JPH03271806A - Fuzzy controller - Google Patents

Abstract

PURPOSE:To easily change an arithmetic function to an AND operation used from conventionally, while realizing an arithmetic function of a limit product as an antecedent part operation by constituting an antecedent part arithmetic means of a hierarchical network structure part to which a value by which a limit product value of an input signal value is outputted as an internal state of an internal coupling from an output layer is set. CONSTITUTION:A hierarchical network part 20 operates to calculate a limit product value of an input signal value inputted to an input layer and outputs it to a consequent part arithmetic part 18 in accordance with a data converting function prescribed by a hierarchical network structure which said part has by itself and an internal state value allocated to an internal coupling of its hierarchical network structure. In such a way, an antecedent part arithmetic part 17 executes a processing to calculate a limit product value of a truth value of an antecedent part membership function calculated by an antecedent part truth value calculating part 16 and to determine it as an application value to the consequent part. Accordingly, the motor appropriate control is realized, and by changing the value of the internal state value, its arithmetic contents can be changed freely.

Description

[Detailed Description of the Invention] [Summary] Regarding a fuzzy controller that calculates and outputs a control manipulated variable corresponding to a control state quantity according to a fuzzy control rule, a marginal product calculation function is realized as an antecedent calculation of the fuzzy control rule. However, with the aim of making it possible to easily change to the logical product operation that has been used conventionally, the antecedent operation means is divided into an input unit that distributes input signal values, an input from the previous layer, and an input unit that distributes input signal values. A basic unit that receives an internal state value to be multiplied by the input to obtain a product-sum value, and converts the product-sum value into an output to obtain an output. One or more intermediate layers are provided as an input layer and one or more basic units as an intermediate layer, one basic unit is an output layer, and an intermediate layer is provided between the input layer and the foremost intermediate layer. An internal connection is formed between the layers and between the final intermediate layer and the output layer, and a value at which the marginal product value of the input signal value is output from the output layer as the internal state value of the internal connection. It consists of a hierarchical network structure section in which the following information is set.

Regarding the fuzzy controller that calculates and outputs the control operation amount, in particular, it is possible to realize the marginal product calculation function as an antecedent operation described in the fuzzy control rule, and also to implement the conventional logical product operation. The present invention relates to a fuzzy controller that can be easily modified.

Fuzzy control is becoming popular as a new control processing method. This fuzzy control expresses a control algorithm that includes ambiguity such as human judgment in a 1f-then format, and executes this control algorithm according to fuzzy reasoning to calculate the control operation amount from the detected control state amount. It calculates and controls the controlled object. The fuzzy controller that implements this fuzzy control needs to be configured so that the arithmetic functions described in the control algorithm can be easily changed in order to absorb the ambiguity of the control algorithm.

[Industrial application field]

The present invention provides control corresponding to control state quantities by executing fuzzy inference according to fuzzy control rules. [Prior art] The fuzzy control rules are: if x, is big and x, is
5lIllall then y, is big
(The IF part is called the antecedent part and is the part that describes the conditions for control state quantities such as temperature data, and T
The HEN part is called the consequent part, and is a part that describes the conditions for the control operation amount of the operating end, etc.), and the fuzzy controller describes the control logic according to the control logic of the controlled object. In addition to managing such multiple fuzzy control rules that are prepared, the meaning of ambiguous linguistic expressions such as "large" and "small" written in each fuzzy control rule is quantified and managed as a membership function. We will adopt a configuration that does this.

When the fuzzy controller is given control state quantities such as temperature data and water level data from the controlled object, it first calculates the membership function (antecedent membership function) of the control state quantity that it is managing. Calculate the membership function value (truth value) of the control state quantity given from 0
Next, in accordance with the antecedent computation of selecting the minimum value, a process is executed to determine the applicable value for the consequent in each fuzzy control rule. In other words, to explain using the fuzzy control rule example above, 'x isbig'
The truth value of is 'o, g', rxz is small'
If the truth value of is "0.5", processing is performed to determine this "0.5" as the applied value to the consequent part of the fuzzy control rule according to the antecedent part operation. .

Subsequently, the fuzzy controller follows each fuzzy control rule given for the same membership function (which becomes the consequent membership function) of the same control manipulated variable according to the consequent operation of selecting the maximum value. A process is executed to determine the applied value for the membership function from the applied value for the subject. In other words, in the above fuzzy control rule, the consequent “ylisbig” is set to “0”.
．． 5" is the applied value, and on the other hand, if another fuzzy control rule is used to set "0.6" as the universal value for 'y+ is bigJ, then this "0.
6" is determined as the value applied to the member-parts function of r big J of the control operation amount yI.

