JPH01224842A - Inferential control system for knowledge-based system - Google Patents

Abstract

PURPOSE:To shorten an inference executing time by discriminating and processing only necessary minimum sensor data corresponding to the executing condition of inference. CONSTITUTION:A priority table 9 analyzes and prepares the content of a knowledge base 6 in advance. An inference control part 5 notifies which rule is in executing presently, for instance, to a discriminating part 8. The discriminating part 8 discriminates sensor data necessary for the next inference from the inputted sensor data and transmits them to the inference control part 5 based on the dependence between the rules of the priority table 9 and a notified rule name in executing. Thus, in the inference control part 5, only the necessary minimum data are inputted corresponding to the proceeding condition of the inference, processing is attained, and a inference executing time can be shortened.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an inference control method for a knowledge-based system, and more particularly to an inference control method suitable for achieving high-speed inference.

[Conventional technology]

In a knowledge-based system, the knowledge of experts is
For example, in fault diagnosis, inferences are made based on the rule description and the state of the system at the time of the failure.

Determine the cause of the failure.

FIG. 2 shows an example of the conventional configuration of this type of knowledge-based system that monitors and diagnoses the diagnostic target in real time while it is in operation.The diagnostic target device 2 is equipped with a plurality of sensors 3.
The operating status of each component constituting the diagnostic target device 2 is collected at certain time intervals and sent as sensor data to the communication control section 9!4. The knowledge base system 1 receives the operating status of the diagnostic target device E2 through the communication control bag M4,
When a failure occurs, the inference control unit 5 performs inference processing based on the knowledge base 6 and outputs the diagnosis result to the terminal 7.
Knowledge base 6 contains r knowledge of experts regarding fault diagnosis.
It is stored in the rule format of "IF Condition THEN Conclusion". Based on the rules expressing this expert's knowledge, the inference control unit 5 refers to sensor data in a condition part (IF) and generates a diagnosis result in a conclusion part (THEN). In that case, it is difficult to draw a final conclusion immediately from sensor data, so an intermediate conclusion is generated from a rule that has sensor data in its conditional part, and then a final conclusion is generated by another rule that has this intermediate conclusion in its conditional part. Steps are taken to generate a conclusion. Therefore, it is necessary to execute multiple rules in sequence to arrive at the final conclusion. FIG. 3 shows an example of rule description. Rules 1, 2.3 generate intermediate inferences from sensor data. Rule 4 generates a final conclusion from intermediate conclusions. If there are multiple rules whose condition parts are satisfied in comparison with sensor data,
One rule is selected from among the rules and executed.

In a system that monitors and diagnoses in real time, the operating state of the system changes in real time, so it is necessary to always refer to the latest data. In this case, the conventional knowledge base system checks all data sent from the communication control device against the condition part of the rule.

[Problem to be solved by the invention]

In the conventional system, since it was not clear which part information of the target device was necessary for diagnosis, the inference control unit checked the condition part of the rule with respect to all input sensor data. For this reason, data that is not directly related to the progress of diagnosis is also subject to inference, resulting in a problem that the time required for inference becomes longer.

For example, in Figure 3, data C is not necessary when diagnosis is performed based on data a, but in conventional knowledge-based systems, data C is also checked against rules. . This is because the data required during the execution of inference depends on the written content of the rules and the execution situation, so it could not be selected using a general-purpose human-powered mechanism.

An object of the present invention is to provide a knowledge base system that stores the knowledge of experts regarding diagnosis as a knowledge base and performs inference processing based on the contents of the knowledge base, in which only the minimum necessary sensor data is collected according to the execution status of the inference. The purpose is to reduce the inference execution time by performing discrimination and processing.

[Means to solve the problem]

In order to achieve the above object, in the present invention.

In addition to a knowledge base and an inference control unit that performs diagnosis based on the knowledge base, a discrimination unit that discriminates sensor data to be passed to the inference control unit, and a dependency relationship between the rules described in the knowledge base. Provide a priority table.

[For production]

The priority table is created by analyzing the contents of the knowledge base in advance. The inference control unit is for the discrimination unit.

For example, it notifies you which rule is currently being executed. The discrimination unit discriminates only the sensor data necessary for the next inference from the input sensor data based on the dependencies between the rules in the priority table and the notified name of the executed rule, and passes it to the inference control unit. This makes it possible for the inference control unit to input and process only the minimum necessary data according to the progress of inference, thereby reducing inference execution time.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows a configuration diagram of an embodiment of a knowledge base system according to the present invention. Knowledge base system 1 is knowledge base 6
In addition to the inference control unit 5 that performs diagnosis based on the inference control unit 5, the inference control unit 5
It is equipped with a discrimination unit 8 that determines the sensor base to be passed to. The inference control unit 5 notifies the discriminator 8 which rule is currently being executed, and receives only necessary sensor data from the discriminator 8. The discrimination unit 8 distinguishes between the diagnosis target device 2 and the communication control device 4 based on the priority table 9 representing the dependency relationship between rules described in the knowledge base 6 and the name of the rule being executed that is notified from the inference control unit 5. It predicts and discriminates the sensor data necessary for the next inference from the sensor data sent via the inference controller 5, and sends it to the inference control unit 5. The inference control unit 5 outputs the diagnosis result to the terminal 7.

Here, we will explain the contents of the rules and the priority order of sensor data selection.

