デ ル 形成 モ ー ド の 処理 プ ロ グ ラ ム を 高級言語 で書 い た と き に 特 に 意味 を 持 つ 。 イ ナ リ イ · ェ ン コ 一デ ィ ン グ す る 第 1 の 理 由 は , ア ク シ ョ ン · サ ン プ ル の 分解能 を 容 易 に 変 え る こ と 力 で き る こ と で あ る 。 た と え ば , 1 6 ビ ッ ト の デ ー タ の 下位 8 ビ ッ ト を 捨 て る こ と に よ り 8 ビ ッ ト · デ ー タ と す る こ と が で き る 。 第 2 の 理 由 は 最適 化 ァ -ル ゴ リ ズ ム の 一 つ で あ る ジ エ ネ テ ィ ッ ク · ァ ノレ ゴ リ ズ ム の 最 も 適 し た 表 現方法 が バ イ ナ リ ィ · ス ト リ ン グ で あ る と い う こ と で あ る 。
ス テ ッ プ 1 0 3
サ ブ ス テ ッ プ 1 0 3 - 1
ま ず , 入 力 変数 ご と に , 最初 の ( イ ニ シ ャ ル ) ァ ク シ ヨ ン · サ ン プ ル と 最 終 の ( 夕 一 ミ ナ ル ) ァ ク シ ョ ン · サ ン プ ノレ と を 抽 出 し , そ れ ら の 差 を と る こ と に よ り 比较 す る 。 ま た , 入 力 変数 ご と に 最 大 値 と 最小値 を 算 出 す る 。
第 9 図 に 示す 例 に お い て は , 時 間 t = 0 . 0 0に お け る ァ ク シ ョ ン · サ ン プ ル が最初 の ァ ク シ ョ ン · サ ン プル (値) で あ り , 時 間 t = 2 9 . 1 G お け る も の が最終 の ァ ク シ ヨ ン · サ ン プノレ (値) で あ る 。 こ れ ら が破線の 長 方形で囲 ま れて い る 。 ま た , 入力変数 0 , d 0 の 最大 値, 最小値が実線の長方形で囲 ま れ て い る 。 入力変数 X の 最 大 値, 最 小 値 は 破線 の 長 方 形 で 囲 ま れ た 最終 値, 最初値で あ る 。
サ ブス テ ッ プ 1 0 3 - 2
サ ブス テ ッ プ 1 G 3 - 1 で得 ら れ た 結果 の リ ス ト を作成 す る 。 こ の リ ス ト は も ち ろ ん デ ー タ · メ モ リ Πに お い て作成 さ れ る 。
第 9 図 に 示す例 に つ い て 得 ら れ た リ ス ト が第 1 1図 に 示 さ れて い る 。
サ ブ ス テ ッ プ 1 G 3 - 3
(最大値) 一 (最小値) = 0 で あ れ ば, そ の 入力変 数 を 無視す る 。 こ の 入力変数 は マ ス ク さ れ る 。
最 大 値 と 最小 値 が 等 し い 入 力 変 数 は , 制 御対象 の 操 作 に お い て 全 く 変 化 し な い も の で あ り , 操 作 や 制 御 に 何 の 影 響 も 与 え て い な い も の で あ る か ら で あ る 0
サ ブ ス テ ッ プ
最初 の 値 と 最終 の 値 と の 差 が零で な い 場 台 に は, そ の 入力変数 に つ い て ゴ ー ル 条件 は 次の よ う に 表現 さ れ る 0
ゴ ー ル 】 : X j 「最終状態」 の 値
= (最終の 値) ゴ ー ル j + i : x i : 「過渡状態」 幅 = [最小値, 最大値 ]
制 御対象 の 実際の操作に お い て入力変数 X i に つ い て最初の 値 と 最終の値 と が異な る と い う こ と は, こ の 操作 は 入力変数 X i を (最初の 値) か ら (最終 の 値) ま で変化 さ せ る よ う に (オ ペ レ ー タ に よ っ て) 意図 さ れて 行 な わ れ た に ち 力く い な い と い う こ と で あ る 。 そ し て, そ の と き の 入力変数 の 変動幅が (最小値) と
(最大値) と の 間 に 収 ま る よ う に 意図 さ れ て行 な わ れ た に ち 力く い な い , と 推 定 す る 。 し た 力く つ て , 入 力 変数 X i に つ い て の ゴ ー ル ( 制 御 目 的 ) は , 入 力 変数 X j の 値 を ( 最 初 の 値 ) か ら 出 発 し て 最 終 的 に は ( 最 終 の 値 ) に も っ て い く こ と で あ り , そ の と き X i の 変 動 幅 を ( 最 小 値 ) と ( 最 大 値 ) と の 間 に 収 ま る よ う に す る こ と で あ る , と い う こ と に な る 。
第 11図 に 示す例 で は, 入力変数 X に お い て, 最初 の 値 と 最終 の 値が異 な る 。 し た 力く つ て入力変数 X に つ い て 次 の ゴ ー ル条件が仮 に 設定 さ れ る 。
ゴ ー ノレ 1 : X : 「最終状態」 の 値 = 9. 467
ゴ ー ル 2 : X : 「過渡状態」 幅
= [ 0. 000 , 9. 461 ]
サ ブ ス テ ッ プ 103 - 5
最初 の 値 と 最終の 値 と の 差が零で あ る 場 合 に は, そ の 入力変数 に つ い て ゴ ー ル条件 は次 の よ う に 記述 さ れ る o
ゴ一ノレ j : X i : 「過渡状態」 幅
= [最小値, 最大値 ]
あ る 入力変数 に つ い て そ の 最初の 値 と 最終 の 値 と が 同 じ で め O 口 に は, 最初の 値 お よ び最終 の 値の い ず れ カヽ ら も ゴ ー ル を導 き 出す こ と は で き な い 。 こ の 場合 に は む し ろ , 操作の 過程 に お い て入力変数を一定の 範 囲 に 保つ よ う な 意図があ つ た の で あ ろ う と 推定す る 。 こ れが , 上述 の ゴ ー ル条件の 意味で あ る 。
第 11図 に 示す 例 で は振れ角 0 お よ び そ の 微分値 d Θ の い ずれ に お い て も , そ の最初 の 値 と 最終 の 値 と が 同 じ で あ る ( と ち に : 0. 000 ) 。 し た 力く つ て , 入 力 変数 Θ , d S に つ い て 次の ゴ ー ル条件が仮 に 設定 さ れ る 。
ゴ一ノレ 3 : Θ : 「過渡状態」 幅
= [ 一 0. 029 , 0. 027 ]
ゴ ー ル 4 : d Θ : 「過渡状態」 幅
= [ - 0. 039 , 0. 042 ]
サ ブ ス テ ッ プ 103-6
上述 し た サ ブ ス テ ツ プ 103-3 , 103 - お よ び 10ト 5 の 結果が, プ リ ン タ 35に よ っ て プ リ ン ト さ れ る , ま た は 表示装 置 34に 表示 さ れ る 0
サ ブ ス テ ッ プ 103- 7
上述 し た ゴ ー ル条件が次の 配列 で バ イ ナ リ ィ · ス ト リ ン グ に ェ ン コ 一 ド さ れ る 。
「 ゴ ー ル番号 入力変数 タ イ プ · フ ラ グ 値」 こ の 配列 に つ い て は 次の ス テ ツ プ 1 0 4 の 説明 に お い て詳述す る 。
ス テ ッ プ 1 0 4
プ リ ン タ 3 5に よ っ て プ リ ン ト さ れ た , ま た は表示装 置 3 4に 表示 さ れた ゴ ー ル の 仮説 (仮 り に 設定 さ れ た 上 記 ゴ ー ル 1 〜 4 ) を み て . ユ ー ザ は こ れ ら を 評価 し , 必要 に 応 じ て こ れ に 新 た な ゴ ー ル条件を 加 え る か, い ずれか の ゴ ー ル条件 を 削除す る か, ま た は ゴ ー ル条件 を修正す る 。
ク レ ー ン の 制御 シ ス テ ム の 例 に お い て は, 設定 さ れ た上述の 仮説 ( ゴ ー ル 1 〜 4 ) の う ち , ゴ ー ル 2 が不 要 な も の と し て 削 除 さ れ る 。 ゴ ー ル 2 は 入 力 変数 X
(位置 を表 わ す) の 過渡状態幅 に 関す る も の で あ り , 位置 X は最初の値か ら 最終の 値 に 向 っ て単調 に 増加す れ ば よ い の で, 過渡状態幅 に 関す る ゴ ー ル仮説 は 不要 で あ る か ら で あ る 。 こ れ に よ り , ゴ ー ル 3 お よ び ゴ ー ル 4 の ゴ ー ル番号が 1 つ ずつ 繰上 り , こ れ ら は そ れぞ れ ゴ ー ル 2 お よ び ゴ ー ル 3 と な る 。
ユ ー ザ に よ る 評 価 を 経 て 最 終 的 に 得 ら れ た も の を ゴ ー ノレ . ケ ー ス と い う 。 上記の 例 に お い て最終的 に 得 ら れ た 3 つ の ゴ ー ル · ケ ー ス が, そ の バ イ ナ リ ィ · ス ト リ ン グ と と も に 第 1 2図 に 示 さ れ て い る 。 バ ィ ナ リ イ · ス ト リ ン グ に お け る 各項 の 配列 お よ び そ の バ イ ナ リ イ · デ ー タ に つ い て 説 明 し て お く ( サ ブ 'ス テ ッ プ 1 0 3 - 7 参照) 。
先頭 の ゴ ー ル 1 , ゴ ー ル 2 , ゴ ー ル 3 の 1 , 2 . 3 が 「 ゴ一 ル · ケ 一 ス 番号 ( ゴ ー ル番号) 」 で あ り , 2 ビ ッ ト で 表 わ さ れ る o
次 の X , Θ , d 0 力く 「 入力 変数」 の 種類 で あ り , こ れ ら も 2 ビ ッ ト で 表現 さ れ る 。 X は 0 0 , S は 0 1 , d 0 は 1 0 で あ る o
第 3 番 目 の 「 夕 イ ブ · フ ラ グ」 は , 「最終状態」 の 値 力、 「過渡状態」 の 幅 か を 表 わ す も の で あ る 。- こ の 夕 ィ プ · フ ラ グ は 2 ビ ッ ト で あ り , 「最終 状態」 の 値 が 0 1 , 「過渡状態 」 の 幅 力く 1 0 に よ っ て 表現 さ れ る 。
最 後 の 「値」 は タ イ プ ' フ ラ グ に よ っ て 異 な る 。 夕 ィ プ · フ ラ グ 力く 0 1 の 場 合 に は 「値」 は 「最終 の 値」 ( 8 ビ ッ 卜 で ¾ わ さ れ る ) と な る 。 「最終 の 値」 は 一 つ し か な い の で . こ の 8 ビ ッ ト , デ ー タ 力く 2 回繰返 さ れ る 。 タ イ プ · フ ラ グ 力く 1 0 の 場 合 に は 「値」 は 最 小 値 , 最 大 値 ( い ず れ も 8 ビ ッ ト で表 わ さ れ る ) の 順序 で 配列 さ れ た 1 6 ビ ッ ト · デ ー タ と な る ( 第 9 図, 第 1 0 図, 第 1 1図 を 参照せ よ ) 。
