JP2559879B2 - Fuzzy controller - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Outline] A fuzzy controller that calculates and outputs a control operation amount corresponding to a control state amount according to a fuzzy control rule, and realizes an arithmetic function of a marginal product as an antecedent operation of the fuzzy control rule. At the same time, for the purpose of easily enabling the change to the logical product operation which has been conventionally used, the antecedent operation means is composed of an input unit for distributing an input signal value, an input from the preceding layer and A plurality of input units are used as input units with a basic unit that receives an internal state value to be multiplied with respect to an input to obtain a product sum value and obtains an output by function-converting the product sum value. age,
And one or a plurality of basic units as an intermediate layer and one or a plurality of intermediate layers, and one basic unit as an output layer, between the input layer and the foremost intermediate layer, between the intermediate layers and between the final layers. A hierarchical network in which an inner coupling is configured between the intermediate layer and the output layer, and a value at which a limit product value of the input signal value is output from the output value is set as the internal state value of the inner coupling. It is composed of a structural part.

[Industrial applications]

The present invention relates to a fuzzy controller that calculates and outputs a control operation amount corresponding to a control state amount by executing fuzzy inference according to a fuzzy control rule, and in particular, as an antecedent operation described in the fuzzy control rule. The present invention relates to a fuzzy controller that realizes a function of calculating a marginal product and can easily change to a logical product operation that has been conventionally used.

Fuzzy control is spreading as a new control processing method. This fuzzy control expresses a control algorithm including ambiguity such as human judgment in if-then format, and executes this control algorithm according to fuzzy inference, so that the control operation amount from the detected control state amount Is calculated to control the controlled object. In order to absorb the ambiguity of the control algorithm, the fuzzy controller that realizes this fuzzy control needs to be configured so that the arithmetic function described in the control algorithm can be easily changed.

[Conventional technology]

The fuzzy control rule is of the form if x ₁ is big and x ₂ is small then y ₁ is big (the IF part is called the antecedent part and is the part that describes the conditions for the control state quantity such as temperature data. , THEN part is called the consequent part, which is the part that describes the condition of the control operation amount such as the operation end), and the fuzzy controller describes the control logic of the controlled object. It manages such multiple fuzzy control rules, and quantifies the meaning of ambiguous linguistic expressions such as “large” and “small” described in each fuzzy control rule as a membership function. It will take a configuration to manage.

When a fuzzy controller receives a control state quantity such as temperature data or water level data from a control target, first, a membership function of the control state quantity being managed (becomes a membership function of the antecedent part). The membership function value (truth value) of the control state quantity given by Next, in accordance with the antecedent operation of selecting the minimum value, the process of determining the applied value to the antecedent in each fuzzy control rule is executed. That is, to explain using the example of the fuzzy control rules above, "x ₁ is big"
The truth value of "0.8" and the truth value of "x ₂ is small" are "0.
In the case of 5 ", processing is performed so as to determine this" 0.5 "as the applicable value to the consequent part of the fuzzy control rule in accordance with the antecedent part calculation.

Then, the fuzzy controller follows each fuzzy control rule given for the same membership function of the same control operation amount (becomes the consequent membership function) according to the consequent operation of selecting the maximum value. A process of determining an applicable value for the membership function from the applied value for the subject part is executed. That is, in the fuzzy control rule described above, “0.5” is applied to the consequent part “y ₁ is big”.
When "0.6" is applied to "y ₁ is big" by another fuzzy control rule by using the fuzzy control rule, this "0.6" is used as the control value y according to the consequent operation. It is processed so as to be determined as the applied value to the membership function of "big" of ₁ .

Subsequently, the fuzzy controller reduces the membership function of the control operation amount according to the determined applied value, and obtains the center of gravity of the function sum of the reduced membership functions for the same control operation amount, etc. In accordance with the process (1), a process of calculating a control operation amount which is a fuzzy inference value and outputting it to the operation end or the like is executed. That is, the membership function of "big" of the control operation amount y ₁ is reduced according to the determined applied value, and the membership function such as "small" of the control operation amount y ₁ obtained from another fuzzy control rule is reduced to The control value is reduced according to the determined application value, and the control operation amount that is the output as the fuzzy controller is calculated according to the processing such as obtaining the barycenter of the figure of the function sum of the membership functions.