Subsequently, the fuzzy controller reduces the membership function of the control manipulated variable according to the determined application value, and
In accordance with the process of finding the center of gravity of the figure of the function sum of this reduced membership function for the same control operation amount, the process of calculating the control operation amount, which is a fuzzy inference value, and outputting it to the operation terminal etc. I will do it. That is, the membership function of the control operation amount y, ``r big'' is reduced according to its determined application value, and the control operation amount rsm of the control operation amount y1 obtained from another fuzzy control rule is
According to the process of reducing the membership functions such as "all" according to the determined application value, and determining the center of gravity of the figure of the sum of the functions of those membership functions,
Processing is performed to calculate the control operation amount that is the output of the fuzzy controller.

Conventionally, fuzzy controllers with such a configuration are equipped with an antecedent calculation function that performs calculation processing to select and output the minimum value of the input signal value, as described above. Ta. The antecedent calculation function is implemented by a program means.

(Problems to be Solved by the Invention) However, fuzzy control rules are generated according to the knowledge of the target process of an operator who is familiar with the control of the controlled target.

The reality is that it is not possible to generate fuzzy control rules that can achieve the desired control from the beginning, and we will tune them through trial and error while evaluating the generated fuzzy control rules through simulations and on-site tests. Therefore, it is necessary to complete the fuzzy control rules that are suitable for the object to be controlled.

From now on, the antecedent part calculation function will also need a different type of calculation function, rather than a logical product operation that selects and outputs the minimum value of input values. However, conventional fuzzy controllers have a problem in that they cannot meet such demands because they only have the function of calculating the antecedent part of the AND operation. In order to cope with this request, it is possible to prepare the necessary antecedent calculation function as a subroutine in the device in advance. Another problem will arise.

The present invention has been made in view of the above circumstances, and while realizing the function of calculating a marginal product as an antecedent operation described in a fuzzy control rule, it also provides a function for calculating a marginal product as an antecedent operation described in a fuzzy control rule. The purpose is to provide a new fuzzy controller that can be easily modified.

[Means to solve the problem]

FIG. 1 is a diagram showing the principle configuration 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 control logic for a controlled object,
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 an antecedent membership function management unit that selects fuzzy control rules managed by the fuzzy rule management unit 11 as processing targets; 14 is a consequent membership function management unit which is described in the fuzzy control rule. 15 is an input data reception unit for accepting control state quantity data collected from a controlled object; The unit that executes the processing is an antecedent truth value calculation unit 16, which calculates the truth value of the antecedent membership function of the input control state quantity using the selected fuzzy control rule as a processing unit. 17 is an antecedent calculation unit which performs an antecedent calculation on the truth value of the calculated antecedent membership function to apply the antecedent calculation to the consequent of the fuzzy control rule to be processed. 18 is a consequent part calculation unit which determines the applied value of the consequent part from the applied value for the consequent part of each fuzzy control rule given for the consequent part membership function of the same -. 19 is an inference value calculation unit that determines the applied value for the membership function, and for example, the consequent part members regarding the same control operation amount according to the applied value determined by the consequent part calculation part 18. Control operation amount data, which is a fuzzy inference value, is calculated by reducing the ship function and finding the graphic center of gravity of the function sum of the reduced consequent membership functions.

The antecedent calculation unit 17 of the present invention receives one or more inputs and an internal state value by which the human power is to be multiplied, obtains a product sum value, and converts the product sum value into a function. A hierarchical network structure section 20 configured by a hierarchical network structure of a plurality of basic units 1 to obtain an output value and executes an antecedent operation, and this hierarchical network structure section 20
and an internal state value management unit 21 that manages internal state values assigned to internal connections of the hierarchical network structure. This hierarchical network structure section 20 has a basic unit 1 and an input unit 1° that distributes input signal values as constituent units, an input layer composed of a plurality of input units 1'-h, and one or more input layers. It includes an intermediate layer composed of basic units 1-i and provided in one or more stages, and an output layer composed of one basic unit 1-j, and an input unit 1''-h and a basic unit) 1-
A hierarchical network structure is adopted in which internal connections are made between basic units 1-i, 1-i, and 1-j, and between basic units 1-i and 1-j. The internal state value management unit 21 is configured to manage internal state values that cause the hierarchical network structure unit 20 to operate to output the marginal product value of the input signal values.

