JPH0619509A - Fuzzy controller - Google Patents

Abstract

(57) [Summary] [Purpose] In a fuzzy controller that executes fuzzy inference operations using membership functions and multiple inference rules and outputs inference results, inference rules are organized and integrated to improve inference operation processing efficiency. Let [Structure] A membership function and all inference rules or some inference rules are used for each input information to execute a simulation operation for the antecedent operation, and the inference rule validity determining means determines each input information. And the respective inference values are compared and compared to determine the validity of each inference rule stored in the user memory. Further, the inference rule redundancy judgment means judges the redundancy of the combination inference rule from the combination of the inference rules used for the antecedent part calculation and the combination of the inference values stored in the simulation calculation result storage part corresponding to this combination. To do.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a fuzzy controller for executing a fuzzy inference operation using a membership function and a plurality of inference rules for input information and outputting an inference result.

[0002]

2. Description of the Related Art In recent years, a control device using fuzzy reasoning has been adopted also in process control for controlling each production facility in a factory. In such a fuzzy controller,
Member zip functions and a plurality of inference rules are stored, and membership functions and inference rules are stored for each input information consisting of physical quantities such as temperature, pressure, flow rate, etc. input from each facility as a control target. A fuzzy inference operation is executed by using this to infer and output a manipulated variable for a controlled object such as valve opening and valve closing.

The information required when defining fuzzy control is a membership function and inference rules. Of these, the membership function is a probability function showing the characteristics of each input / output, and is usually defined in the form of a polygonal line function as shown in FIG. On the other hand, the inference rule indicates the relationship between one or more inputs and outputs, and is generally defined in the IF-THEN format, as shown in FIGS.

Of the membership functions shown in FIG. 5 and the inference rules shown in FIGS. 6 and 7, it is the inference rules that are difficult to define in actual use. That is, in the inference rule, since the information that is the basis for the judgment of the person who actually makes the judgment is expressed in the form of input (IF) and output (THEN), it is expressed about the part that the human is conscious of. Will be things.

It is very complicated to extract the know-how possessed by a skilled engineer who cannot be expressed by such a mathematical expression from the human knowledge level, and convert and express it into a form that a machine (computer) can understand. It's also difficult.
Then, in reality, an expert called a knowledge engineer (KE) collects information from an engineer in an interview format, arranges it, and writes it in a storage device as inference rules.

[0006]

However, as described above, even if a specialized knowledge engineer draws out each inference rule that is the basis of control / judgment from the knowledge of a person who specializes in the target technical field, It is very complicated to organize and convert it into an expression format understood by the machine. As a result, the following problems occur.

If all the knowledge heard from the engineer in charge is converted into the form of inference rules, the number of inference rules naturally increases. As the number of inference rules increases, the inference calculation processing time increases. As a result, there is a concern that the control response performance of the process control device may deteriorate.

Further, when the pieces of knowledge received from the engineer in charge are made into rules without rearranging them, as described above, not only the number of inference rules increases but also the expressions are different. There is a concern that a large number of inference rules that produce different results will be created, and that inconsistent inference rules that result in mutually opposite conclusions will occur between multiple inference rules that have the same meaning.

The present invention has been made in view of the above circumstances, and is unnecessary by adding a function of determining whether each inference rule once set in the user memory is a valid rule. An object of the present invention is to provide a fuzzy control device capable of removing the inference rule, increasing the inference operation speed while maintaining high control accuracy, and improving the control response performance.

[0010]

In order to solve the above problems, the present invention provides a user memory for storing a membership function and an inference rule, and a certainty factor of the input information from a controlled object. Calculated using a membership function, and control from the antecedent calculation unit that calculates the inference value using the inference rule for this calculated certainty factor, and the inference value calculated by this antecedent calculation unit In a fuzzy control device including a consequent operation unit that calculates an operation amount for an object,

An input information storage unit that stores each input information that is sequentially input from the controlled object, and a membership function by an antecedent operation unit for one or more sets of input information stored in the input information storage unit. And a simulation operation means for performing an inference operation using part or all of the inference rules, a simulation result storage section for storing each inference value obtained by the simulation operation means, and each input stored in the input information storage section. The inference rule validity determining means for determining the validity of each inference rule stored in the user memory using the information and each inference value stored in the simulation result storage section, and the inference rule used in the simulation computing means. Redundancy of combination inference rules from combinations of combinations and inference values stored in the simulation result storage corresponding to this combination It is obtained by a reasoning rule redundancy determination means for determining.