ス テ ツ プ 1 0 5
出 力 変数 に つ い て ァ ク シ ョ ン · サ ン プ ル を 比铰す る こ と に よ り ス ト ラ テ ジ 一 が 抽 出 さ れ る 。 具体 的 に は , ま ず 出 力 変数 の 最初 の 値 と 最終 の 値 が デ ー タ · メ モ リ 32に ス ト ア さ れ る 。 続 い て , 時刻 t に お け る 出 力 デ ー タ ( ア ク シ ョ ン · サ ン プ ル ) と そ の 次 の 時刻 t + 1 に お け る 出 力 デ ー タ と が比較 さ れ る 。 も し , こ れ ら の 出 力 デ ー タ が 異 な れ ば , 時 刻 t の 出 力 デ ー タ と t + 1 の 出 力 デ ー タ と 力く ス ト ア さ れ る 。 こ の 処理 は t = 0 か ら t = 最終 時刻 ま で , t の 値 を 1 つ ず つ イ ン ク レ メ ン ト し な 力《 ら 繰返 さ れ る 。 こ の よ う に し て メ モ リ に ス ト ア さ れ た デ ー タ を 「 ス ト ラ テ ジ ー * サ ン プ ル 」 と 呼 ぶ。
ク レ ー ン 制 御 シ ス テ ム の ァ ク シ ョ ン · サ ン プ ル を 用 い て 具体 的 に 説 明 し ょ う 。 第 1 G図 に 示 す バ イ ナ リ ィ · ス ト リ ン グ を 参照 す る 。
第 13図 に は 最終 的 に 得 ら れ た ス ト ラ テ ジ ー · サ ン プ ル カく 示 さ れ て い る 。 こ こ に は , 時 刻 お よ び ァ ク シ ョ ン · サ ン プ ゾレ N 0. も 参照 の た め に 示 さ れ て い る 。
出 力 変 数 の 最初 の 値 1 0 0 0 0 0 0 0 ( 時刻 t = 0. 00 , サ ン プ ル N O. = 1 ) と 最 終 の 値 1 0 0 0 0 0 0 0 ( 時刻 t = 29. 10 , サ ン プ ル N 0. 98) と 力く ス ト ァ さ れ る 。
時 刻 t = G . Q Q に お け る 出 力 変 数 値 1 0 0 0 0 0 0 0 と 時刻 t = ϋ . 30に お け る 出 力 変数 値 1 1 1 1
1 1 1 1 と は 異 な る の で, 時刻 t = 0. 30に お け る 出 力 変数値 が ス ト ァ さ れ る ( 時刻 t = 0. (Hの 出 力 変数 値 は 既 に ス ト ァ 済) 。
時刻 t = 0. 30の 出 力変数値 1 1 1 1 1 1 1 1 と 時 刻 t = 0. 60の 出 力変数値 1 0 0 0 0 0 0 0 と が比較 さ れ る 。 こ れ ら は 相互 に 異 な る の で こ れ ら の 値が ス ト ァ さ れ る 。
同 じ よ う に , 時 刻 t = : 0. δ Qに お け る 値 と 時 刻 t = 0. 90に お け る 値 と が相互 に 異 な り , 時刻 t = 0. 9 に お け る 値 と 時刻 t = 1. 20に お け る 値 と も 相 互 に 異 な り , さ ら に 時間 t = 1. 20に お け る 値 と 時刻 1. 5 Gに お け る 値 も 相 互 に 異 な る の で , :: れ ら の 値 力 { ス ト ァ さ れ て い ぐ α
時刻 1. 5 Qに お け る 値 か ら 時刻 14. 4 Q に お け る 値 ま で は 同 じ で あ る 。 し た 力く つ て こ れ ら の 時刻 に お け る 出 力 変数値 は ス ト ァ さ れ な い o
時刻 14. 4 Q に お け る 値 と 時刻 H . 70 に お け る 値 と は 異 な る 。 し た 力《 つ て . こ れ ら の 時刻 に お け る 出 力 変数 値力く ス ト ァ さ れ る 。
時刻 14. 70 に お け る 値 と 時刻 15. (1 G に お け る 値 は 同 じ で あ る 。 し た 力く つ て こ れ ら の 値 は ス ト ァ さ れ な い 。
時刻 15. Q Q に お け る 値 と 時刻 15. 3 ΰ に お け る 値 は 異 な り , 時刻 15. 3 Q に お け る 値 と 時刻 15. 6 G に お け る 値 と は 異 な り , さ ら に 時 刻 15. 6 Q に お け る 値 と 時 刻 15. 90 に お け る 値 と も 異 な る の で, こ れ ら の 値 力く ス ト ァ さ れ る 。 時刻 15. 9 G に お け る 値 と 時刻 16. 2 Q に お け る 値 は 同 じ で あ る か ら こ れ ら の 値 は ス ト ァ さ れ な い。
時刻 16. 2 Q に お け る 値 と 時刻 16. 5 Q に お け る 値 は 相 互 に 異な る の で, こ れ ら の値がス ト ア さ れ る 。
時刻 16. 50 以 降 に お け る 出 力 変数 の 値 は 最終 時刻 29. 10 に 至 る ま です ベ て 同 じ で あ る か ら こ れ ら の 値 は ス ト ァ さ れ な い 。
こ の よ う に し て , 出 力 変数 が変 化 す る 位 置 を 見 付 け, そ の 前後 の 時点 に お け る 出力変数値 を メ モ リ に ス ト ァ す る 。
上記の 例 で は 出 力変数が一種類の場合 に つ い て の み 説明 さ れて い る が . 出 力変数の 種類が複数で あ っ て も 同 じ で あ る 。 こ の 場合 に は い ずれか一つ の 出 力変数 の 値が変化す れ ば 出 力変数が変化 し た と し て, そ の 変化 の 前 後 に お け る す べ て の 出 力 変数 の 値 が ス ト ァ さ れ る o
ス テ ッ プ 106
こ の よ う に し て 抽 出 さ れ た ス ト ラ テ ジ ー · サ ン プ ル を, ア ク シ ョ ン · サ ン プ ル を参照 し て次の 形式 を も つ ス ト ラ テ ジ ー · ケ ー ス に 変換す る 。
厂 d 0 ··· h e n … u n t i l … m i n ··· m a i … 」 - d oに は 出 力変数 ( 1 種類 ま た は複数種類) の 値 が続 く 。 w h e 11に は 入力変数 の 値 ( 1 個 ま た は 複数個) が铳 く 。 u n t i l の 後 に も 入力変数の 値が続 く 。 m i n の 後 に は各入力変数 に つ い て の最小値が配列 さ れ, m a I の 後 に は各入力変数 に つ い て の最大値が配列 さ れ る 。
第 1 4図か ら 第 1 6図 を 参照 し て簡単な 例を用 い て説明 す る 。 こ れ ら の 図 に お い て, 入力変数 は X l , X 2 の 2 種類, 出 力変数 は u の 1 種類で あ る 。 時刻 は t p 〜 t p + i iで示 さ れて い る 。
第 1 4図 は ァ ク シ ョ ン · サ ン プ ル の 簡単な 例 を 示 し て い る 。 第 H図の ァ ク シ ョ ン · サ ン プ ル に 基づ い て作成 さ れ た ス ト ラ テ ジ ー · サ ン プ ルが第 1 5図 に 示 さ れ て い る 。 第 1 4図 に お い て , 出 力変数 に 着 目 し て, 出力変数 が変化す る と き に そ の 前後の 時点の 出力変数の 値 を 抽 出 す る こ と に よ り 第 1 5図 に 示す ス ト ラ テ ジ ー · サ ン プ ル が 得 ら れ る の は ス テ ッ プ 1 Q 5 で 説 明 し た 通 り で あ る 0
第 1 6図 は 入力変数空間 に お け る ァ ク シ ョ ン · サ ン プ ル の グ ラ フ を 示 し て い る 。 直交す る 2 つ の軸 は入力 変 数 Χ ι , X 2 で あ る 。 出 力変数 u が一定値 を 保つ 範囲 が直方形で示 さ れて い る 。
入力変数の 種類数が 3 個 の 場 台 に形成 さ れ る 3 次元 入力変数空間 に お い て, 出 力変数が一定値を 示す 範囲 は 直方体で表 わ さ れ る 。 4 次元以上の 入力空間 に お い て こ の 範囲 は ハ イ パ ー ボ ッ ク ス (超直方体) で表わ さ れ る 。 以下の 説明 で は, 入力変数空間が 2 次元で あ ろ う と , 3 次元で あ ろ う と , 長方形, 直方体 を 含 め て す ベ て の 場合 に つ い てハ イ パ ボ ッ ク ス と い 用 ¾ 用 い る 。
第 16図 に お い て ハ イ パ ー ボ ッ ク ス Q は 出力変数 u が
90で あ る 範囲で あ り , ヽ ィ 一 ボ ッ ク ス Q + 1 は u が
100 で あ る 範囲で あ る 。
—つ の ハ イ パ ー ボ ッ ク ス を規定す る 入, 出 力変数の 集 合 を ス ト ラ テ ジ ー · ケ ー ス と い う 。 ス ト ラ テ ン 一 · ケ ー ス Q を例 に 挙げ る と , れ は 次 の よ う に 表現 さ れ る 。
d o u = 90 wh e n x
1 = 3. 0 , X 2 20
u n t i l x
1 = 4. 5 , X 2 45 m i n x
1 = 3. 0 , X 2 10 m a I x
1 一 . 5 9 4 δ こ れ は 次の こ と を 意味 し て い る
「入力力《 X i = 3 , X , 20の と き 力、 ら ' χ 1 =
4. 5 , x 9 = 45の と き ま で 出 力 u を 90と せ よ 。 た だ し , 入力 X i に は最小値 3. 0 力、 ら 取大 鼠 5. 5 ま で, 入 力 X 2 に は 最 小値 1 Gか ら 最 大 値 45ま で 許 容 範 囲 が あ る 。 」