[Effect]

The hierarchical network structure unit 20 of the present invention processes input input to the input layer according to a data conversion function defined by its own hierarchical network structure and internal state values assigned to internal connections of the hierarchical network structure. It operates to calculate the marginal product value of the signal value and output it to the consequent calculation section 18. That is, zl as an input value. 22
．． ..., 2. When l (0≦28≦1) is input, a marginal product value Y defined by setting pt in Yager's equation below is calculated and output.

Y=z, Az, △...△Z++ =1-[1△((1-z+)'+(1-zz)'+-+
(1-za)') "'] In this way, the antecedent part calculation unit 17 of the present invention calculates the marginal product value of the truth value of the antecedent membership function calculated by the antecedent part truth value calculation unit 16. Since the process calculates and determines it as a common value for the consequent, it can be used in cases where the conventional antecedent calculation function that performs a logical AND operation cannot perform appropriate control. This will provide a way to achieve more appropriate control.

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 calculation can be changed freely 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 section 21,
It is also possible to immediately return to the same configuration as the conventional fuzzy controller.

〔Example〕

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

FIG. 2 illustrates an embodiment of the present invention. As explained in FIG. 1, 16 in the figure is an antecedent part truth value calculation section, 17 is an antecedent part calculation section, 20 is a hierarchical network structure section, and 21 is an internal state B value management section.

The antecedent truth value calculation unit 16 performs processing to calculate the truth value of the antecedent membership function of the input control state quantity. That is, if the antecedent part of the selected fuzzy control rule is the control state quantity χ1X,,X, IF L is Small, L is Medium
m, −, L is Big, and on the other hand, from the controlled object, χ+ = A +, X! = A −,...
,X,,=A.

Assuming that the control state quantity data is given, the antecedent part truth value calculation unit 16 calculates that the value A1 has according to the membership function that quantifies 'Small' of the control state quantity XI, as shown in FIG. Membership function value Z1
Specify the control state quantity X.

The membership function value Z that the value A2 has according to the membership function that quantifies the rMed i um of
2, and a process is performed to specify the membership function value ZR of the value A7 according to the membership function that quantifies the control state quantity "rBig" of the control state quantity X8. In addition, the second
In the figure, these membership function values (
Although the illustration shows that the truth values) are found in parallel, in reality they are found one by one in chronological order.

The hierarchical network structure section 20 performs an antecedent operation that performs a marginal product operation on the truth value given from the antecedent truth value calculation section 16, thereby calculating the fuzzy control that is the target of the truth value. Processing is performed to determine the applicable value for the consequent of the rule.

FIG. 3 illustrates the basic configuration of the basic unit 1 that will constitute the hierarchical network structure section 20. As shown in this figure, the basic unit 1 is a multi-car output system, which multiplies multiple inputs by the weight value of each internal connection (corresponding to the internal state value explained in Fig. 1). A multiplication processing section 2 that adds all the multiplication results, an accumulation processing section 3 that adds up all the multiplication results, and a threshold processing section 4 that performs nonlinear dark value processing on this accumulated value and outputs one final output. Assuming that the h layer is the first layer and the first layer is the second layer, the accumulation processing unit 3 of the i-th basic unit 1 of the i layer has the following (1
), and the threshold processing unit 4 processes to execute the calculation of equation (2) below.

Xpi=Σy□W,, (1) Formula y pi! 1/(1+exp(-x pi+θ,)
) (2) where, h : Unit number of h layer p : Input signal pattern number 8.Darkness value Wll of il unit of 21st layer, : Weight value of internal connection between h-i layer) 'phiP
The output from the hth unit of the h layer in response to the input signal of the hth pattern.Then, the hierarchical network structure section 20 distributes the truth values calculated by the basic unit 1 having such a configuration and the antecedent truth value calculation section 16. As shown in FIG. 4, a plurality of input units 1
an input layer constituted by °-h, an intermediate layer constituted by one or more basic units 1 and provided in one or more stages (in this example, one stage), and one basic unit 1-j. and an output layer configured by
Between the human crab unit 1'-h and the basic unit 1-i,
It is configured to have a hierarchical network structure formed by mutually internally connecting the basic units 1-i and between the basic units 1-1 and 1-j. Here, the number of input units 1'-h is determined by the number of units of the fuzzy control rule that describes the largest number of antecedent conditions when the antecedent calculation section 17 is implemented in one hierarchical network structure section 20. They are prepared according to the condition number (that is, the number of truth values required).