[0012]

With the fuzzy control device thus constructed, in a normal operating state, a manipulated variable is obtained by fuzzy arithmetic processing with respect to input information input from the controlled object, and the manipulated variable is sent to the controlled object. This fuzzy arithmetic process is
As is well known, it is composed of antecedent part arithmetic processing and consequent part arithmetic processing.

In the present invention, the input information sequentially input from the controlled object in the above-mentioned normal operating state is separately accumulated and stored in the input information storage section. A membership function set in the user memory by activating the antecedent operation unit for one or more sets of input information stored in the input information storage unit when the operating time is idle or offline, for example. And inference rules are used to simulate inference operations. Each inference value obtained by this simulation is stored in the simulation result storage unit. The validity of the inference rule used for this inference is judged by comparing and comparing each stored inference value with the previously input information.

Therefore, among a large number of inference rules set by the user in the user memory, an invalid inference rule that does not actually satisfy the condition from a physical point of view or the inference result is affected. If there is an inference rule that does not exist, the inference rule is specified by the inference rule validity determining means.

If there is a combination of inference rules that have an equivalent effect on the inference result, or if there is another inference rule that has the same effect on the inference result, the inference rule redundancy judgment means selects those rules. Inference rules are specified. Therefore, the operator may be specified and the inference rules may be organized and integrated.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the schematic arrangement of a fuzzy control apparatus according to the embodiment. This fuzzy control device is composed of a kind of computer, and has a CPU 2 and an OS for executing various control operations on the system bus line 1.
Process data as input information consisting of each physical quantity such as temperature, pressure, flow rate, etc. is input from each control target such as ROM 3 for storing control programs such as programs and various fixed data, each manufacturing facility of the factory, etc. Interface 4 for converting the data into a digital value, an output interface 5 for sending out the inference result as a manipulated variable to each controlled object, a transmission interface for exchanging various information with an external host computer via a transmission path 6 etc. are connected.

Further, the system bus line 1 has a system memory 7 for storing a system program and a fuzzy arithmetic program, a user memory 8 for a user who actually uses the fuzzy controller to set various data, and a user for the user. A man-machine interface 10 for controlling a man-machine device 9 for inputting various data and operation commands to the memory 8 is connected.

As shown in FIG. 2, a system program area 7a for storing a normal program and a fuzzy inference program area 7b for storing various programs related to fuzzy inference are formed in the system memory 7. In the system program area 7a, processing programs for operating the interfaces 4, 5, 6, 10 are stored. Further, in the system program area 7a, a user execution processing program set by the user is stored.

In the fuzzy reasoning program area 7b,
A fuzzy inference execution program 11a for executing a normal fuzzy inference operation, an inference rule validity judgment program 11b for judging the validity of each inference rule once set in the user memory 8, and the inference rule An inference rule redundancy judgment program 11c for judging redundancy is stored.

In the user memory 8, as shown in FIG. 3, a membership function area 8a and an inference rule area 8 are provided.
b, an input information storage unit 8c and a simulation result storage unit 8d are formed.

In the membership function area 8a, for example, FIG.
The membership functions shown in are stored. The inference rule area 8b stores a large number of inference rules shown in FIGS. The membership functions and the inference rules differ depending on the type of each control target to which the fuzzy control device is connected, and are generally set by the user of the fuzzy control device using the man-machine device 9.

The input information storage unit 8c stores input information such as temperature, pressure, flow rate, etc., which is input via the input interface 4 in a constant cycle in a time-series manner. Further, in the simulation result storage device 8d,
The inferred value obtained by the antecedent operation in the simulation processing is accumulated and stored.

Although not shown, this user memory 8 has
A control cycle T for performing control on each control target at a certain fixed cycle is stored. Further, a user program set by the user is stored.

The operation of the fuzzy control device thus configured will be described. First, using the man-machine device 9, each membership function and each inference rule shown in FIGS. 5 and 6 and 7 are set in the membership function area 8a and the inference rule area 8b of the user memory 8. When the operator completes the membership function and inference rule setting process for the user memory 8, the operator activates the fuzzy controller.

The CPU 2 fetches each input information such as a physical quantity from each controlled object via the input interface 4 with the control cycle T stored in the user memory 8 and executes a fuzzy control process for the corresponding input information. To do.

That is, the input information is temporarily stored in the user memory 8
It is stored in the input information storage unit 8c. The fetch cycle, storage address, etc. are specified by the user program. Moreover, each input information other than this input information is created as needed. For example, the difference between physical input information is calculated as new analog deviation information, and the physical input information is differentiated as new change rate information, the input information storage unit 8c.
Store to.