In the fuzzy controller having such a configuration, conventionally, as described above, the fuzzy controller is configured to have the antecedent operation function for executing the operation of selecting and outputting the minimum input signal value. It was Then, the operation function of the antecedent part is configured to be implemented by a program means.

[Problems to be Solved by the Invention]

However, the fuzzy control rule is generated according to the knowledge of the target process of the operator who is familiar with the control of the control target. In the future, it is not possible to generate a fuzzy control rule that can achieve the desired control from the beginning, and the fuzzy control rule generated will be tuned by trial and error while being evaluated by simulations and field tests. Therefore, there is no choice but to take the procedure of completing the fuzzy control rule suitable for the controlled object.

From this, it may happen that the antecedent part arithmetic function also requires another kind of arithmetic function, rather than the logical product operation of selecting and outputting the minimum value of the input values. However,
Since the conventional fuzzy controller has only the antecedent operation function of the logical product operation, there is a problem that it cannot meet such a demand. In order to deal with this demand, it is possible to deal with the necessary antecedent part arithmetic function by preparing it as a subroutine in the device in advance. However, if such a corresponding method is adopted, the memory capacity will be compressed. Another problem will come up.

The present invention has been made in view of such circumstances,
A new fuzzy controller that realizes the arithmetic function of the marginal product as the antecedent operation described in the fuzzy control rules, and that can easily change to the logical product operation that has been used conventionally. It is intended to be provided.

[Means for solving the problem]

 FIG. 1 is a block diagram showing the principle of the present invention.

In the figure, 10 is a fuzzy controller equipped with the present invention, 11 is a fuzzy rule management unit that manages fuzzy control rules that describe the control logic for a control target, 12
Is a fuzzy rule selection unit that sequentially selects the fuzzy control rules managed by the fuzzy rule management unit 11 to be processed. 13 is a membership function management unit of the antecedent part. It manages the truth value of the antecedent part membership function that quantifies the linguistic expression about the described control state quantity, and 14 is the consequent part membership function management part, which is described in the fuzzy control rule. That manages the truth value of the consequent part membership function that quantifies the linguistic expression regarding the controlled operation amount,
Reference numeral 15 is an input data acceptance unit that executes acceptance processing of the control state quantity data collected from the controlled object, and 16 is an antecedent part truth value calculation unit that processes the selected fuzzy control rule as a processing unit. , Which calculates the truth value of the antecedent membership function of the input control state quantity,
Reference numeral 17 denotes an antecedent part operation part, which is an applied value to the consequent part of the fuzzy control rule to be processed by performing an antecedent part operation on the calculated truth value of the antecedent part membership function. 18 is the latter half operation part, and from the applied value to the consequent part of each fuzzy control rule given for the same consequent part membership function, to the subsequent part membership function Determining the applicable value of
Reference numeral 19 denotes an inference value calculation unit, which applies the fuzzy control rules that do not have the same consequent part output from the antecedent part calculation part 17 and the same consequent part output from the consequent part calculation part 18. With the applied values of the fuzzy control rules having and, for example, the consequent membership function is reduced according to those applied values, and the figure centroid of the functional sum of the contracted consequent membership functions is calculated. The control manipulated variable data, which is a fuzzy inference value, is calculated by obtaining it.

The antecedent operation unit 17 of the present invention receives one or more inputs and an internal state value to be multiplied with respect to the inputs to obtain a product sum value, and performs a function conversion of the product sum value. A hierarchical network structure unit 20 configured to execute an antecedent operation by being configured by a hierarchical network structure of a plurality of basic units 1 for obtaining output values, and an internal state assigned to an internal connection of the hierarchical network structure of this hierarchical network structure unit 20. It is composed of an internal state value management unit 21 that manages values.
This hierarchical network structure unit 20 has a basic unit 1 and an input unit 1'which distributes an input signal value as a structural unit, and an input layer composed of a plurality of input units 1'-h and one or a plurality of input layers. An input layer 1'-h and a basic unit 1-i are provided, which are provided with an intermediate layer composed of the basic unit 1-i and provided in one or a plurality of stages, and an output layer composed of one basic unit 1-j. , Between the basic units 1-i, and the basic unit 1-i
A hierarchical network structure in which i and the basic unit 1-j are internally connected to each other is adopted. And the internal state value management unit
Reference numeral 21 has a configuration for managing the value of the internal state value that the hierarchical network structure unit 20 operates to output the marginal product value of the input signal value.