The hierarchical network structure section 20 configured in this way is
To perform a function of converting an incoming input signal into a corresponding output signal in accordance with a data conversion function defined by a hierarchical network structure and weight values assigned to internal connections of the hierarchical network structure. , and the antecedent part calculation will be executed according to this data conversion function. The present invention is configured such that the antecedent calculation section 17 calculates and outputs the marginal product value of the input signal values in accordance with the data conversion function of the hierarchical non-work structure section 20.

The hierarchical network structure section 20 can be constructed by either a program means or a hardware means, but in the case of constructing it by a hardware means, it can be constructed as described in Japanese Patent Application No. 63-216865 filed by the present applicant. 8, 1988
Filed on March 31st, ゛Network configuration data processing device''
)” can be used.

That is, as shown in FIG. 5, the basic unit 1 is a multiplication type D that multiplies the output from the previous layer input via the input switch section 7 by the weight value held by the weight value holding section 8.
/A converter 2a, an analog adder 3a that adds the output value of the multiplication type D/A converter 2a and the previous accumulated value to calculate a new accumulated value, and holds the addition result of the analog adder 3a. A sample-and-hold circuit 3b that converts the data held in the sample-and-hold circuit 3b into nonlinear form when the accumulation process is completed, and an analog signal of the nonlinear function generation circuit 4a that outputs the data to the subsequent layer. This is realized by including an output holding section 5 that holds a value, an output switching section 6 that outputs the data held by the output holding section 5, and a control circuit 9 that controls each of these processing sections.

The hierarchical network structure section 20 is realized in a configuration in which the basic units 1 having this configuration are electrically connected by one common analog lotus 70, as shown in FIG. Here, in the figure, 71 is a weight output circuit that gives a weight value to the weight value holding section 8 of the basic unit 1, 72 is an initial signal output circuit corresponding to the input unit 1, and 73 is a synchronous control signal that is a control signal for data transfer. Signal weighting output circuit 7
1. A synchronous control signal line 74 that transmits the initial signal output circuit 72 and the control circuit 9 is a main control circuit that sends out the synchronous control signal.

In the hierarchical network structure section 20 having this configuration, the main control circuit 74 sequentially selects the basic units 1 in the previous layer in chronological order, and in synchronization with this selection process, the output holding section of the selected basic unit 1. 5 is transmitted via the analog bus 70 in a time-division manner to the multiplication type D/A converter 2 of the basic unit 1 in the subsequent layer.
Process to output to a. 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 value, multiplies the input value and the weight value, and connects the analog adder 3a and sample hold. The accumulation processing section 3 constituted by the circuit 3b sequentially accumulates the multiplied values. Subsequently, when all the accumulation processing for the basic unit 1 in the previous layer is completed, the main control circuit 74 activates the nonlinear function generating circuit 4a of the basic unit 1 in the subsequent layer to calculate the final output. , the output holding unit 5 processes to hold the final output of the conversion processing result. Then, the main control circuit 74 sets this latter layer as a new former layer and repeats the same process for the next latter layer, thereby changing the output cover turn (output signal) corresponding to the input cover turn (input signal). 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, its data conversion function can be set according to the weight values assigned to the internal connections of the hierarchical network structure. In the present invention, this feature is utilized so that the antecedent part calculation section 17 executes the marginal product calculation.

FIG. 7 illustrates an embodiment of a learning system for learning weight values assigned to internal connections of this hierarchical network structure section 20. Next, according to an embodiment of this learning system, a learning process for determining internal connection weight values of the hierarchical network structure section 20 will be described in detail.