— ボ ッ ク ス Q の - — つ の 頂 点 は ( X 1 の 最小 値, X 2 の 最小値) ( こ の 点 は 第 16図 で M I Ν と 表わ さ れ て い る ) で 表 わ さ れ, 二 れ に 対 向 す る も う 一 つ の 頂点 は ( X i の 最大値, X 2 の 最大値) (第 16図で
M A X で示 さ れて い る ) で表わ さ れ る ス ト ラ テ ジ ー · ケ ー ス Q + 1 は次の よ う に し て表わ さ れ る の は 容易 に 理解で き ょ う 。
do u = 100 wh e n x , = 3. 5 , χ = 47 un t i l χ ^ = 1. 0 , χ = 34 m 1 η χ = 1 χ = 34 m a x x = 3. 5 , x = 48 ク レ ー ン 制御の 上述 の 例 に つ い て, 入力空間 に お け る ァ ク シ ョ ン · サ ン プ ル の グ ラ フ が第 17図 に 示 さ れて い る 。 簡単の た め に , 位置 X と 振れ角 S を 直交す る 二 軸 と す る 2 次元入力変数空間が示 さ れて い る ( も ち ろ ん, 正確 に は X , Θ , d 0 を相互 に 直交す る 軸 と す る 3 次元入力変数空間 に 描 く べ き で あ る ) 。
第 13図 に 示す ス ト ラ テ ジ ー · サ ン プル を 参照 し て得 ら れ る ス ト ラ テ ジ ー · ケ ー ス が第 a 図 お よ び第 18 b 図 な ら び に 第 19図 に 示 さ れて い る 。 第 18 a 図 お よ び第 18 b 図 は 10進 数 の 数 値 を 用 い て 表 現 さ れ (第 9 図 参 照) , 第 19図 は バ イ ナ リ ィ ' ス ト リ ン グ に 変換 さ れ た も の で あ る (第 1 G図参照) 。
第 18 a 図 お よ び第 18 b 図 に お い て, do : に fee レヽ 出 力変数値が表わ さ れ て い る 。 ま た, wh e n : に铳 い て 入 力 変 数 値 ( wh e nの 項 の 入 力 変数値 ) が 入 力 変数 X , Θ , d 0 の 順 に 配列 さ れて い る 。 同 じ よ う に , u n t i l : , m i n : , m a i : に 続 い て 入 力 変 数値 ( u n t i l の 項, m i n の 項 お よ び m a x の 項 の 入 力 変数値) 力 X , θ , d S の 順 に 配列 さ れ て い る 。
第 19図 に お い て , 2 進化 コ ー ド は 各変数値 に つ い て 8 ビ ッ ト で 表 わ さ れ て い る 。 そ の 順序 は , wh e n © 項 に お け る X , Θ , d 0 の 入力 変数値, u n t i l の 項 に お け る x , Θ , d 0 の 入 力 変 数値, m i n の 項 に お け る X , Θ , の 入 力 変 数 値 , m a x の 項 に お け る x , Θ d 0 の 入力 変数値, そ し て 最後 に ( の 項 に お け る 出 力 変数値 と な っ て い る 。
第 17図 に お い て , ス ト ラ テ ジ ー · ケ ー ス 1 , 2 , 3 , 4 , 7 お よ び 8 は 長方形 — ボ ッ ク ス ) で は な く , 一 点 で 表 わ さ れ て い る ( こ の 点 も ハ イ ボ ッ ク ス の 概念 に 含 ま せ る こ と が で き る ) 。
ス テ ッ プ 107
上 述 の 処 理 に よ っ て 得 ら れ た す べ て の ス ト ラ 丁 ジ一 . ケ ー ス の う ち , 不要 な ス ト ラ テ ジ ー . ケ 一 ス を 無視 (削除 ) す る こ と に よ り , ス ト ラ テ ジ ー * ケ一 ス を 最適化 す る 。
ま ず , 同 じ 出 力 変 数 値 を も つ 2 つ 以 上 の ス ト ラ テ ジ ー · ケ ー ス が あ る か ど う か を み ¾ 。 同 じ 出 力 変数 値 を も つ 2 つ 以上 の ス ト ラ テ ジ ー · ケ ー ス が見付 か つ た 場 合 に は , そ れ ら の ケ ー ス の 近 さ を 入 力 変数値 ( ハ ィ パ ー ボ ッ ク ス ) に 基 づ い て 算 出 す る 。 近 さ が近 い ^¾ '口 に は ス ト ラ テ ジ ー · ケ ー ス の 一つ を 残 し て他が削 除 さ れ る か, ま た は こ れ ら の ス ト ラ テ ジ ー ' ケ ー ス が 合体 さ れ る 。
た と え ば 2 つ 以上 の ス ト ラ テ ジ ー · ケ ー ス に 関 す る ハ イ ノ、。 ー ボ ッ ク ス が 完 全 に 重 な っ て い る 場 台 に は , そ れ ら の う ち の 一つ を 残 し て他 の も の を 削 除す る 。
さ ら に た と え ば 2 つ 以上の ハ イ パ ー ボ ッ ク ス の 距離 を 算 出 し , 得 ら れ た 距離 が所定 の し き い 値以 下 の 場 合 に は , こ れ ら の ス ト ラ テ ジ ー · ケ ー ス を 合体 ( ま た は 融 合) さ せ る か, ま た は 一方 を 削除 す る 。 距離 と し て は , た と え ば w h e n ( ま た は u n U 1 ) の 項 に 記述 さ れ た 入 力 変数値 に よ つ て 表 わ さ れ た 入力 変数空間 に お け る 点 ( ま た は M I N の 点 も し く は M A X の 点 ) に 着 目 し , 2 つ の ケ ー ス に お け る こ れ ら の 点 間 の 距離が 用 い り れ よ つ o
ス ト ラ テ ジ ー · ケ ー ス の 数 を で き る だ け 少 な く す る 方法 に は 多 く の ア プ ロ ー チ が あ る 。
ク レ ー ン 制 御 シ ス テ ム の 例 で は , 出 力 変数 U の 値 が 等 し い ケ ー ス は 多 数 あ る が , そ れ ら の 入 力 変 数 値 に よ っ て 表 わ さ れ る 点 は か な り 遠 い の で, す べ て の ス ト ラ テ ジ一 · ケ ー ス 力く 必要 な も の と し て残 さ れ る 。
ス テ ッ プ 1 0 8
上述 の よ う に し て 作 成 さ れ た ゴ ー ル · ケ ー ス と ス ト ラ テ ジ一 · ケ ー ス と を 次 の よ う に し て 関連づ け る ( ま た は リ ン ク , 連鎖) 。
サ ブ ス テ ッ プ 1 0 8 - 1 第 1 番 目 の ス ト ラ テ ジ ー . ケ ー ス に つ い て (第 1 8 a 図 ま た は 第 1 9図参照) , そ の 状態量 を 各 ゴ ー ル · ケ 一 ス の ゴ ー ル 条 件 と 比 較 す る 。 状 態 量 と は , ス ト ラ テ ジ ー · ケ ー ス に お い て , w h e nの 項 の 入力 変数値 ( ま た は u n t i l の 項 の 入 力 変数値 ) で あ る 。
サ ブ ス テ ッ プ 1 0 8 - 2
状態量が ゴ ー ル 条 件 を 満 た さ な い 場 合 に は , 第 1 番 目 の ス ト ラ テ ジ ー · ケ ー ス を そ の ゴ ー ル · ケ ー ス に 関 連 づ け る 。 ゴ ー ル 条 件 を 満 た し た 場 合 に は , 関連 づ け を し な い 。 ゴ ー ル 条 件 を 満 た し た と き に は も は や そ の ゴ ー ノレ に 向 う ス ト ラ テ ジ 一 は 不要 で あ る カヽ ら で あ る 。
サ ブ ス テ ッ プ 1 Q 8 - 3
サ ブ ス テ ッ プ 1 (U - 1 お よ び 1 0 8 - 2 の 処理 を す ベ て の ス ト ラ テ ジ ー · ケ ー ス に つ い て繰返す。
状態量 と し て w h e nの 項 の 入 力 変数 値 を と り , 上述 し た ク レ ー ン 制 御 シ ス テ ム の 例 で 具 体 的 に こ の 関連 づ け を 行 っ て み る 。
第 1 2図 に 示 す ゴ ー ル · ケ ー ス と 第 1 8 a 図 お よ び 第 1 8 b 図 に 示す ス ト ラ テ ジ ー · ケ ー ス を 参 照す る 。
ス ト ラ テ ジ ー · ケ ー ス 番 号 1 に お い て w h e nの 項 に 記 述 さ れ た 入 力 変 数 X の 値 は 0. (H Q で あ る 。 こ の 値 は ゴ ー ル 番 号 1 の ゴ ー ル 条 件 ( 「 最 終 状 態 」 の 値 =
9. 4 6 7 ) を 満 た さ な い 。 し た 力く つ て , ス ト ラ テ ジ ー ' ケ ー ス 1 は ゴ ー ル ' ケ ー ス 1 と 連鎖 す る 。 ス ト ラ テ ジ ー · ケ ー ス 1 の wh e nの 項 に 記述 さ れ た 入 力変数 0 の 値 は 0. 000 で あ る 。 こ の 値 は ゴ ー ル ' ケ 一 ス 2 の ゴ ー ル条件 ( 「過渡状態」 の 幅 = [— 0. 029 , 0. 02 ? ] ) を 満 た す 。 し た 力く つ て , ス ト ラ テ ジ ー ' ケ ー ス 1 は ゴ ー ノレ · ケ ー ス 2 と は 連 鎖 し な い。
同 じ よ う に , ス ト ラ テ ジ ー · ケ ー ス 1 の wh e nの 項 に 記述 さ れ た 入 力 変数 d 0 の 値 は O. Q Q Q で あ る 。 こ の 値 は ゴ ー ノレ · ケ ー ス 3 の ゴ ー ル 条 件 ( 「過 渡状態」 の 幅 = [ - 0. 039 , 0. 042 ] ) を 満 た す 。 した が っ て , ス ト ラ テ ジ ー · ケ ー ス 1 は ゴ ー ノレ ' ケ ー ス 3 と も 連鎖 し な い 。 - 結局, ス ト ラ テ ジ ー ' ケ ー ス 1 は ゴ ー ノレ ' ケ ー ス 1 と の み 連鎖 ( リ ン ク ) す る こ と に な る 。
同 様 に し て, ス ト ラ テ ジ ー · ケ ー ス 2 〜 10に つ い て 上述 し た サ ブ ス テ ッ プ 1 Q 8 - 1 , 1 08 - 2 を 適用 す る と 次 の 結 果 が 得 ら れ る 。
ス ト ラ テ ジ ー · ケ ー ス 2 , 3 , 4 , 5 , 6 , 8 , 9 お よ び 1 0は ゴ ー ル · ケ ー ス 1 と の み 連 鎮 す る 。