[Action]

The hierarchical network structure unit 20 of the present invention inputs the input to the input layer according to the data conversion function defined by its own hierarchical network structure and the internal state value assigned to the internal connection of the hierarchical network structure. It operates so as to calculate the limit product value of the signal values and output it to the consequent operation unit 18. That is, z ₁ , z ₂ , ..., z _n as input values
When (0 ≦ z _i ≦ 1) is input, the following Yager
Limit product value Y defined by setting p = 1 in the equation
Is calculated and output.

_{Y = z 1 ∧z 2 ∧ ...} ∧z n = 1- [1∧ ((1-z 1) p + (1-z 2) p + ... + (1-z n) p) 1 / P] This As described above, the antecedent part computing section 17 of the present invention calculates the marginal product value of the truth value of the antecedent part membership function calculated by the antecedent part truth value calculating part 16 to obtain the consequent part. Since it is processed so that it is determined as an applicable value, there is a way to realize more appropriate control in the case where the antecedent part arithmetic function that executes the conventional AND operation cannot perform appropriate control. Will be given.

The data conversion function of the hierarchical network structure unit 20 is characterized in that even if the hierarchical network structure is the same, the content of the operation can be freely changed by changing the value of the internal state value assigned to the internal connection. Therefore, by registering the value of the internal state value that operates to output the logical product of the input signal values in the internal state value management unit 21, it is possible to immediately return to the same configuration as the conventional fuzzy controller. Of.

〔Example〕

 Hereinafter, the present invention will be described in detail according to examples.

FIG. 2 shows an embodiment of the present invention. As described with reference to FIG. 1, reference numeral 16 in the figure is an antecedent part truth value calculation part, 17 is an antecedent part calculation part, 20 is a hierarchical network structure part, and 21 is an internal state value management part.

The antecedent part truth value calculation unit 16 performs processing so as to calculate the truth value of the antecedent part membership function of the input control state quantity. That is, the antecedent part of the selected fuzzy control rule describes IF X ₁ is Small, X ₂ is Medium, ..., X _n is Big for the control state quantities X ₁ , X ₂ , X _n. On the other hand, if the controlled object is given the control state quantity data of X ₁ = A ₁ , X ₂ = A ₂ , ..., X _n = A _n , the antecedent part truth value calculation part 16 As shown in FIG. 2, the control state quantity X ₁
A according to the membership function that quantifies the "Small" of A
_The membership function value Z ₁ with the value ₁ is specified, and the membership function value Z ₂ with the value A ₂ is specified according to the membership function that quantifies “Medium” of the control state quantity X ₂ and the control state According to the membership function that quantifies “Big” of the quantity X _n , the processing for identifying the membership function value Z _n having the value A _n is performed. In FIG. 2, for the sake of convenience of description, these membership function values (truth values) are illustrated in parallel, but in reality, they are calculated one by one in time series.

The hierarchical network structure unit 20 performs the antecedent part operation for executing the marginal product operation on the truth value given from the antecedent part truth value calculation part 16, thereby performing the fuzzy control which is the object of calculation of the truth value. Processing will be performed so as to determine the applied value to the consequent part of the rule.