In FIG. 7, 30 is a hierarchical network simulator built on a computer, which simulates the data processing function of the hierarchical network structure section 20, and 40 is a learning signal presentation device, which is a hierarchical network structure section 20. It manages a group of learning signals (the basic unit is a pair of a learning presentation signal and a learning teacher signal) used for learning the weight values of the internal connections of In addition to presenting the learning teacher signal to the network simulator 30, the learning teacher signal is also presented to the learning processing device 50, which will be described later. In order to realize this process, the learning signal presentation device 40 reads the learning signals for learning from the learning signal storage section 41 that manages the learning signal group and the learning signal storage section 41, and selects the learning presentation signals among them. a learning signal presentation unit 42 that presents the learning signal to the hierarchical network simulator 30 and presents the other learning teacher signal forming the pair to a learning processing device 50 (described later) and a learning convergence determination unit 43 (described next);
Upon receiving the output signal outputted from the hierarchical network simulator 30 and the learning teacher signal from the learning signal presentation section 42, it is determined whether the error of the data processing function of the hierarchical network simulator 30 is within an allowable range. , and a learning convergence determination unit 43 that notifies the learning signal presentation unit 42 of the determination result.

Reference numeral 50 denotes a learning processing device implementing a back propagation method known as a learning algorithm for weight values of a hierarchical network configuration data processing device, which outputs a learning presentation signal in response to presentation of a learning presentation signal by a learning signal presentation unit N40. 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 studied is calculated based on the error value. By sequentially updating , the weight value of the internal connection that makes the error value fall within the allowable range is learned.

That is, the learning processing device 50 follows the back propagation method to create three layers, h-layer-1-j layer, shown in FIG.
To explain using the layered network structure unit 20, the output signal ypj from the output layer output when a learning input signal (learning presentation signal) is presented, and the output signal 3'pj take Learning teacher signal d l which is a power signal
When lj is determined, first, the difference value (dpJ-yp=) between the output signal ypJ and the learning teacher signal dpj is calculated, and then αpj""yp, (1-7p;) (dpJ-
y□), and then ΔW, 1(t)=εΣαpj )' pi 0ζΔW,,
(t-1), update amount Δ of weight values between layer 1 and layer j
W.

Calculate (1). Here, t represents the number of learning times, and the reason for adding the weight value update amount determined at the previous update cycle is to speed up the learning.

Next, the learning processing device 50 uses the calculated αpJ to first calculate βp,=yet(1-y,t)Σα,,WJi(t-1
), and then ΔWth(t)=εΣβpiyl'h+ζΔWih(t
-1), update amount ΔWi of weight values between layer h and layer 1
Calculate k(1). Then, according to the calculated update amount, the weight value W,,(t)= for the next update cycle.
WJ, (t-1)+ΔWJi(t)W=h(t)=W
By repeating the method of determining th(t-1)+ΔWth(t), the output signal y2 from the output layer that is output when the learning presentation signal is presented, and the output signal y, Learning teacher signal d pj that is the signal that J should take
The weight values Wjl, W, and the threshold values θ, , θ1 are learned so that they match.

In addition, the present applicant previously applied for the “Japanese Patent Application No. 62-3334”.
No. 84 (filed on December 28, 1988, "Network configuration data processing device"), "1" is always output to the h layer on the input side, and a threshold value θ is set for the output. It has been proposed that by providing a unit for assigning weight values, the threshold value θ is incorporated into the weight value W and the threshold value θ is handled as a weight value. From now on, the threshold value θ will also be learned by the same learning process as the learning of the weight 4fLW.

The hierarchical network simulator 30, as shown in FIG.
a network structure management section 31 that manages internal coupling relationships between the basic units 1 and 1, an arithmetic control section 32 that controls the entire arithmetic processing, and each unit of the input unit 1 and basic unit l of the hierarchical network structure section 20. an input data expansion section 33 that temporarily expands input values input to the , a weight value management section 34 that manages weight values assigned to each internal connection, and an operation unit 34 that is provided to simulate the operation function of the basic unit l. It is configured to include an execution unit 35 and an output data management unit 36 that manages the calculation results of the calculation execution unit 35. The calculation execution unit 35 includes a multiplication value calculation unit 37 that calculates a multiplication value of the weight value and the input value, and a summation value calculation unit 38 that calculates the summation value of the calculation values of the multiplication value calculation unit 37. It is configured to include a threshold value calculation unit 39 that performs threshold value processing on the value calculated by the total sum value calculation unit 38.

With this configuration, the hierarchical network simulator 30 simulates the data processing function executed by the hierarchical network structure unit 20 installed in the fuzzy controller. Note that the hierarchical network structure section 20
When configuring by 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 that executes marginal product calculations, the user first generates the input/output signal relationship for the marginal product calculations and uses the learning signal presentation device 4.
0 is executed. That is, a process is performed in which an input/output signal relationship is generated in which the input signals and the marginal product value of those input signals are used as output signals, and is registered in the learning signal storage section 41.