Next, the CPU 2 retrieves the current input information stored in the input information storage section 8a in accordance with the fuzzy inference execution program 11a in the fuzzy inference program area 7b and stores the membership function stored in the user memory 8. And fuzzy inference using inference rules.

Specifically, as shown in FIG. 8, when one input value t is input, the certainty factor (grade) of this input value t for each membership function 10a, 10b is calculated. Next, the output-side member jib functions 11a and 11b in each of the calculated certainty factors using the inference rule.
, The inference value 12a. Calculate 12b. The above is the antecedent part arithmetic processing.

When this antecedent part arithmetic processing is completed, as shown on the right side of FIG. 8, each inference value 12a. For example, the area of the shaded area is calculated from 12b, the center of gravity thereof is calculated, and the operation amount Y for the controlled object is calculated. The above is the consequent part arithmetic processing. Then, the manipulated variable Y calculated in the consequent part arithmetic processing is converted into a form that can be understood by the controlled object, and is sent to the controlled object via the output interface 5.

As described above, the fuzzy inference operation is executed on the input information input from the controlled object using the membership function and the inference rule, and the inference result is transmitted to the controlled object. Therefore, the controlled object is well controlled.

Further, in this embodiment, when a verification command for an inference rule is input via the man-machine device 9, for example, the simulation process shown in FIG. 4 is executed. Each determination process for this inference rule is executed during the period in which the control operation for the controlled object is stopped, or during the idle time within the control cycle T described above.

In FIG. 4, a set of input information stored in the input information storage section 8c is read (P1), and the inference operation is performed using the membership function and one inference rule stored in the user memory 8. Is executed (P2.P
3). When the inference operation using all the inference rules for this one set of input information ends (P4), the simulation operation for the antecedent operation for the corresponding input information ends, so the obtained inference values are stored in the simulation result storage unit. Store in 8d.

Then, the consequent part arithmetic part is activated to infer the consequent part calculation for each inferred value (P7).
Then, the output process for the obtained manipulated variable is performed (P7). If a large amount of input information is stored in the input information storage unit 8c, FIG.
The simulation calculation shown in is executed.

Therefore, when the simulation calculation is performed a plurality of times, the simulation result calculation unit 8d stores each inferred value for each combination of the input information stored in the input information storage unit 8c. Next, CP
U2 executes the inference rule validity judgment program 11b and the inference rule redundancy judgment program 11c.

At a specific time, each combination of each input information, inference value, and inference rule used for this inference is compared and compared. Then, depending on the combination, the condition of the inference rule is not satisfied at all, so that the case where the inference value remains 0 or the case where no inference is possible is detected. In this way, an inference rule in which the conditions of the inference rule are not satisfied at all is judged to be an ineffective inference rule.
Then, the inference rule determined to be ineffective is displayed on the display screen of the man-machine device 9, for example.

An example of this ineffective inference rule is shown in FIG.
Will be explained. FIG. 6 is a diagram showing an inference rule for inferring the operation amount Y indicated by the certainty factor (grade) of the membership functions of the input A and the input B. In this case, input A
Is the temperature input and pressure C is the pressure input. Since there is a physically constant relationship between the temperature input and the pressure input, the two inputs A. There is a certain limit on the range of values that the input C can take. For example, if the temperature is very high (A = P
The combination of B) and very low pressure (C = NB) does not exist as an actual physical condition, so the inference rule (1) shown in FIG. 6 is an ineffective rule. (1) IF A = PB AND C = NB THEN Y = ZR Depending on the type of the controlled object, the inference rule of (1) may be valid, but in such a special case, an exception is made. And ignore this time.

When the validity of the inference rule is determined in this way, when focusing on a specific inference rule, no inference result can be obtained in that inference rule, or a comparison is made with a combination of other inference rules. Then, since the output value is small, it can be determined that this inference rule is not valid.

When the validity judgment processing for each of the above inference rules is completed, the CPU 2 activates the inference rule redundancy judgment program 11c. That is, as described above, after the inference operations for all combinations of the inference rules are completed, when each inference rule and the inference result are compared and compared,
There are multiple inference rules that give the same inference result. In this case, even if the different parts of the inference rules are different in terms of the expression of the inference rules, the actual physical phenomenon may have the same content. Further, the different part of each inference rule may be a condition in which one includes the other.

In such a case, the inference calculation time naturally becomes long. Therefore, as described above, the inference rule that produces the same result under substantially the same condition is determined as the inference rule having redundancy and is displayed on the display screen of the man-machine device 9.