FIG. 3 illustrates the basic configuration of the basic unit 1 that constitutes the hierarchical network structure unit 20. As shown in this figure, the basic unit 1 is a multi-input / single-output system, in which a plurality of inputs are assigned weight values of internal coupling (corresponding to the internal state values described in FIG. 1). A multiplication processing unit 2 for multiplying, an accumulation processing unit 3 for adding all the multiplication results thereof, and a threshold processing unit 4 for subjecting the accumulated value to nonlinear threshold processing and outputting one final output. When the h layer is the front layer and the i layer is the rear layer, the accumulation processing unit 3 of the i-th basic unit 1 of the i layer executes the operation of the following equation (1) and the threshold processing unit In step 4, processing is performed so as to execute the operation of the following expression (2).

y _pi = 1 / (1 + exp (−x _pi + θ _i )) (2) where, h: unit number of layer p: input signal pattern number θ _i : threshold value of unit i of layer W _ih : h -Weight value of internal coupling between the i layers y _ph : output from the h-th unit of the h-layer with respect to the input signal of the p-th pattern Then, the hierarchical network structure unit 20 uses the basic unit 1 and the antecedent unit having such a configuration. As shown in FIG. 4, a plurality of input units 1 ′ are used as a structural unit with the input unit 1 ′ that distributes the truth value calculated by the truth value calculation unit 16.
-H, an input layer formed by one or a plurality of basic units 1-i, an intermediate layer (one stage in this embodiment) formed by one or a plurality of stages, and one basic unit 1- and an output layer constituted by j, between the input unit 1'-h and the basic unit 1-i, between the basic units 1-i, and the basic unit 1-i.
i and the basic unit 1-j are internally connected to each other to have a hierarchical network structure. Here, the number of units of the input unit 1'-h is
When the antecedent part operation part 17 is implemented by one hierarchical network structure part 20, it is matched with the condition number (that is, the required number of truth values) of the fuzzy control rule that describes the most antecedent part conditions. Will be prepared.

The hierarchical network structure unit 20 configured in this way is
To exhibit a function of converting an input signal that is input to a corresponding output signal according to a data conversion function that is defined by a hierarchical network structure and a weight value assigned by the internal connection of the hierarchical network structure. That is, processing is performed so as to execute the antecedent operation according to this data conversion function. According to the present invention, according to the data conversion function of the hierarchical network structure unit 20, the antecedent operation unit 17 is configured to calculate and output the limiting product value of the input signal value.

The hierarchical network structure unit 20 can be configured by either program means or hardware means. However, when it is configured by hardware means, the applicant has filed “Japanese Patent Application No. 63-216865 ( It is possible to use the one disclosed in “Network configuration data processing device” filed on August 31, 1988.

That is, the basic unit 1 is, as shown in FIG.
Multiplication type for multiplying the output from the preceding layer input via the input switch unit 7 and the weight value held by the weight value holding unit 8
The addition result of the D / A converter 2a, the analog adder 3a that adds the output value of the multiplication D / A converter 2a and the previous accumulated value to calculate a new accumulated value, and the analog adder 3a A sample-hold circuit 3b for holding, a non-linear function generating circuit 4a for performing non-linear conversion of the held data of the sample-hold circuit 3b when the accumulation processing is completed, and an analog of the non-linear function generating circuit 4a which is an output to a subsequent layer. It is realized by including the output holding unit 5 that holds the signal value, the output switch unit 6 that outputs the held data of the output holding unit 5, and the control circuit 9 that controls each of these processing units.

The hierarchical network structure unit 20 is realized by a structure in which the basic unit 1 having this structure is electrically connected by a single common analog bus 70 as shown in FIG. Here, in the figure, 71 is a weight output circuit for giving a weight value to the weight value holding unit 8 of the basic unit 1, 72 is an initial signal output circuit corresponding to the input unit 1 ', and 73 is a synchronization signal which is a control signal for data transfer. A synchronous control signal line for transmitting the control signal to the weight output circuit 71, the initial signal output circuit 72 and the control circuit 9, and a main control circuit 74 for transmitting the synchronous control signal.

In the hierarchical network structure unit 20 of this configuration, the main control unit 74 sequentially selects the base unit 1 of the preceding layer in time series and, in synchronization with this selection process, the output holding unit of the selected base unit 1. The final output held by 5 is processed to be output to the multiplication type D / A converter 2a of the basic unit 1 of the subsequent layer via the analog bus 70 according to the time-division transmission format. Upon receiving this input, the multiplication type D / A converter 2a of the basic unit 1 in the subsequent layer sequentially selects the corresponding weight values and performs the multiplication processing of the input value and the weight value, and the analog adder 3a and the sample hold. The accumulation processing unit 3 including the circuit 3b sequentially accumulates the multiplied values. Then, when all the accumulation processing for the basic unit 1 of the previous layer is completed, the main control circuit 74 activates the nonlinear function generating circuit 4a of the basic unit 1 of the subsequent layer to calculate the final output, The output holding unit 5 performs processing so as to hold the final output of this conversion processing result. Then, the main control circuit 74 repeats the same processing for the next succeeding layer by using this latter layer as a new preceding layer, so that the output pattern (output signal) corresponding to the input pattern (input signal) is obtained. It is processed so that it can be output.