ス ト ラ テ ジ ー · ケ ー ス 7 は ゴ ー ル · ケ ー ス 1 お よ び
7 と 連 鎖す る 。
逆 に ゴ ー ル · ケ ー ス の 観点 力、 ら い え ば , 次 の よ う に な る 。
ゴ ー ル · ケ ー ス 1 は ス ト ラ テ ジ ー · ケ ー ス 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 お よ び 10と 連鎖す る 。 ゴ一ル • ケ ー ス 2 は い ず れ の ス ト ラ テ ジ ー · ケ ー ス と も 連鎖 し な い 。
ゴ一ル • ケ ー ス 3 は ス ト ラ テ ジ ー · ケ ー ス 7 と の み 連鎖す る
こ れ り の 連 鎖 を メ モ リ 上 で記憶す る た め に , 各 ス ト ラ テ シ • ケ ー ス は そ れ に 連鎖 す る す べ て の ゴ ー ル · ケ ー ス へ の ポ イ ン タ を も つ (各 ス ト ラ テ ジ ー · ケ ー ス を 表 わす デ ー タ に 対 応 し て こ れ ら の ボ イ ン タ が記憶 さ れ る ) o ま た , 各 ゴ ー ル · ケ ー ス は そ れ に 連鎖す る す ベ て の ス ト ラ テ ジ ー · ケ ー ス へ の ポ イ ン タ を も つ 。
第 2 0図 は 各 ゴ ー ル · ケ ー ス に 設 け ら れ た ボ イ ン タ を 示 し て い る 。 P L は ス ト ラ テ ジ ー · ケ ー ス へ の ポ イ ン タ で あ る こ と を 示 し , P L に 続 く 矢 印 で 示 さ れ た 番 号 が そ の ゴ一 ノレ · ケ ー ス に 連鎖す る ス ト ラ テ ジ一 · ケ ー ス の 番 で あ る 。 ポ イ ン タ P L は 各 ゴ ー ゾレ * ケ ー ス 番 号 ご と 設 け ら れ て い る 。
第 2 1 は 各 ス ト ラ テ ジ ー · ケ ー ス 番号 に 対応 し て 設 け ら れ た ポ イ ン タ を 示 し て い る 。 ポ イ ン タ P H は ゴ ー ル · ケ ス へ の ボ イ ン タ で.あ り , ボ イ ン 夕 P L は ァ ク シ ヨ ン ケ ー ス へ の ポ イ ン タ ( 次 に 説 明 す る ) で あ る 。 P H に 続 く 矢 印 で 示 さ れ た 番号 が, そ の ス ト ラ テ ジ 一 · ケ ー ス に 連 鎮 す る ゴ ー ル · ケ ー ス の 番 号 で あ 0 , P L に 続 く 矢 印 で 示 さ れ た 番 号 が そ の ス ト ラ テ ジ 一 · ケ ー ス に 連鎖す る ァ ク シ ョ ン · ケ ー ス の 番号 で あ る 。
こ の よ う な ス ト ラ テ ジ ー · ケ ー ス と ゴ ー ノレ · ケ ー ス の 連鎖 は 必要 に 応 じ て 表示 さ れ, ま た は プ リ ン ト さ れ る 。 ユ ー ザ は 必 要 に 応 じ て , 新 し い ス ト ラ テ ジ ー · ケ ー ス を 加 え た り , 不要 な ス ト ラ テ ジ ー · ケ ー ス を 削 除 し た り , 既存 の ス ト ラ テ ジ ー · ケ ー ス を 修正 し た り す る こ と が で き る 。
ス テ ッ プ 109
ス ト ラ テ ジ ー · ケ ー ス と ァ ク シ ョ ン · ケ ー ス と が 次 の よ う に し て 関 連 づ け ら れ る ( ま た は リ ン ク , 連 鎖) 。 ア ク シ ョ ン · ケ ー ス と は . 各 時点 ご と の 入 力 変 数値 と 出 力変数値 の 組 を 指 し , ア ク シ ョ ン ' サ ン プ ル と ほ ぼ 同 義 で あ る 。 た だ し , 全 く 同 じ 値 を も つ ァ ク シ ョ ン · サ ン プ ル は ァ ク シ ョ ン · ケ ー ス か ら は 取 除 か れ る 。
サ ブ ス テ ッ プ 109 - 1
第 1 番 目 の ア ク シ ョ ン · ケ ー ス に つ い て (第 9 図 ま た は 第 10図 参 照 ) , そ の 状 態 量 を 各 ス ト ラ テ ジ ー · ケ ー ス の ス ト ラ テ ジ ー 条件 と 比較 す る 。 伏態量 と は ァ ク シ ヨ ン · ケ ー ス に お け る 入 力 変数値 の 組 を 指す。 ス ト ラ テ ジ 一 条件 と は , ス ト ラ テ ジ ー · ケ ー ス に お け る wh e nの 項 に 記述 さ れ た 入 力 変数値 の 組 か ら u n t i l の 項 に 記述 さ れ た 入 力 変数値 の 組 ま で の 範囲 を さ す ( こ れ を 以下, wh e i!〜 un t i l の 範囲 と い う ) 。 サ ブ ス テ ッ プ 1 ΰ 9 - 2
状態量が ス ト ラ テ ジ ー 条件 を 満 た し た 場 合 に は , 第 1 番 目 の ァ ク シ ョ ン · ケ ー ス を そ の ス ト ラ. テ ジ ー · ケ ー ス に 関連 づ け る 。
サ ブ ス テ ッ プ 109 - 3
サ ブ ス テ ッ プ 109- 1 お よ び 109-2 の 処理 を す ベ て の ァ ク シ ョ ン · ケ ー ス に つ い て繰返 す 。
上述 し た ク レ ー ン 制 御 シ ス テ ム の 例 に つ い て こ の 関 連 づ け を 試 み て み ょ う 。 第 9 図 と 第 18 a 図 お よ び 第 18 b 図 を 参 照 す る 。 第 9 図 に 示 さ れ た す べ て の ァ ク シ ヨ ン · サ ン プ ル 力く ア ク シ ョ ン · ケ ー ス で あ る 。 た だ し , 第 1 番 目 の ア ク シ ョ ン ' ケ ー ス は ト ロ リ の 静止状 態 を 表 わ し て い る の で, モ デ ル形成 お よ び 問題解決 に お い て 意 味 を も た な い の で, 関連 づ け は 特 に 行 う 必要 は な い 。 し た 力く つ て , 第 2 番 目 の ア ク シ ョ ン · ケ ー ス か ら 関連 づ け を 行 う こ と に す る 。
第 2 番 目 の ア ク シ ョ ン · ケ ー ス ( 番号 2 ) に お い て は 状態量 は x = Q. 0 , ^ = 0. 000 , d e = Q. 0 () () で あ り , こ れ は ス ト ラ テ ジ ー · ケ ー ス 番 号 1 の wh e i!〜 u n t i l の 範 囲 ( X = 0. 000 〜 0. 000 , Θ = 0. 000 〜 0. 000 , d Θ = 0. 000 〜 0 · 000 ) に 入 る 。 ま た , ァ ク シ ョ ン · ケ ー ス 2 の 状態量 は 他 の ス ト ラ テ ジ ー · ケ ー ス の w h e n〜 ! 1 n t i 1 の 範 囲 内 に は 入 ら な い 。 し た 力く つ て , ア ク シ ョ ン · ケ ー ス 2 は ス ト ラ テ ジ ー · ケ ー ス 1 O
4 4
に の み 連鎖す る 。
同 様 に 上述 の 関連づ け 処理 ( サ ブ ス テ ッ プ 1 09 - 1 , 109 - 2 ) を 他 の ア ク シ ョ ン ' ケ ー ス に つ い て行 う .と , 第 22図 に 示 す連鎖 が 得 ら れ る 。
第 22図 に お い て , ァ ク シ ョ ン · ケ ー ス 番号 ご と に , そ の ァ ク シ ョ ン · ケ ー ス に 連 鎖 す る ス ト ラ テ ジ ー · ケ ー ス 力《 ポ イ ン タ P H で 示 さ れ て い る 。 ポ イ ン タ P H は 連 鎖 す る ス ト ラ テ ジ ー · ケ ー ス へ の ァ ク シ ョ ン · ケ ー ス の ポ イ ン タ で あ る 。 ポ イ ン タ P H の 矢 印 に 続 い て連鎖す る ス ト ラ テ ジ ー · ケ ー ス 番号 が記述 さ れ て い る o
各 ス ト ラ テ ジ ー · ケ ー ス ご と に , そ れ に 連鎖す る ァ ク シ ヨ ン · ケ ー ス へ の ポ イ ン タ も ま た 設 け ら れ る 。 こ れ 力く 第 2 1図 に P L で 示 す も の で あ る 。 各 ス ト ラ テ ジ ー · ケ ー ス 番号 ご と に 設 け ら れ た ボ イ ン タ P L の 矢 印 で 示 さ れ る 番号 が , そ の ス ト ラ テ ジ ー · ケ ー ス に 連 鎖す る ァ ク シ ョ ン · ケ ー ス の 番号 で あ る 。
ゴ ー ル · ケ ー ス , ス ト ラ テ ジ ー · ケ ー ス お よ び ァ ク シ ョ ン · ケ ー ス 力、 ら 構成 さ れ る モ デ ル は , 第 23図 に 示 す よ う に ヒ エ ラ ノレ キ ー 構造 を も ち , か つ こ れ ら が相 互 に 関 連 づ け ら れ て い る こ と が分 る 。 ゴ ー ノレ ( ケ ー ス ) が 最 も 上 位 の レ ベ ル に あ る ( 概 念 的 意 志 決 定 の レ べ ル) 。 ス ト ラ テ ジ ー ( ケ ー ス ) は 中 間 の レ べ ノレ に あ る ( 状 況 に 応 じ た 意 志 決 定 の レ ベ ル ) 。 ア ク シ ョ ン ( ケ 一 ス ) が最 も 下位の レ ベ ルで あ る (経験的意志決 定 の レ ベ ル ) 。
ス テ ツ プ 1 1 0
こ の よ う に し て作成 さ れ た モ デ ル は, も し 必要で あ れ ば, 検証 さ れ る 。
こ の 検 証 の た め {: : , 上記 の ァ ク シ ヨ ン · サ ン プ ル (教師 デ ー タ ; 第 9 図 ま た は第 1 0図) が用 い ら れ る 。 ァ ク シ ヨ ン · サ ン プ ル の 入力 デ ー タ を 用 い て, 後述す る 問題解決 モ ー ド に お け る 手順 と 同 じ よ う に し て ( ス テ ツ 7 1 2 1 の 入力変数値取込み お よ びェ ン コ 一デ ィ ン グ は 必ず し も 必要な い で あ ろ う ) , 先 に 作成 し た モ デ ル に し た が っ て 出力 が計算 さ れ る 。 得 ら れ た 出 カ デ ー 夕 と ァ ク シ ヨ ン · サ ン プ ル に お け る 出 力 デ ー タ と が比 較 さ れ る 。 両 出 力 デ 一 夕 が等 し い 値 ま た は近 い値で あ れ ば ( た と え ば両 デ 一 夕 の 差が所定 し き い 値以下) , モ デ ル は正 し く 作成 さ れ た も の と み な さ れ よ う 0