A dependency relationship exists between the rules in the knowledge base 6 based on the description contents of the condition part and the conclusion part. Another rule (rule b) that refers to data that is generated or changed in the conclusion part of a certain rule (referred to as rule a).
In order to explain that "Rule b depends on Rule a", we create a directed graph (referred to as a "rule dependency graph") in which the rules are nodes and the dependencies are edges. Expressed as Figure 4 shows an example of the rule dependence graph. In the 4th vtU, P, q*
r represents sensor data, and A, B, C, and D represent other data.The rule at the top in the rule dependency graph is called the "ancestral rule," and the rule directly below it is called the r child rule. . The rule dependency graph is created by analyzing the written contents of the rules before starting the execution of inference.

Inference is controlled in a depth-first manner on the rule dependency graph, ie, rules that are children of the most recently executed rule are tried next. If the condition part of the child rule is not satisfied, the process goes back one level and tries another branch.

From the rule dependency graph, the priority of sensor data to be input and processed is determined for each rule immediately after the rule is executed. The rules are shown below.

Rule 1: The sensor data referenced in the condition part of the child rule has the highest priority.

Rule 2: Regarding ancestor rules, on the rule dependency graph,
Priority is determined starting from the closest rule. Data referenced by a closer rule has a higher priority.

Rule 3: Priority is not given to data referenced by rules other than child and ancestor rules.

Rules 1, 2.3 are performed at the same time as the rule dependency graph is created, and a sensor data priority table is created for each rule.

FIG. 5 shows a detailed configuration of the discrimination section 8, which is composed of a filter section 10 and a sorter section 11. The inference control unit 5 is the discriminator 8
Notifies the name of the rule that was recently executed (or immediately before execution). The filter section 10 of the discriminator section 8 searches the priority order table 9, and among the sensor data received from the communication control device 4, only N pieces of sensor data having a priority order are passed to the sorter section 11, and the other data are simply held. The sorter unit 11 orders the data received from the filter unit 10 according to the priority order, and passes the data to the inference control unit 5 in order from data having the same priority order.

The inference control unit 5 determines the next rule to be executed based on the data received from the discrimination unit 8.

When the next rule to be executed is changed by the input from the discrimination unit 8, the inference control unit 5 executes the rule. If no change is to be made, a request is made to the discriminator 8 for data of the next priority order.

In order to shorten the execution time, the inference control unit 5 and the discrimination unit 8 are realized on independent processors or as multi-processes. When the inference control unit 5 determines the rule to be executed, it passes the rule name to the discrimination unit 8 and starts executing the rule. When the data in the reworking memory is changed by executing a rule, the next rule to be executed is determined based on the changed content.

While the inference control unit 5 is deciding on the rule execution, the discrimination unit 8 predicts the next rule to be executed and selects sensor data to be passed to the inference control unit 5.

FIG. 6 shows a specific process flow. FIG. 6 (1) shows a state where no sensor data has been input and Rule 1 is being executed. In Figure 6 (2), sensor data Pe
q has occurred, but only the data p referenced by rule 2, which is a child of rule 1, is selected by the discrimination unit 8 and sent to the inference control unit 5 for processing. 6th Ii! Although the sensor data p is generated in U(3), the sensor data p is not referenced in the rule 4 currently being executed, so the data p is held in the discriminator 8. In Figure 6 (4), the data referenced by rule 8, which is a child of rule 4, is
Newly generated data r and previously generated data q
is passed to the inference control unit 5, and as a result, rule 8 is executed.

〔Effect of the invention〕

As is clear from the above description, the inference control method of the present invention provides the following effects.

(1) Since only data directly related to the flow of inference is passed to the inference system, execution time is shorter than in the conventional case where all data is processed by the inference control unit.

(2) Data selection is performed based on the description of the rule and the current progress of inference. Therefore, more accurate selection is possible than in the conventional method in which selection is made using a general-purpose input mechanism based only on the interrupt level without recognizing the description content or progress status.

(3) By implementing the discrimination unit as a separate processor from the inference control unit or as a multi-process, data can be narrowed down without degrading the performance of the original inference processing.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the knowledge base system according to the present invention, FIG. 2 is a block diagram of a configuration example of a conventional knowledge base system, and FIG. 3 is a diagram showing an example of writing rules representing knowledge. , FIG. 4 is a diagram showing an example of a rule dependency graph expressing the dependence relationships between rules, FIG. 5 is a diagram showing the detailed configuration of the discriminator in FIG. 1, and FIG. It is a figure which shows the example of a flow. 1...Knowledge base system, 2...Diagnosis target device. 3...Sensor, 4...Communication control device. 5... Inference control unit, 6... Knowledge base, 7...
Terminal, 8... Discrimination unit, 9... Priority order table. Figure 1 Figure 2 Figure 5 Figure 6 Figure 6

Claims (1)

[Claims] (1) It is equipped with a knowledge base that has accumulated the knowledge of experts, and an inference control unit that infers the state of the diagnosis target based on the contents of the knowledge base and the diagnosis target data that is input asynchronously from the outside. In the knowledge base system, a table representing dependencies between rules described in the knowledge base and a discrimination unit for discriminating the diagnosis target data inputted asynchronously are provided, and the inference control unit is configured to cause the discrimination unit to The discrimination unit is characterized in that the discrimination unit discriminates the diagnosis target data based on the contents of the table and the notified rule and passes it to the inference control unit. An inference control method for knowledge-based systems.