The hierarchical network structure unit 20 is characterized in that even if the hierarchical network structure is fixed, the data conversion function can be set according to the weight value assigned to the internal connection of the hierarchical network structure. In the present invention, by utilizing this feature, the antecedent operation unit 17 is configured to execute the marginal product operation.

FIG. 7 shows an embodiment of a learning system for learning the weight value assigned to the internal connection of the hierarchical network structure unit 20. Next, the learning process for determining the weight value of the internal connection of the hierarchical network structure unit 20 will be described in detail according to the embodiment of the learning system.

In FIG. 7, reference numeral 30 is a hierarchical network simulator built on a computer, which simulates the data processing function of the hierarchical network structure unit 20, and 40 is a learning signal presentation device, which is a hierarchical network structure unit 20. Manages a learning signal group (a basic unit is a pair of a learning presentation signal and a learning teacher signal) used for learning the weight value of the internal coupling of, and further, the learning presentation signal in the learning signal group is hierarchically arranged. The learning teacher signal is presented to the network simulator 30 and the learning processing device 50 to be described later. In order to realize this processing, the learning signal presentation device 40
A learning signal storage unit 41 that manages a learning signal group, and a learning signal for learning is read from the learning signal storage unit 41, and the learning presentation signal in the learning signal is presented to the hierarchical network simulator 30, and another pair is formed. Learning teaching signal of the learning processing device 50 described later, a learning signal presentation unit 42 presenting the learning convergence determination unit 43 described below, an output signal output from the hierarchical network simulator 30, and a learning signal presentation unit 42.
The learning signal is received from the learning signal presenting unit 42, and it is determined whether the error of the data processing function of the hierarchical network simulator 30 is within the allowable range.
And a learning convergence determination unit 43 for notifying the.

Reference numeral 50 denotes a learning processing device that implements a back propagation method known as a weight value learning algorithm of a hierarchical network configuration data processing device, and is output in response to the presentation of the learning presentation signal by the learning signal presentation device 40. The error value between the output signal group from the hierarchical network simulator 30 and the learning teacher signal group presented from the learning signal presentation device 40 is calculated, and the weight value to be learned based on the error value. Is sequentially updated to learn the weight value of the internal coupling within which the error value is within the allowable range.

That is, the learning processing device 50 uses the h-layer-i-layer-3 shown in FIG. 4 according to the back propagation method.
If explained in the layered hierarchical network structure section 20,
An output signal y _pj from the output layer that is output when a learning input signal (learning presentation signal) is presented, and the output signal y
_When the learning teacher signal d _pj , which is the signal to be taken by _pj , is determined, first, the difference value [d _pj −y _pj ] between the output signal y _pj and the learning teacher signal d _pj is calculated, and then α _pj = y _pj (1-y _pj ) (d _pj -y _pj ) is calculated, and then According to the update amount ΔW _ji (t) of the weight value between the i layer and the j layer
Is calculated. Here, t represents the number of times of learning, and the reason for adding the value related to the update amount of the weight value determined in the previous update cycle is to speed up the learning.

Subsequently, the learning processing device 50 uses the calculated α _pj ,
First of all, And then According to the update amount ΔW _ih (t) of the weight value between the h layer and the i layer
Is calculated. Then, the weight value W _ji (t) = W _ji (t−1) + ΔW _ji (t) W _ih (t) = W _ih (t−1) + ΔW for the next update cycle according to the calculated update amount. By repeating the method of determining _ih (t), it is the output signal y _pj from the output layer that is output when the learning presentation signal is presented, and the output signal y _pj is the signal to be taken. The weight values W _ji and W _ih and the thresholds θ _i and θ _j that will be in agreement with the learning teacher signal d _pj will be learned.