さ ら に 必要で あ れ ば, 入力 デ ー タ を ラ ン ダ ム に 生成 し, こ の 入力 デ ー タ を シ ス テ ム に 与え て も よ い 。 シ ス テ ム は モ デ ル に し た が っ て 出 力 デ ー タ を 算 出 し , か つ 出 力す る 。 出 力 デ ー タ が妥 当 な も の か ど う 力、 は ユ ー ザ に よ つ て判断 さ れ よ ラ o
ス テ ツ プ 1 1 1
最後 に , 作成 さ れ た モ デ ル を 表わ す デ ー タ 力 メ モ リ の 作業 エ リ ア か ら フ ア イ ル · エ リ ア に 転送 さ れて 保存 さ れ る 。
( 4 ) 問題解決 モ ー ド に お け る 処理
問題解決 モ ー ドで は . シ ス テ ム が上述 し た モ デ ルを 用 い て実際 に 制御対象 を制御 (運転 ま た は操作) す る (第 2 図参照) 。 そ の処理手順が第 2 4図 に 示 さ れ て い る
ス テ ッ プ 1 2 1
セ ン サ 1 1, 1 2 , l k (第 2 図) , ま た は 位 置 X を 表わ す 回転 ト ラ ン ス デ ュ ー ザ の 出力, 振れ角 を 表わす 角 度 セ ン サ 等 (第 4 図, 第 5 図) か ら の 入力変数値が シ ス テ ム 1 に 取込 ま れ る 。 こ れ ら の 入力値 は 適 当 な 分解能 で 2 値 デ ー タ に 変換 さ れ る 。
ス テ ッ プ i 2 2
得 ら れ た 入 力 デ ー タ を モ デ ル に 含 ま れ る ゴ ー ル ' ケ ー ス の ゴ ー ル条件 と 比較す る 。 入力 デ ー タ に よ っ て 満 た さ れ な い ゴ ー ル条件を も つ ゴ ー ル · ケ ー ス が あ れ ば , そ の ゴ ー ル · ケ ー ス に 連 鎖 し て い る ス ト ラ テ ジ 一 · ケ ー ス を リ ス ト · ア ッ プす る 。
こ の ス テ ッ プ に お い て , —シ ス テ ム は ¾ 切 な ゴ ー ル · ケ ー ス を 決定す る こ と に よ り 概念 レ ベ ル の 意志決定を 行 な お う と し て い る 。 こ れ に よ つ て 適 切 な ス ト ラ テ ジ ー · ケ ー ス が決定 さ れ る こ と に な る 。
ス テ ッ プ 1 2 3
入力 デ ー タ を上記 ス テ ッ プ 1 2 2 で選択 さ れ た ス ト ラ テ ジ ー · ケ ー ス と 比較す る 。 入 力 デ ー タ と マ ッ チ す る ス ト ラ テ ジ ー 条件 ( ff h e i!〜 u n t i l の 範囲 ) を も つ ス ト ラ テ ジ一 · ケ ー ス が あ れ ば, そ の ス ト ラ テ ジ ー · ケ ー ス に 連鎖 す る ア ク シ ョ ン ' ケ ー ス を リ ス ト · ア ッ プす る o
こ の 段 階 で シ ス テ ム は 適切 な ス ト ラ テ ジ ー · ケ ー ス を 決定す る こ と に よ り 状況 的 な 意志決定 を 行 な お う と し て い る 。 こ の 結果, 適切 な ア ク シ ョ ン · ケ ー ス が決 定 さ れ る 。
ス テ ッ プ 1 2 4
ス テ ッ プ 1 2 3 で選択 さ れ た ァ ク シ ョ ン · ケ ー ス の う ち , 入 力 デ ー タ と 完全 に 一致 し て い る 入 力 変数値 を も つ も の , ま た は 入 力 デ ー タ に 最 も 近 い 入 力 変数値 ( 入 力変数空 間 に お け る 距離 が最 も 短 い も の ) を も つ も の を 候補 と し て 選択す る 。
ス テ ッ プ Π 5
一般 的 に は ス テ ッ プ 1 2 4 で一つ の 候捕 が選択 さ れ る の で , そ の 候補 の ア ク シ ョ ン · ケ ー ス の 出 力 変数値 を 出 力 デ ー タ と し て 出 力 す る 。 2 つ 以上 の 候補 が も し 存 在 し た 場 合 に は 最 も 良 い も の を 選択 し , そ の ァ ク シ ョ ン · ケ ー ス の 出 力 変 数 値 を 出 力 デ ー タ と し て 出 力 す る 。 た と え ば , 2 つ の ス ト ラ テ ジ ー · ケ ー ス の ハ イ パ ー ボ ッ ク ス が一部 重複 し て お り , こ の 重 複 し て い る 空 間 に 入 力 デ ー タ が 含 ま れ て い る と き に は , 2 つ の ス ト ラ テ ジ ー · ケ ー ス か ら 異 な る 出 力 変 数値 が導 き 出 さ れ る 。 こ の 場 合 に は , た と え ば入力 デ ー タ と ハ ィ パ 一 ボ ッ ク ス の 端点 ( M I N ま た は M A X 点) と の 距離 の 近 い 方 の ヽ ィ パ ー ボ ッ ク ス か ら 得 ら れ る 出 力値が選択 さ れ る 。 ま た は , 処理 に お い て 最初 に 得 ら れ た 出 力 値 が無 条件 に 選択 さ れ る 。
第 25図 は ク レ ー ン 制 御 シ ス テ ム を モ デ ル に し た が つ て 自 動運 転 し た と き に 得 ら れ た 入力 デ ー タ , モ デ ル に し た が っ て 計 算 さ れ た 出 力 デ ー タ ( ク レ ー ン 制 御 に 用 い ら れ た ) , お よ び 出 力 値 を 得 る 過 程 で 選 択 さ れ た ケ ー ス の 番 号 を 示 し て い る 。 最 も 右側 の 力 ッ 内 の 3 つ の 数字 ( 矢 印 で つ な が れ て い る ) は , 前 の 方 力、 ら , 選 択 さ れ た ゴ ー ル · ケ ー ス 番号 . ス ト ラ テ ジ 一 • ケ 一 ス 番 号 , お よ び ア ク シ ョ ン · ケ ー ス 番 号 を 示 し て い る c
た と え ば 入 出 力 デ ー タ 番 号 1 の 入 力 デ 一 タ は X = 0. 000 , ^ = 0. 000 , d 0 = O. O O O で あ る o れ ら の 入力 デ ー タ は ゴ 一 ソレ · ケ ー ス 1 の ゴ 一 ゾレ 条件 を た さ ず, 他 の ゴ ー ル ' ケ ー ス 2 , 3 の ゴ ー ノレ 条件 を た し て い る ( 第 Π図参照 ) 。 し た が っ て , ゴ ー ル · ケ ー ス 1 に 連 鎖 す る 10個 の ス ト ラ テ ジ ー · ケ ー ス 1 . 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 お よ び 1 0が 選 択 さ れ る ( 第 20図参照) ( ス テ ッ プ 1 Π ) 。
こ れ ら の ス ト ラ テ ジ ー · ケ ー ス 1 〜 10の う ち , 上記 入 力 デ ー タ を 満 た す ス ト ラ テ ジ 一 条件 を 持つ も の は ス ト ラ テ ジ ー · ケ ー ス 1 の み で あ る ( 第 1 8 a 図参照) 。 し た 力く つ て , ス ト ラ テ ジ ー · ケ ー ス 1 に 連鎖す る ァ ク シ ヨ ン · ケ ー ス 2 力く 選 択 さ れ る ( 第 2 1図 参 照 ) ( ス テ ツ プ 1 2 3 ) o
し た 力《 つ て , ア ク シ ョ ン · ケ ー ス 2 の 出 力 変 数 値 u = 4 0 . 0 0 が 出 力 デ ー タ と し て 出 力 さ れ る ( 第 9 図 参 照) ( ス テ ッ プ 1 2 4 , 1 2 5 ) 。
第 2 5図 に お い て 入 , 出 力 デ ー タ 番 号 8 〜 1 0の 入 力 デ ー タ X は ゴ ー ノレ · ケ ー ス 1 の ゴ ー ル 条 件 を 満 た さ な い ば 力、 り か, 入 力 デ ー タ S 力く ゴ ー ル · ケ ー ス 2 の ゴ ー ル 条 件 も 満 た し て い な い 。 し た が っ て , ゴ ー ル · ケ 一 ス 1 と 2 力く選択 さ れ る 。 し 力、 し な 力《 ら , ゴ ー ル · ケ 一 ス 2 に 連鎖 す る ス ト ラ テ ジ ー · ケ ー ス は 存在 し な い 。 し た 力く つ て , ゴ ー ノレ · ケ ー ス 1 に 連 鎖 す る ス ト ラ テ ジ ー · ケ ー ス 1 〜 1 0の 中 力、 ら , 入力 デ ー タ 力く マ ッ チ す る ス ト ラ テ ジ ー · ケ ー ス が選択 さ れ る 。 入, 出 力 デ ー 夕 番号 8 〜 1 0の 全 入 力 デ ー タ X , Θ , d 0 を 満 た す ス ト ラ テ ジ 一 条 件 を も つ ス ト ラ テ ジ ー · ケ ー ス は 存在 し な い 。 こ の よ う な 場 合 に は , 入 力 デ ー タ に 最 も 近 い ( た と え ば一 部 の 入 力 変数 に つ い て の 入 力 デ ー タ 力く ス ト ラ テ ジ 一 条 件 を 満 た す, ま た は , 入 力 デ ー タ に 最 も 近 い 距 離 に あ る ハ イ ノヽ0 — ボ ッ ク ス を 持 つ ) ス ト ラ テ ジ ー · ケ ー ス が選択 さ れ る 。 ス ト ラ テ ジ ー · ケ ー ス 5 が選択 さ れ る 。
ス ト ラ テ ジ ー · ケ ー ス 6 に 連 鎖 す る ア ク シ ョ ン · ケ ー ス の う ち で (第 21図参照) , 各入, 出 力 デー タ 番 号 8 , 9 , 10の 入力 デ ー タ に最 も 近 い 入力変数値を も つ ア ク シ ョ ン · ケ ー ス 9 . 10 , 11が そ れ ぞ れ 選 択 さ れ, そ れ ら の ア ク シ ョ ン · ケ ー ス の 出 力 変数値 ( u = 40, 000 ) が 出 力 デ ー タ と し て 出 力 さ れ る 。
第 24図の ス テ ッ プ 121 〜 125 の 処理 は , 制御対象か ら 入力 デ ー タ を取込 む ご と に (一定時間 ご と に ) 繰返 さ れ る 。