The marginal product value Y for two values z I + z z (0≦zt≦1) is expressed as y=ov (zl+zz 1). From now on, in FIG. 8, the input unit 1'-h of the input layer of the hierarchical network structure section
The input/output signal relationship of this marginal product operation when the number of units is two is illustrated. In this example of Fig. 8, 8
This figure shows an example in which one input/output signal relationship is generated.

Next, the user activates the learning signal presentation device 40 by instructing that the input signal in the registered input/output signal relationship be used as a learning presentation signal and the output signal associated with this as a learning teacher signal. At the same time, by activating the learning processing device 50 in response to this activation process, the weight values of the weight value management unit 34 of the hierarchical network simulator 30 should realize this input/output signal relationship (before entering learning). to instruct.

In accordance with this activation instruction, the hierarchical network simulator 30 operates as the hierarchical network structure unit 20 that is implemented (or is scheduled to be implemented) in the fuzzy controller, and converts the presented learning presentation signal into data defined by its weight value. While converting and outputting according to the function,
The learning processing device 50 receives the output signal from the hierarchical network simulator 30, updates the weight value of the inner connection (managed by the weight value management unit 34) according to the hack propagation method described above, and presents the learning signal. The device 40 will perform a process of repeatedly presenting the learning signal until the output signal from the hierarchical network simulator 30 approximately matches the learning teacher signal. Through this process, the weight values of the internal connections of the hierarchical network structure section 20 that realize the input/output signal relationship of the registered marginal product calculation are stored in the weight value management section 34 of the hierarchical network simulator 30.

FIG. 9 shows learning data when learning is performed using the input/output signal relationship shown in FIG. 8 as a learning signal. Here, FIG. 9(a) shows the output signal data of the hierarchical network simulator 30 for presentation of each learning presentation signal when the number of learning times is 0 (that is, at the start of learning), and FIG. 9(b) is the output signal data of the hierarchical network simulator 30 for the presentation of each learning presentation signal when the number of learning times is 500, and FIG. 9(C) is the output signal data for the presentation of each learning presentation signal when the number of learning times is 1628. 3 represents output signal data of the hierarchical network simulator 30. Here, this learning was performed using a hierarchical network structure section 20 whose middle layer consisted of 10 units in one stage. Furthermore, at the start of learning, random numbers generated by a random number generator were assigned as initial values for the weight values (including dark values) of each internal connection. In this figure, for example, when the learning presentation signal of rBO9J, which is the learning signal in FIG. represents what has been done. Here, "0.100" associated with this "0.098" in parentheses is the learning teacher signal of "BO2".

FIG. 10 shows a plot of the number of times of learning and the error of the data processing function of the hierarchical network simulator 30. As shown in this figure, when the number of times of learning exceeds 500, the hierarchical network simulator 30 will output a learning teacher signal roughly corresponding to the presentation of the learning signal shown in FIG. In this state, even when something that is not in the learning signal is presented,
It operates to output an output signal that is close to the content of the marginal product calculation.

FIG. 11 shows the weight values and threshold values (hierarchical network simulator 30
The learning values of (managed by the weight value management unit 34) are illustrated. Here, "-1,111794" is the weight value of the internal connection between the first unit of the input layer and the first unit of the hidden layer, and "-1,494546" is the weight value of the internal connection between the unit of the output layer and the first unit of the hidden layer. The weight value of the internal connection with the first unit is “0.
156501” is the darkness value of the first unit in the middle layer, “3
．． 352931″ represents the darkness value of the output layer unit.

As can be seen from Figures 9(c) and 10, this first
If the weight values and darkness values shown in FIG. The calculation unit 17 executes the antecedent calculation of calculating and outputting the marginal product value of the truth values calculated by the antecedent truth value calculation unit 16.

Due to the execution of the marginal product calculation process executed by this antecedent part calculation unit 17, there may be cases where appropriate control cannot be executed by the antecedent part calculation of the logical product operation used in the conventional fuzzy controller. This will also provide a way for more appropriate control to be exercised.

On the other hand, when the weight values and dark values shown in FIG. 13 obtained from the learning signal that defines the input-output relationship of the AND operation shown in FIG. 12 are registered in the internal state value management unit 21, the hierarchical network The structure unit 20 outputs an output signal that provides the AND value of the input signals as shown in FIG. 14 in response to the input of the learning presentation signal shown in FIG. 12. As described above, in the present invention, while realizing the function of calculating the marginal result as an antecedent calculation described in the fuzzy control rule, by simply changing the management data of the internal state value management section 21, It is also possible to return to the antecedent operation of the logical AND operation provided in the fuzzy controller.