In addition, the applicant of the present invention, as disclosed in the previous application “Japanese Patent Application No. 62-333484 (filed on Dec. 28, 1987,“ network configuration data processing device ”)”, has the h-layer on the input side. Proposed to incorporate threshold value θ into weight value W and treat threshold value θ as a weight value by providing a unit that always outputs “1” and assigns threshold value θ as a weight value to the output. did. From this, the learning of the threshold value θ is also learned by the same learning process as the learning of the weight value W.

As shown in FIG. 7, the hierarchical network simulator 30 includes an input unit 1'of the hierarchical network structure unit 20.
And a network structure management unit 31 that manages the internal connection relationship of the basic unit 1, an operation control unit 32 that controls the entire arithmetic processing, an input unit 1 ′ of the hierarchical network structure unit 20 and each unit of the basic unit 1. An input data expansion unit 33 that temporarily expands input input values, and a weight value management unit 34 that manages the weight values assigned to each inner join.
And an arithmetic execution unit 35 provided to simulate the arithmetic function of the basic unit 1, and an output data management unit 36 for managing the arithmetic result of the arithmetic execution unit 35. And
The calculation execution unit 35 includes a multiplication value calculation unit 37 that calculates a multiplication value of a weight value and an input value, a sum total value calculation unit 38 that calculates a sum value of the calculation values of the multiplication value calculation unit 37, and a sum total of these sum values. Value calculator
It is configured to include a threshold value calculation unit 39 that performs threshold processing on the calculated value of 38.

With such a configuration, the hierarchical network simulator 30 simulates the data processing function executed by the hierarchical network structure unit 20 installed in the fuzzy controller. When the hierarchical network structure unit 20 is configured by a program means, it can be realized by using such a hierarchical network simulator 30, for example.

In order to construct the hierarchical network structure unit 20 as one for executing the marginal product operation, the user first generates the input / output signal relation of the marginal product operation and learns the learning signal presentation device 40.
The process of registering in the learning signal storage unit 41 is executed. That is, the input signal and the input / output signal relationship in which the intense product value of those input signals is used as the output signal are generated and registered in the learning signal storage unit 41.

Limit product value Y for _two values z ₁ and z ₂ (0 ≦ z _i ≦ 1)
Will be represented by Y = 0∨ (z ₁ + z ₂ −1). From this, FIG. 8 shows the input / output signal relationship of this marginal product operation when the number of input units 1'-h in the input layer of the hierarchical network structure section 20 is two. In the example of FIG. 8, an example in which 81 input / output signal relationships are generated is shown.

Subsequently, the user activates the learning signal presenting device 40 by instructing to use the input signal in the registered input / output signal relationship as the learning presentation signal and the output signal associated with this as the learning teacher signal. At the same time, by starting the learning processing device 50 in correspondence with this starting process, the weight value of the weight value management unit 34 of the hierarchical network simulator 30 is set to start learning so as to realize this input / output signal relationship. Give instructions.

In accordance with this activation instruction, the hierarchical network simulator 30 operates as the hierarchical network structure unit 20 implemented (or scheduled to be implemented) in the fuzzy controller to convert the presented learning presentation signal into the data conversion specified by the weight value. The learning processing device 50 converts this according to the function and outputs it.
Upon receiving the output signal from 30, the internal connection weight value (managed by the weight value management unit 34) is updated according to the back propagation method described above, and the learning signal presentation device 40 receives the output from the hierarchical network simulator 30. The learning signal is repeatedly presented until the output signal substantially matches the learning teacher signal. By this processing, the weight value of the internal engagement of the hierarchical network structure unit 20 that realizes the registered input / output signal relationship of the marginal product operation is stored in the weight value management unit 34 of the hierarchical network simulator 30.