An example of the inference rule having this redundancy will be described with reference to FIG. FIG. 7 shows input A, input C, and input D.
It is a figure which showed the inference rule which infers the operation amount Y shown by the certainty factor (grade) of the membership function with.

Since there is a physically constant relationship between the temperature input A and the pressure input C, these two inputs A. There is a certain limit on the range of values that the input C can take. For example, under constant volume conditions, the pressure rises as the temperature rises. Therefore, high temperature conditions (A = PB)
And that the pressure is high (C = PB) are described as independent inference rules different in word expression form, or can be regarded as substantially the same inference rule. The input D is a physical quantity other than temperature and pressure.

Therefore, in the four inference rules (1), (2), (3) and (4) shown in FIG. 7, the inference rules of (1) and (2) are the same inference rule, and further, (3 The inference rules of () and (4) are the same inference rule. Therefore, 4 inference rules are 2
It is possible to replace with one inference rule.

In this way, when the redundancy of the inference rules is judged, the inference rules are different from each other by comparing and contrasting each inference rule with the inference result, and the same inference result is obtained. By detecting the combination, it can be determined that there is redundancy for this inference rule. In this way, the inference rule validity determination process and the inference rule redundancy determination process specify, for example, the inference rules corresponding to the following items. (a) the existence of an invalid rule that does not actually satisfy the condition, (b) the existence of an inference rule that does not affect the result of inference, (c) the one included in other large inference rules, (d) there are multiple inference rules that affect the inference results equivalently, and (d) other inference rules that affect the inference results the same.

Each inference rule extracted by the inference rule validity determination process and the inference rule redundancy determination process as described above is displayed on the display screen of the man-machine device 9.
The operator deletes the inference rule pointed out to be ineffective from the inference rule area 8b of the user memory 8. Also,
A necessary inference rule among the plurality of inference rules that are pointed out to be redundant is left, and an erasing operation is performed on other inference rules.

[0046]

As described above, according to the fuzzy control device of the present invention, the simulation of the antecedent operation is performed using the input information stored in the input information storage unit, and the inference value of the simulation operation result is executed. And the input information are compared with each other to determine whether the inference rule is valid or redundant with respect to each inference rule set in the user memory. Therefore, by consolidating invalid inference rules and redundant inference rules, the number of inference rules stored in the user memory can be limited to the necessary minimum, and the control response can be achieved without degrading the control accuracy. Performance can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of a fuzzy control device according to an embodiment of the present invention,

FIG. 2 is a diagram showing stored contents of a system memory of the apparatus of the embodiment,

FIG. 3 is a diagram showing stored contents of a user memory of the apparatus of the embodiment,

FIG. 4 is a flow chart showing the operation of the apparatus of the embodiment,

FIG. 5 is a diagram showing a general membership function,

FIG. 6 is a diagram showing ineffectiveness inference rules,

FIG. 7 is a diagram showing inference rules with redundancy,

FIG. 8 is an explanatory diagram showing the relationship between general antecedent part arithmetic and consequent part arithmetic.

[Explanation of symbols]

2 ... CPU, 7 ... System memory, 7a ... System program area, 7b ... Fuzzy inference program area, 8 ...
User memory, 8a ... Membership function area, 8b ... Inference rule area, 8c ... Input information storage section, 8d ... Simulation result storage section, 9 ... Man-machine device.

Claims (1)

[Claims] 1. A user memory that stores a membership function and an inference rule, and a certainty factor of the input information with respect to input information from a control target is calculated by using the membership function, and the calculated certainty is calculated. Degree, the antecedent part computing part for computing the inferred value using the inference rule, and the consequent part computing part for computing the operation amount for the controlled object from the inferred value computed by the antecedent part computing part In a fuzzy control device including: an input information storage unit that stores each input information that is sequentially input from the control target; and a 1 stored in the input information storage unit.
Simulation operation means for performing inference operations on the input information of more than one set by the antecedent operation section using a part or all of the membership function and inference rules, and each inference obtained by the simulation operation means. Of the inference rules stored in the user memory using a simulation result storage unit that stores a value, each input information stored in the input information storage unit, and each inference value stored in the simulation result storage unit A combination of the inference rule validity judgment means for judging the effectiveness, the combination of the inference rules used in the simulation calculation means, and the combination of the inference values stored in the simulation result storage unit corresponding to this combination, the combination inference rule Fuzzy control device provided with an inference rule redundancy judgment means for judging the redundancy of the.