JPH05134873A - Dynamic rule type knowledge changing system - Google Patents

Abstract

PURPOSE:To improve the validity of the inferred result and to reduce the load of the tuning and maintenance of knowledge by dynamically changing rule type knowledge in the middle of inference at an expert system corresponding to the conditional change of fact type knowledge. CONSTITUTION:In the preceding step of using rule type knowledge 11 handling fact type knowledge 12 for which the condition is changed in the middle of inference, a fact type knowledge analyzing mechanism 13 analyzes the condition surrounding the objective fact type knowledge 12. Based on this analyzed result, a rule type knowledge changing mechanism 15 changes the rule type knowledge 11 corresponding to the conditional change of the fact type knowledge 12. The analyzing method of the fact type knowledge is previously defined as fact type knowledge analyzing knowledge 14, and the changing method of the rule type knowledge is defined as rule type change knowledge 16 in advance.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a dynamic rule type knowledge changing method in an expert system for reasoning using fact type knowledge and rule type knowledge.

[0002]

2. Description of the Related Art An expert system that separates expert knowledge into fact type knowledge and rule type knowledge, and uses the fact type knowledge and rule type knowledge to make inferences. As a method for handling changing factual knowledge, for example, there is a description as 6034 "Job Shop Scheduling Priority Determination Method by Fuzzy Logic" in Proceedings of 34th Annual Conference of the Society of System Control Information Studies, 487-488. Here, the job shop (job-sho
In p) type scheduling, we have proposed a method that gives more fine priority by using fuzzy inference for the processing that gives the priority of allocation based on the evaluation scale of multiple candidates.

[0003]

The fuzzy rules used in the above conventional method can be created relatively easily, but in order to obtain good results, it is necessary to finely change the membership function expressing the fuzzy proposition. ..

In the conventional method, it is premised on the fuzzy inference that the state of fact-type knowledge does not deviate from the domain of the membership function that is set in advance. However, the state of fact-type knowledge is predetermined during the inference. It is possible that a given membership function is out of the domain, and the issue is how to derive an appropriate inference result from the state of factual knowledge that changes during this inference.

The object of the present invention is to analyze the state in which rule-based knowledge is used in an expert system, and modify the rule-based knowledge so as to match the situation based on the result. It is intended to improve rule-based knowledge and save labor for maintenance of rule type knowledge.

[0006]

In order to achieve the above object, the present invention provides a factual knowledge analysis mechanism for analyzing a situation surrounding factual knowledge, and a rule-based knowledge change for changing rule-based knowledge from the analysis result. Provide a mechanism. The situation surrounding the target factual knowledge before the rule-based knowledge that handles the factual knowledge whose state changes during inference is used.
The fact type knowledge analysis mechanism analyzes and based on the analysis result, the rule type knowledge changing mechanism changes the rule type knowledge which is currently used to suit the situation.

[0007]

In the present invention, it is possible to obtain a proper inference result by changing the rule type knowledge in advance from the analysis result of the situation surrounding the fact type knowledge so as to match the state. Also, since it is possible to dynamically change rule-based knowledge, it is possible to change the state during inference.
Rule-based knowledge about fact-based knowledge whose states are difficult to define can be easily expressed.

[0008]

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

FIG. 1 is a functional block diagram of an embodiment of the present invention. Expert knowledge is separated into factual knowledge 12 and rule-based knowledge 11 and stored in an external storage device or the like. The rule type knowledge 11 is used for a set of one or more fact type knowledge 12 that satisfy the condition. Factual knowledge analysis mechanism 1
Reference numeral 3 is a portion for analyzing the situation surrounding the factual knowledge 12, and the analysis method is stored in advance in the factual knowledge analysis knowledge storage unit 14 such as an external storage device. The rule type knowledge changing mechanism 15 is a part for changing the rule type knowledge 11, and the changing method is stored in advance in the rule type knowledge changing knowledge storage unit 16 such as an external storage device.

During the inference, the state of the factual knowledge 12 often changes. The fact type knowledge analysis mechanism 13 preliminarily describes the situation surrounding the target fact type knowledge 12 before the rule type knowledge 11 that handles the fact type knowledge 12 whose state changes during the inference is used. Knowledge analysis The fact type analysis knowledge stored in the knowledge storage unit 14 is read and analyzed. The rule type knowledge changing mechanism 15 reads out the rule type knowledge changing knowledge stored in the rule type knowledge changing knowledge storage unit 16, and based on the analysis result of the fact type knowledge analyzing mechanism 13, the rule type knowledge changing mechanism 11 To
Change to match the situation of factual knowledge 12.