Claims

請求 の 範囲
1 . 対象 を 操 作 ま た は 制 御 す る ;二 と に よ り 得 ら れ る 入, 出力変数値の 組 か ら な る 一連の ア ク シ ョ ン , サ ン プルを取込み,
取込ん だ上記 ァ ク シ ョ ン · サ ン プ ル の う ち の 入力変 数値 に 着 目 し て, 入力変数値の 最初 の 値, 最後の 値, 最大値お よ び最小値 を 用 い て規定 さ れ る ゴ ー ゾレ · ケ ー ス を 入力変数 ご と に 作成 し ,
取込ん だ上記 ア ク シ ョ ン · サ ン プ ル の う ち の 出 力変 数値 に 着 目 し て, 出 力変数値が同一性を 保つ 入力変数 の 範囲 を 抽 出 し , 各範囲 に つ い て , 出 力変数値 に 入力 変数の 範囲を結 びつ け る ス ト ラ テ ジ ー · ケ ー ス を 作成 し ,
上記 ゴ ー ル · ケ ー ス の 条件 を 満 た さ な い 入力変数範 囲を も つ ス ト ラ テ ジ ー · ケ ー ス を そ の ゴ ー ル · ケ ー ス に連鎖 さ せ,
上記 ス ト ラ テ ジ ー · ケ ー ス に お け る 入力変数範囲 に 入 る 入力変数値 を も つ ア ク シ ョ ン • ケ ー ス を そ の ス ト ラ テ ジ 一 · ケ ー ス に 連鎖 さ せ ,
こ れ に よ つ て, 相互 に 連鎖 し た ゴ ー ル · ケ ー ス , ス ト ラ テ ジ ー · ケ ー ス お よ び ァ ク シ ョ ン · ケ ー ス 力ヽ ら 構 成 さ れ る ヒ エ ラ ル キ ー 構造の モ デルを形成す る 推論方 法 0
2 . 取込ん だ上記 ア ク シ ョ ン · サ ン プル を 所定桁の バ イ ナ リ イ · デ ー タ に 変換す る ス テ ッ プを 含む, 請求 の 範囲第 1 項 に 記載の 方法。
3 . ゴ ー ル · ケ ー ス を 作成す る 上記 ス テ ッ プが,
仮の ゴ ー ノレ · ケ ー ス を作成す る ス テ ッ プ と ,
作成 し た ゴ ー ル · ケ ー ス を 出力す る ス テ ッ プ と , ユ ー ザ に よ る 評価 を 受 け た後の ゴ ー ル · ケ ー ス を最 終的 な ゴ ー ル . ケ ー ス と す る ス テ ッ プ と 力、 ら 構成 さ れ る , 請求の 範囲第 1 項 に 記載の方法。
4 . 上記の ユ ー ザ に よ る 評価が, ゴ ー ル · ケ ー ス の 追 加, 削除 お よ び変更 を 含 む, 請求の 範囲第 3 項 に 記載 の 方法。
5 . ス ト ラ テ ジ ー · ケ ー ス を 作成 す る 上 記 ス テ ッ プ が,
時間順序で配列 さ れ た 出 力変数値の 隣 り 合 う も の を 順次比絞 し , 変化が あ つ た と き に そ の 前後の 出 力変数 値 を ス ト ァ し て ス ト ラ テ ジ一 · サ ン プ ル を 作成す る ス テ ツ 7 と ,
上記 ス ト ラ テ ジ ー · サ ン プ ル を, 上記 ア ク シ ョ ン · サ ン プ ノレ を 参 照、 し て , d 0 ··· w h e n ··· u n t i I ··· in i n "- m a x の 形 式を も つ ス ト ラ テ ジ ー · ケ ー ス に 変換す る ス テ ツ プ と ,
変換後の 上記 ス ト ラ テ ジ ー · ケ ー ス を最適化す る ス テ ツ プ と か ら 構成 さ れ る , 請求の範囲第 1 項 に 記載 の 方法
6 . 上記 ス ト ラ テ ジ ー · ケ ー ス を 最適化 す る ス テ ッ プ は , 同 じ 出 力 変数 値 を も つ 2 つ 以上 の ス ト ラ テ ジ ー · ケ ー ス の 数 を 減少 さ せ る も の で あ る , 請求 の 範囲 第 5 項 に 記載 の 方 法。
7 . ス ト ラ テ ジ ー ' ケ ー ス と ゴ ー ノレ ' ケ ー ス の 連 鎖 を , 各 ス ト ラ テ ジ ー · ケ ー ス に 設 け ら れ た ゴ ー ル ' ケ ー ス へ の ポ イ ン タ と , 各 ゴ ー ル · ケ ー ス に 設 け ら れ た ス ト ラ テ ジ ー · ケ ー ス へ の ポ イ ン タ と に よ り 実現 す る , 請求 の 範囲 第 1 項 に 記載 の 方法。
8 . ア ク シ ョ ン · ケ ー ス と ス ト ラ テ ジ ー , ケ ー ス と の 連鎖 を , 各 ア ク シ ョ ン · ケ ー ス に 設 け ら れ た ス ト ラ テ ジ ー · ケ ー ス へ の ポ イ ン タ と , 各 ス ト ラ テ ジ ー ' ケ 一 ス に 設 け ら れ た ァ ク シ ョ ン · ケ ー ス へ の ポ イ ン タ と に よ り 実現 す る , 請求 の 範 囲 第 1 項 に 記載 の 方法。
9 . 全 く 同 じ 入 , 出 力 変 数 値 の 組 力、 ら な る ァ ク シ ョ ン * サ ン プ ル を , そ の 一つ を 残 し て 他 を 削 除 す る こ と に よ り ア ク シ ョ ン · ケ ー ス を 作 成 す る , 請 求 の 範 囲 第 1 項 に 記載 の 方 法。
1 0 . 上記一連 の ア ク シ ョ ン · サ ン プ ル を , 上記 モ デ ル を 検証 の た め の デ ー タ と し て 用 い , 上記 ア ク シ ョ ン · サ ン プ ル を そ の 時 間 の 順 序で一つ ずつ 取込 み ,
取込 ん だ ァ ク シ ョ ン · サ ン プ ル の 入力 変数値 に よ つ て 満 た さ れ な い 条 件 を も つ ゴ ー ル · ケ ー ス を 捜 し 出 し , 捜 し 出 し た ゴ ー ル · ケ ー ス に 連鎖 す る ス ト ラ テ ジ ー · ケ ー ス を リ ス ト し ,
上 記 の リ ス ト さ れ た ス ト ラ テ ジ ー · ケ ー ス の 中 か ら , 上記取込ん だ入力変数値 に よ っ て満 た さ れ る 条件 を も つ ス ト ラ テ ジ ー · ケ ー ス を捜 し 出 し , 捜 し 出 し た ス ト ラ テ ジ ー · ケ ー ス に 連鎖す る ァ ク シ ョ ン • ケ ー ス を リ ス ト し ,
上 記 の リ ス ト さ れ た ァ ク シ ョ ン · ケ ー ス の う ち で上 記の取込ん だ入力変数値 と 一致す る ま た は最 も 近 い 入 力変数値を も つ ァ ク シ ョ ン · ケ ー ス の 出 力変数値 を 出 力 し ,
こ れ に よ つ て上記の モ デルを検証す る , 請求の 範囲 第 1 項 に 記載の 方法。
1 1 . 請求の 範囲第 1 項 に規定 さ れた推論方法 に よ っ て 作成 さ れた モ デ ル を 用 い て対象を操作 ま た は制御す る た め の 方法で あ り ,
対象か ら 得 ら れ る 入力 デ ー タ を順次取込み
取込ん だ入力 デ ー タ に よ っ て満た さ れ な い 条件 を も つ ゴ ー ル · ケ ー ス を 捜 し 出 し , 捜 し 出 し た ゴ ー ル · ケ ー ス に連鎖す る ス ト ラ テ ジ ー · ケ ー ス を リ ス ト し , 上 記 の リ ス 卜 さ れ た ス ト ラ テ ジ ー · ケ ー ス の 中 か ら , 上記の 取込ん だ入力 デ ー タ に よ っ て 満 た さ れ る 条 件を も つ ス ト ラ テ ジ ー · ケ ー ス を捜 し 出 し , 捜 し 出 し た ス ト ラ テ ジ ー · ケ ー ス に 連鎖す る ァ ク シ ョ ン · ケ 一 ス を リ ス ト し ,
上記 の リ ス ト さ れ た ア ク シ ョ ン · ケ ー ス の う ち で 上 記 の 取込 ん だ入 力 デ ー タ と 一致 す る ま た は 最 も 近 い 入 力変数値 を も つ ァ ク シ ョ ン · ケ ー ス の 出 力変数値 を 出 力 し ,
上記処理 を 入 力 デ ー タ を 取込 む ご と に 繰返 す, 制御 方 法。
1 2 . 対 象 を 操 作 ま た は 制 御 す る こ と に よ り 得 ら れ る 入, 出 力 変数値 の 組 力、 ら な る 一連 の ア ク シ ョ ン * デ一 夕 を 取込 み ,
取込 ん だ 上記 ァ ク シ ョ ン · デ ー タ の う ち の 入 力 変数 値 に 着 目 し て , 上記操作 ま た は 制 御 が め ざ す ゴ ー ル を 入 力変数 ご と に 作成 し ,
取込 ん だ上記 ァ ク シ ョ ン · デ ー タ の う ち の 出 力 変数 値 に 着 目 し て , 上記 ゴ ー ル を 達成す る た め の ス ト ラ テ ジ ー を 作 成 し ,
上記 ゴ ー ル と そ の ゴ ー ル に 未 だ達 し て い な い ス ト ラ テ ジ 一 と を 連鎖 さ せ ,
上 記 ス ト ラ テ ジ 一 と そ れ を 実 現 す る ァ ク シ ョ ン · デ ー タ と を 連鎖 さ せ ,
こ れ に よ つ て , 相 互 に 連 鎖 し た ゴ ー ル , ス ト ラ テ ジ ー お よ び ァ ク シ ョ ン カ、 ら 構成 さ れ る ヒ エ ラ ノレ キ ー 構 造 の モ デ ル を 形成す る 推論方 法。