Although the illustrated embodiment has been described, the present invention is not limited thereto. For example, in the embodiment, an example of application to the antecedent part calculation unit 17 is explained, which is powered by two people. The arithmetic processing in No. 1 is not limited to dark value conversion processing. In addition, the present applicant previously filed Japanese Patent Application No. 1, No. 1, 1983-227825 (filed on September 12, 1986, "Network configuration data processing"). ``Device Learning Processing Method'') disclosed an invention that improves the back propagation method to realize weight value learning processing in a shorter time. It is also possible to use other weight value learning methods other than the hack propagation method and back propagation method.And, without using a simulator etc.
Hierarchical network structure unit 20 implemented in fuzzy controller
It is also possible to use that information to learn the weight value and the darkness value.

(Effects of the Invention) As explained above, according to the present invention, it is possible to realize the calculation function of the marginal score as the antecedent part calculation of the fuzzy control rule, and it is also easy to change to the conventionally used logical product operation. As a result, it becomes possible to construct a fuzzy controller that is more suitable for the controlled object, and it is also easier to use when it needs to operate as a conventional fuzzy controller. It will be possible to respond to

[Brief explanation of drawings]

Figure 1 is a diagram of the principle configuration of the present invention, Figure 2 is an embodiment of the present invention, Figure 3 is a diagram of the basic configuration of the basic unit, Figure 4 is a diagram of the basic configuration of the hierarchical network structure, and Figure 5 is a diagram of the basic configuration of the hierarchical network structure. FIG. 6 is an example of the basic unit; FIG. 6 is an example of the hierarchical network structure; FIG. 7 is an explanatory diagram of the learning system used in the present invention; FIG. 8 is an explanatory diagram of the generated learning signal; and Fig. 10 is an explanatory diagram of learning data, Fig. 11 is an explanatory diagram of learning data of weight values and dark values, Fig. 12 is an explanatory diagram of learning signals generated when assigning logical product operations, and Fig. 13 is an explanatory diagram of learning data. FIG. 14 is an explanatory diagram of the learning data of the weight value and dark value of the AND operation. FIG. 14 is an explanatory diagram of the output data when operating as the AND operation function. In the figure, 1 is a basic unit, 1 is an input unit, 10 is a fuzzy controller, 16 is an antecedent truth value calculation unit, 17 is an antecedent calculation unit, 18 is a consequent calculation unit, and 20 is a layer 2 -/)
21 is an internal state value management unit; 30 is a hierarchical network simulator; 40 is a learning signal presentation device;
50 is a learning processing device. Principle of the present invention a-temporary diagram Figure 1 Figure 5 g 1 4 or R Hajika F Umebyou network λ town-2 and Arizinrimoto 4 strategies, sword (Figure 4 Explanatory diagram of learning data Figure 10 Logic Divide 1i operation (1 digit becomes 8) Explanation diagram of 5 learning signals Figure 12

Claims (1)

[Claims] An antecedent part truth value calculation unit (16) that calculates a membership function value for the input control state quantity, and a logical operation on the calculated value of the antecedent part truth value calculation unit (16). an antecedent part calculation unit (17) that obtains an output value by applying
), which performs a logical operation on the output value of the consequent part (18) to obtain an output value, and performs a control operation based on the output value of the consequent part calculation part (18) and the membership function value for the control operation amount. In the fuzzy controller, the antecedent part calculation part (17) includes an input unit (1') which distributes the input signal value, and an inference value calculation part (19) that calculates and outputs a quantity. a basic unit (1) that receives one or more inputs and an internal state value by which the inputs are to be multiplied to obtain a product-sum value, and obtains an output by functionally converting the product-sum value; as a structural unit, a plurality of the input units (1'-h) as an input layer, and one or more of the basic units (1-i) as an intermediate layer, comprising one or more intermediate layers, And one of the basic units (1-j) is used as an output layer, and internal connections are formed between the input layer and the foremost intermediate layer, between the intermediate layers, and between the final intermediate layer and the output layer. and a hierarchical network structure unit (20) in which a value is set at which a marginal product value of input signal values is output from the output layer as an internal state value assigned to the internal connection. Fuzzy controller.