FIG. 9 shows learning data when learning is executed using the input / output signal relationship shown in FIG. 8 as a learning signal. Here, FIG. 9 (a) is output signal data of the hierarchical network simulator 30 for presentation of each learning presentation signal when the number of times of learning is 0 (that is, at the start of learning), FIG. 9 (b). Is the output signal data of the hierarchical network simulator 30 for the presentation of each learning presentation signal when the learning frequency is 500 times, and the learning frequency is 16 in FIG. 9 (c).
The output signal data of the hierarchical network simulator 30 for each presentation of each learning presentation signal is shown. Here, this learning was performed when the intermediate layer of the hierarchical network structure unit 20 was composed of 10 units of one-stage structure.
Further, at the start of learning, a random number value generated by the random number generating means is assigned as an initial value of the weight value (including the threshold value) of each internal connection. In this figure, for example, when the learning presentation signal “B09” of the learning signal of FIG. 8 is given to the hierarchical network simulator 30, when the number of learning times is 1628, the hierarchical network simulator 30 outputs “0.098”. It means that it was done. Here, "0.100", which is associated with this "0.098" in parentheses, is the learning teacher signal of "B09".

FIG. 10 shows a plot of the number of learning times and the error of the data processing function of the hierarchical network simulator 30. As shown in this figure, when the number of learning times exceeds 500, in response to the presentation of the learning signal in FIG. 8, the hierarchical network simulator 30 outputs a learning teacher signal roughly corresponding thereto. Then, in this state, even when something not included in the learning signal is presented, an operation is performed so as to output an appropriate output signal close to the content of the marginal product calculation.

FIG. 11 shows the learning value of the weight value and threshold value (managed by the weight value management unit 34 of the hierarchical network simulator 30) assigned to each internal connection of the hierarchical network structure unit 20 when the number of learning times is 1628. Illustrate. here,
“−1.111794” is the weight value of the internal coupling between the first unit of the input layer and the first unit of the intermediate layer, “−1.494546”
Is a weight value of the internal coupling between the unit of the output layer and the first unit of the intermediate layer, "0.156501" is the threshold value of the first unit of the intermediate layer, and "3.352931" is the threshold value of the unit of the output layer. There is.

As can be seen from FIG. 9 (c) and FIG.
If the weight value and the threshold value shown in the figure are registered in the internal state value management unit 21 of the hierarchical network structure unit 20 mounted in the fuzzy controller, the antecedent operation unit that executes the arithmetic processing by the hierarchical network structure unit 20. 17 is the antecedent part truth value calculation part
The antecedent operation of calculating and outputting the marginal product value of the truth value calculated by 16 is executed.

In some cases, due to the execution of the intense product arithmetic processing executed by the antecedent operation unit 17, it may not be possible to execute appropriate control in the antecedent operation of the AND operation used by the conventional fuzzy controller. Will also provide a way to enable more appropriate control.

On the other hand, when the weight value and the threshold value of FIG. 13 obtained by the learning signal that defines the input / output relation of the logical product operation shown in FIG. The section 20 outputs an output signal giving a logical product value of the input signals as shown in FIG. 14 with respect to the input of the learning presentation signal of FIG. As described above, in the present invention, while realizing the arithmetic function of the marginal product as the antecedent operation described in the fuzzy control rule, by simply changing the management data of the internal state value management unit 21, It is also possible to return to the antecedent operation of the logical product operation provided in the fuzzy controller.

Although the illustrated embodiment has been described, the present invention is not limited to this. For example, in the embodiment, the application example to the antecedent operation unit 17 having two inputs has been described, but the present invention is not limited to two inputs and can be applied to a multi-input one as it is. The arithmetic processing of the basic unit 1 is not limited to the threshold conversion processing. Further, the applicant of the present invention has previously filed “Japanese Patent Application No. 63-227825 (filed on September 12, 1988,
In "Learning processing system of network configuration data processing device"), an invention is disclosed in which the learning process of weight values can be realized in a shorter time by improving the back propagation method. The improved back propagation method or another weight value learning method other than the back propagation method can be used. The weight value and the threshold value may be learned using the hierarchical network structure unit 20 itself installed in the fuzzy controller without using a simulator or the like.

〔The invention's effect〕

As described above, according to the present invention, the fuzzy control operation can be easily changed to the conventional AND operation while realizing the operation function of the marginal product as the antecedent operation of the fuzzy control rule. Since it becomes possible to construct a controller, it is possible to configure a fuzzy controller that is more suitable for the controlled object, and it is possible to easily deal with the case where it is necessary to operate as a conventional fuzzy controller. .