1 3 . 対 象 を 操 作 ま た は 制 御 す る こ と に よ り 得 ら れ る 入, 出 力変数値 の 組力、 ら な る 一連の ア ク シ ョ ン · サ ン プ ル を 取込ん で記憶す る 手段,
取込ん だ上記 ァ ク シ ョ ン · サ ン プル の う ち の 入力変 数値 に 着 目 し て, 入力変数値の最初 の 値, 最後の 値, 最大値 お よ び最小値 を 用 い て規定 さ れ る ゴ ー ル · ケ ー ス を入力変数 ご と に 作成す る 手段,
取込ん だ上記 ァ ク シ ョ ン · サ ン プ ル の う ち の 出 力変 数値 に 着 目 し て, 出 力変数値が同一性を 保つ 入力 変数 の 範囲 を抽 出 し , 各範囲 に つ い て, 出力変数値 に 入力 変数の 範囲 を結 びつ け る ス ト ラ テ ジ ー · ケ ー ス を 作成 す る 手段,
上記 ゴ ー ル · ケ ー ス の 条件を満た さ な い 入力変数範 囲 を も つ ス ト ラ テ ジ ー · ケ ー ス を そ の ゴ ー ル · ケ ー ス に連鎖 さ せ る 手段, な ら び に
上記 ス ト ラ テ ジ ー · ケ ー ス に お け る 入力変数範囲 に 入 る 入 力 変数値 を も つ ァ ク シ ョ ン · ケ ー ス を そ の ス ト ラ テ ジ 一 · ケ ー ス に 連鎖 さ せ る 手段を備 え ,
こ れ に よ つ て, 相互 に連鎖 し た ゴ ー ノレ ' ケ ー ス , .ス ト ラ テ ジ ー · ケ ー ス お よ び ァ ク シ ョ ン · .ケ ー ス 力ヽ ら 構 成 さ れ る ヒ エ ラ ル キ ー 構造の モ デ ル を 形成す る 推論 シ ス 7" ム o
1 4 . 上記 ア ク シ ョ ン ' サ ン プ ル の取込み 記憶手段 は, 取込ん だ上記 ァ ク シ ョ ン · サ ン プル を 所定桁の バ イ ナ リ イ · デ ー タ に 変換す る 手段を 含む, 請求 の 範囲第 U 項 に 記載 の シ ス テ ム
1 5 . ゴ一 ル · ケ ー ス を 作成す る 上記手段が,
仮 の ゴ一 ル • ケ 一 ス を 作成す る 手段 と ,
作成 し た ゴ一 ル · ケ ー ス を 出 力す る 手段 と ,
ュ一ザ に よ る 評価 を 受 け た後 の ゴ ー ル · ケ ー ス を最 終的 な ゴ一 ル • ケ 一 ス と し て記憶す る 手段 と か ら 構成 さ れ る , 請求 の 範囲 第 Π項 に 記載の シ ス テ ム 。
1 6 . 上記の ュ一ザ に よ る 評価が, ゴ 一 ノレ · ケ ー ス の 追 加, 削除 お よ び変更 を 含 む, 請求 の 範囲第 1 5項 に 記載 の シ ス テ ム 。
1 7 . ス ト ラ テ ジ 一 · ケ ー ス を作成す る 上記手段が, 時間順序で配列 さ れ た 出 力変数値 の 隣 り 台 う も の を 順次比較 し , 変化が あ つ た と き に そ の 前 後 の 出 力 変数 値 を ス ト ァ し て ス ト ラ テ ジ一 · サ ン プ ル を 作成す る 手 段 と ,
上記 ス ト ラ テ ジ 一 • サ ン プ ル を , 上記 ア ク シ ョ ン · サ ン プ ル を 参照 し て , α 0 ··· w h e n■·· u n t I I ··· m i n ··· m a x の 形 式 を も つ ス 卜 う テ ジ 一 · ケ ー ス に 変 換 す る 手 段 と ,
変換後 の 上記 ス ト ラ テ ジ 一 · ケ ー ス を 最適化す る 手 段 と か ら 構成 さ れ る , 請求 の 範囲第 1 3項 に 記載の シ ス ア ム 。
1 8 . 上記 ス ト ラ テ ジ 一 · ケ ー ス を 最適化す る 手段 は , 同 じ 出 力変数値 を も つ 2 つ 以上の ス ト ラ テ ジ ー · ケ ー ス の 数 を 減少 さ せ る も の で あ る , 請求 の 範囲 第 17項 に 記載 の シ ス テ ム 。
19. 上記 ス ト ラ テ ジ ー · ケ ー ス と ゴ ー ノレ ' ケ ー ス を 連 鎖 さ せ る 手 段 は , ス ト ラ テ ジ ー · ケ ー ス と ゴ ー ル · ケ ー ス の 連鎖 を , 各 ス ト ラ テ ジ ー · ケ ー ス に 設 け ら れ た ゴ ー ノレ · ケ ー ス へ の ポ イ ン タ と , 各 ゴ ー ル · ケ ー ス に 設 け ら れ た ス ト ラ テ ジ 一 ' ケ ー ス へ の ボ イ ン タ と に よ り 実現 す る , 請求 の 範囲第 13項 に 記載 の シ ス テ ム 。
20. 上記 ア ク シ ョ ン · ケ ー ス と ス ト ラ テ ジ ー ' ケ ー ス と を 連鎖 さ せ る 手段 は , ア ク シ ョ ン · ケ ー ス と ス ト ラ テ ジ ー ' ケ ー ス と の 連鎖 を , 各 ア ク シ ョ ン ' ケ ー ス に 設 け ら れ た ス ト ラ テ ジ ー ' ケ ー ス へ の ポ イ ン タ と , 各 ス ト ラ テ ジ ー · ケ ー ス に 設 け ら れ た ァ ク シ ョ ン · ケ ー ス へ の ポ イ ン タ と に よ り 実現す る , 請 求 の 範囲 第 13項 に 記載 の シ ス テ ム 。
21. 全 く 同 じ 入 , 出 力 変 数 値 の 組 力、 ら な る ァ ク シ ョ ン · サ ン プ ル を , そ の 一つ を 残 し て 他 を 削 除 す る こ と に よ り 上記 ア ク シ ョ ン · ケ ー ス が作成 さ れ る , 請求 の 範 囲 第 13項 に 記載 の シ ス テ ム 。
22. 検証 の た め の デ ー タ と し て 用 い る た め に , 上記 ァ ク シ ョ ン * サ ン プ ル を そ の 時 間 の 順序 で 一 つ ずつ 取込 む 手段,
取 込 ん だ ァ ク シ ョ ン · サ ン プ ル の 入 力 変数値 に よ つ て 満 た さ れ な い 条 件 を も つ ゴ ー ル · ケ ー ス を 捜 し 出 し , 捜 し 出 し た ゴ ー ル · ケ ー ス に 連 鎖 す る ス ト ラ テ ジ ー · ケ ー ス を メ モ リ 上で リ ス 卜 す る 手段,
上記 の リ ス ト さ れ た ス ト ラ テ ジ ー · ケ ー ス の 中 か ら , 上記取込ん だ入力変数値 に よ っ て満 た さ れ る 条件 を も つ ス ト ラ テ ジ ー · ケ ー ス を捜 し 出 し , 捜 し 出 し た ス ト ラ テ ジ ー · ケ ー ス に 連鎖す る ァ ク シ ョ ン · ケ ー ス を メ モ リ 上で リ ス ト す る 手段, な ら び に
上記 の リ ス ト さ れ た ア ク シ ョ ン · ケ ー ス の う ち で上 記の 取込ん だ入力変数値 と 一致す る ま た は 最 も 近 い 入 力変数値 を も つ ァ ク シ ョ ン · ケ ー ス の 出 力 変数値 を 出 力す る 手段 を 備 え ,
こ れ に よ つ て上記 の モ デル を検証す る , 請求の 範囲 第 1 3項 に 記載の シ ス テ ム 。
2 3 . 請 求 の 範 囲 第 1 3項 に 規 定 さ れ た 推 論 シ ス テ ム に よ っ て作成 さ れ た モ デ ル を用 い て対象を 操作 ま た は制 御す る た め の シ ス テ ム で あ り ,
対象 か ら 得 ら れ る 入力 デー タ を 順次取込む手段, 取込ん だ入力 デ ー タ に よ っ て満 た さ れ な い 条件 を も っ ゴ ー リレ . ケ ー ス を 捜 し 出 し , 捜 し 出 し た ゴ ー ル · ケ ー ス に 連鎖す る ス ト ラ テ ジ ー · ケ ー ス を メ モ リ 上で リ ス 卜 す る 手段,
上 記 の リ ス 卜 さ れ た ス ト ラ テ ジ ー · ケ ー ス の 中 か ら , 上記 の 取込ん だ 入力 デ ー タ に よ っ て 満 た さ れ る 条 件を も つ ス ト ラ テ ジ ー · ケ ー ス を捜 し 出 し , 捜 し 出 し た ス ト ラ テ ジ ー · ケ ー ス に 連鎖す る ァ ク シ ョ ン · ケ ー ス を メ モ リ 上で リ ス ト す る 手段,
上記 の リ ス 卜 さ れ た ァ ク シ ョ ン · ケ ー ス の う ち で上 記の 取込ん だ入力 デ ー タ と 一致す る ま た は最 も 近 い 入 力変数値を も つ ァ ク シ ョ ン · ケ ー ス の 出力変数値を 出 力す る 手段, な ら び に
上記処理を入力 デ ー タ を取込む ご とに繰返す よ う 制 御す る 手段,
を 備 え た 制 御 シ ス テ ム 。
PCT/JP1995/002562 1994-12-28 1995-12-14 Procede et systeme d'inference mettant en ÷uvre un modele hierarchique et procede et systeme de gestion associes WO1996020438A1 (fr)

Priority Applications (3)

Application Number Priority Date Filing Date Title
KR1019960704169A KR970700877A (ko) 1994-12-28 1995-12-14 계층적 모델을 사용하는 추론 시스템과 방법 및 제어 시스템과 방법(Method and system for inference using hierarchy model, and method and system for control)
DE69532237T DE69532237T2 (de) 1994-12-28 1995-12-14 Verfahren und Vorrichtung für Rückschlüsse, welche ein hierarchisches Modell verwendet und Verfahren und Vorrichtung zur Steuerung
EP95940434A EP0749057B1 (en) 1994-12-28 1995-12-14 Method and system for inference using hierarchy model, and method and system for control

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US36497094A 1994-12-28 1994-12-28
US08/364,970 1994-12-28

Publications (1)

Publication Number Publication Date
WO1996020438A1 true WO1996020438A1 (fr) 1996-07-04

Family

ID=23436932

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP1995/002562 WO1996020438A1 (fr) 1994-12-28 1995-12-14 Procede et systeme d'inference mettant en ÷uvre un modele hierarchique et procede et systeme de gestion associes

Country Status (6)

Country Link
EP (1) EP0749057B1 (ja)
JP (1) JP3196938B2 (ja)
KR (1) KR970700877A (ja)
CN (1) CN1106601C (ja)
DE (1) DE69532237T2 (ja)
WO (1) WO1996020438A1 (ja)

Families Citing this family (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8290721B2 (en) 1996-03-28 2012-10-16 Rosemount Inc. Flow measurement diagnostics
US7949495B2 (en) 1996-03-28 2011-05-24 Rosemount, Inc. Process variable transmitter with diagnostics
US7623932B2 (en) * 1996-03-28 2009-11-24 Fisher-Rosemount Systems, Inc. Rule set for root cause diagnostics
US8112565B2 (en) 2005-06-08 2012-02-07 Fisher-Rosemount Systems, Inc. Multi-protocol field device interface with automatic bus detection
US20070068225A1 (en) 2005-09-29 2007-03-29 Brown Gregory C Leak detector for process valve
US7953501B2 (en) 2006-09-25 2011-05-31 Fisher-Rosemount Systems, Inc. Industrial process control loop monitor
US7750642B2 (en) 2006-09-29 2010-07-06 Rosemount Inc. Magnetic flowmeter with verification
US8898036B2 (en) 2007-08-06 2014-11-25 Rosemount Inc. Process variable transmitter with acceleration sensor
JP5141905B2 (ja) * 2007-12-28 2013-02-13 オムロン株式会社 安全マスタ
US9207670B2 (en) 2011-03-21 2015-12-08 Rosemount Inc. Degrading sensor detection implemented within a transmitter
ITMI20120101A1 (it) * 2012-01-27 2013-07-28 Saipem Spa Sistema elettronico, metodo e programma di controllo di una rampa di varo a configurazione variabile di un natante di posa per varare una tubazione su un letto di un corpo d'acqua
US9052240B2 (en) 2012-06-29 2015-06-09 Rosemount Inc. Industrial process temperature transmitter with sensor stress diagnostics
US9602122B2 (en) 2012-09-28 2017-03-21 Rosemount Inc. Process variable measurement noise diagnostic
CN112200319A (zh) * 2020-10-20 2021-01-08 北京计算机技术及应用研究所 一种实现无人车导航避障的规则推理方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0315902A (ja) * 1989-03-13 1991-01-24 Hitachi Ltd 知識抽出方法およびプロセス運転支援システム
JPH05204405A (ja) * 1992-01-23 1993-08-13 Matsushita Electric Ind Co Ltd 入出力関係推定装置
JPH05257695A (ja) * 1992-03-16 1993-10-08 Hitachi Ltd ルール自動生成方式
JPH06281302A (ja) * 1993-03-26 1994-10-07 Yamatake Honeywell Co Ltd 冷凍機の稼働エネルギー量推定装置
JPH06314104A (ja) * 1993-04-30 1994-11-08 Yamatake Honeywell Co Ltd 非線形制御装置

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2118885C (en) * 1993-04-29 2005-05-24 Conrad K. Teran Process control system

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0315902A (ja) * 1989-03-13 1991-01-24 Hitachi Ltd 知識抽出方法およびプロセス運転支援システム
JPH05204405A (ja) * 1992-01-23 1993-08-13 Matsushita Electric Ind Co Ltd 入出力関係推定装置
JPH05257695A (ja) * 1992-03-16 1993-10-08 Hitachi Ltd ルール自動生成方式
JPH06281302A (ja) * 1993-03-26 1994-10-07 Yamatake Honeywell Co Ltd 冷凍機の稼働エネルギー量推定装置
JPH06314104A (ja) * 1993-04-30 1994-11-08 Yamatake Honeywell Co Ltd 非線形制御装置

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
ON-LINE PROCESS SIMUL TECHNIQUES IND CONTROL, issued 1985, GRASTY R S, "Simulation of Large-Scale Control System by the Use of an Automated Hierarchical Methodology", pp. 97-106. *
PROG CYBERN SYST RES, Vol. 6, issued 1982, CARLSSON C, "Adaptive Multigoal Control: An Approach to Applied Systems Analysis", pp. 195-208. *
PROG CYBERN SYST RES, Vol. 6, issued 1982, KELLERMAYR K H, "On Hierarchical Modelling of Large-Scale Systems", pp. 284-294. *
See also references of EP0749057A4 *

Also Published As

Publication number Publication date
EP0749057A1 (en) 1996-12-18
DE69532237D1 (de) 2004-01-15
CN1106601C (zh) 2003-04-23
CN1142269A (zh) 1997-02-05
JP3196938B2 (ja) 2001-08-06
EP0749057B1 (en) 2003-12-03
EP0749057A4 (en) 1997-07-02
KR970700877A (ko) 1997-02-12
DE69532237T2 (de) 2004-10-21

Similar Documents

Publication Publication Date Title
WO1996020438A1 (fr) Procede et systeme d'inference mettant en ÷uvre un modele hierarchique et procede et systeme de gestion associes
WO2020166303A1 (ja) モデル生成装置、方法及びプログラム並びに予測装置
US20190286990A1 (en) Deep Learning Apparatus and Method for Predictive Analysis, Classification, and Feature Detection
WO2023102962A1 (zh) 一种训练端到端的自动驾驶策略的方法
CN113826051A (zh) 生成实体系统零件之间的交互的数字孪生
CN111199272A (zh) 一种面向智能车间的自适应调度方法
CN113807596B (zh) 一种信息化工程造价的管理方法及系统
CN112053008B (zh) 基于时序复合运算模型的梯级船闸通航调度运行方法
Ngo et al. Hardware Design for Intelligent IoT Approach to Optimize Parking Slots
Jun et al. Evolutionary neural network for learning of scalable heuristics for pickup and delivery problems with time windows
Filipic et al. A combined machine learning and genetic algorithm approach to controller design
JP3400062B2 (ja) プラント制御装置及びトンネル換気制御装置
JPH0635706A (ja) ファジィ推論システムおよび方法,ならびにパターン入力型メンバーシップ値生成装置および方法
Patjoshi Design and implementation of embedded based elevator control system
JPH0981205A (ja) 学習システム
WO2021048243A1 (en) Method for generating an adapted task graph
CN117273395B (zh) 一种基于深度学习的生产计划与调度方法
CN115079706B (zh) 人机协同控制移动式机器人智能避障方法和系统
CN114139592A (zh) 用于训练量化分类器的方法和设备
Li et al. Driverless Planning and Decision-Making based on GRU
Alfa et al. Analysis of fuzzy and neural networks expert systems in forecasting stock prices
CN116467587A (zh) 用于连续机器学习不同任务的序列的方法和设备
Pawar et al. Adaptive Traffic Signal Controlling With Transfer Learning
CN117787291A (zh) 利用常识体现的多模式人工智能问答和对话的系统和方法
Tang et al. Regulating the imbalance for the container relocation problem: A deep reinforcement learning approach

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 95191798.6

Country of ref document: CN

AK Designated states

Kind code of ref document: A1

Designated state(s): CN JP KR

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): AT BE CH DE DK ES FR GB GR IE IT LU MC NL PT SE

WWE Wipo information: entry into national phase

Ref document number: 1995940434

Country of ref document: EP

121 Ep: the epo has been informed by wipo that ep was designated in this application
WWP Wipo information: published in national office

Ref document number: 1995940434

Country of ref document: EP

WWG Wipo information: grant in national office

Ref document number: 1995940434

Country of ref document: EP