[Brief description of drawings]

1 is a block diagram of the principle of the present invention, FIG. 2 is an embodiment of the present invention, FIG. 3 is a block diagram of a basic unit, FIG. 4 is a block diagram of a hierarchical network structure unit, and FIG. An example of a basic unit, FIG. 6 is an example of a hierarchical network structure part, FIG. 7 is an explanatory diagram of a learning system used by the present invention, FIG. 8 is an explanatory diagram of a learning signal to be generated, and FIG. And FIG. 10 is an explanatory diagram of learning data, FIG. 11 is an explanatory diagram of learning data of weight values and thresholds, FIG. 12 is an explanatory diagram of a learning signal generated when assigning a logical product operation, and FIG. 13 is a logical diagram. FIG. 14 is an explanatory diagram of the learning data of the weight value and the threshold value of the product calculation, and FIG. 14 is an explanatory diagram of the output data when operating as the logical product calculation function. In the figure, 1 is a basic unit, 1'is an input unit, 10 is a fuzzy controller, 16 is an antecedent part truth value calculation part, 17 is an antecedent part calculation part, 18 is an antecedent part calculation part, and 20 is a hierarchical network. A structure unit, 21 is an internal state value management unit, 30 is a hierarchical network simulator, 40 is a learning signal presentation device, and 50 is a learning processing device.

Front page continuation (72) Inventor Asahi Kawamura 1015 Kamiodanaka, Nakahara-ku, Kawasaki, Kanagawa Prefecture, Fujitsu Limited (72) Inventor Ryusuke Masuoka 1015, Kamedotachu, Nakahara-ku, Kawasaki, Kanagawa Prefecture (72) Invention Kazuo Asakawa 1015 Kamiodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture, Fujitsu Limited (72) Inventor Shigenori Matsuoka, No. 1, Fujimachi, Hino City, Tokyo Fuji Facom Control Co., Ltd. (72) Hiroyuki Okada, Tokyo Hino City, Tokyo Fuji Town No. 1 within Fuji Facom Control Co., Ltd. (56) Reference JP-A-2-292602 (JP, A) JP-A-3-134704 (JP, A) JP-A-3-222003 (JP, A) International Published 89/11684 (WO, A) Kanno "Fuzzy control" (Sho 63.5.30) Nikkan Kogyo Shimbun Terano et al. "Introduction to applied fuzzy system" (Head 1.9.20) Ohmsha

Claims (1)

(57) [Claims] Claim: What is claimed is: 1. An antecedent part truth value calculating part (16) for calculating a membership function value for an input control state quantity in units of fuzzy control rules, and an antecedent part truth value calculating part (16). To the output value for the fuzzy control rule having the same consequent part output from the antecedent part arithmetic part (17), which performs the logical operation on the calculated value of Consequent operation part (18) that performs logical operation and obtains output value
And an output value for the fuzzy control rule that does not have the same consequent part output from the antecedent part computing part (17), an output value output from the consequent part computing part (18), and a control operation amount. In the fuzzy controller including the inference value calculation unit (19) that calculates the control operation amount from the membership function value for the input unit (1), the antecedent operation unit (17) distributes the input signal value. ′), One or more inputs from the previous layer and an internal state value to be multiplied with respect to the inputs to obtain a sum of products value,
A basic unit (1) that obtains an output by function-transforming the sum of products is used as a structural unit, a plurality of the input units (1′-h) are used as an input layer, and one or a plurality of the basic units ( 1-i) as an intermediate layer, one or more intermediate layers are provided, and one of the basic units (1-j) is used as an output layer, and an intermediate layer between the input layer and the frontmost intermediate layer is provided. The hierarchical network structure part (20) that is internally connected between the intermediate layer and the output layer and the intermediate layer at the final stage, and the area that manages the internal state value referenced by the hierarchical network structure part (20) can be accessed from the outside. And an internal state value management unit (21) that manages an internal state value that operates so that the hierarchical network structure unit (20) outputs a marginal product value of input signal values. And fuzzy controller.