FIG. 2 is a diagram showing the hardware configuration of a computer system to which this embodiment is applied. The central processing unit (CPU) 21 is a part that executes various processes, and the fact type knowledge analysis mechanism 13 and the rule type knowledge changing mechanism 15 of FIG. 1 are included therein. The memory 22 stores programs, data being processed, and the like. The external storage device 23 is used for storing the rule type knowledge 11 and the fact type knowledge 12 of FIG. 1, and as the fact type knowledge analysis knowledge storing section 14, the rule type knowledge change knowledge storing section 16, etc. Accordingly, it is loaded into the memory 22 and provided to the processing of the CPU 21. The keyboard 24 is an input device for commands and data, and the display 25 is an output device for displaying processing results and the like.

FIG. 3 is a conceptual diagram of the problem used in this embodiment, and here shows the problem of making a production plan for producing a plurality of orders by one manufacturing apparatus.

FIG. 4 is an example of a data structure of factual knowledge used in this embodiment. The initial data of the factual knowledge includes order data 41 and manufacturing device data 42. The order data 41 is composed of orders A, B, C and D, and each order is composed of delivery date, product name, production time, and allocation flag. The manufacturing device data 42 includes the end time of the previous production and the product name of the previous production. The analysis result table 43 is an area serving as an interface for passing the analysis result from the fact type knowledge analysis mechanism 13 to the rule type knowledge changing mechanism 15. The evaluation result 44 is the rule-based knowledge changing mechanism 1
An area 5 is an interface for changing a part of the rule type knowledge. The evaluation result 44 is described as part of the rule-based knowledge. The work table 45 is an area for storing names of allocation candidates and evaluation values.

FIG. 5 shows an example of rule type knowledge used in this embodiment. Reference numeral 51 is an expression in the rule form of the allocation know-how "Prioritize candidates with earlier delivery date."
2 and the execution unit 53. This rule is repeatedly used for the number of combinations of fact type knowledge satisfying the condition part 52. The condition section 52 is composed of condition clauses 521, 522, and 523. In the condition clause 521, it is verified whether or not there is an arbitrary allocation candidate name / evaluation value pair in the work table 45, and if it exists, the allocation candidate name corresponding to the variable “? Allocation candidate” and the variable “? Evaluation value” are satisfied. The evaluation value to be set is set. In the conditional clause 522, the delivery date of the factual knowledge indicated by the variable “? Allocation candidate” is set in the variable “? Delivery date”. In the conditional section 523, the variable “? Delivery date evaluation value” used in the execution section 532 is defined. Execution unit 53
Is composed of an execution section 531 and an execution section 532.
In the execution section 531, the rule type knowledge changing mechanism 15 is called in order to set the variable "? Delivery date evaluation value" described as a part of the rule type knowledge. In the execution section 532, the variable “? Delivery date evaluation value” set in the execution section 531 and the conditional section 52 are set.
The variable “? Evaluation value” acquired in 1 is added and stored in the work table 45.

Reference numeral 54 represents an allocation know-how "allocation of the same product in succession" in a rule format, which is composed of a condition section 55 and an execution section 56. This rule is repeatedly used for the number of combinations of factual knowledge that satisfy the condition part 55. The condition section 55 is composed of condition clauses 551, 552 and 553. In the conditional clause 551, it is verified whether or not there is an arbitrary allocation candidate name / evaluation value pair in the work table 45, and if there is, a variable “?
The allocation candidate name corresponding to the above and the evaluation value corresponding to the variable "? Evaluation value" are set. In the conditional section 552, the product name of the factual knowledge indicated by the variable “? Allocation candidate” is the variable “?
Product name ”is set. In the condition clause 553, it is verified whether the product name of the manufacturing apparatus, which is the factual knowledge, and the variable “? Product name” match. In the execution section 56, the variable “? Evaluation value” acquired in the conditional section 521 and 0.4 are added and stored in the work table 45.

The allocation know-how shown by the rule type knowledge of 51 and 54 has a trade-off relationship, the rule type knowledge of 51 is a continuous value from 0.0 to 0.6, and the rule type knowledge of 54 is 0. Discrete evaluation values of 0 or 0.4 can be added, and the balance of superiority when the allocation know-how shown by the rule type knowledge of 51 and 54 competes is a ratio of 6: 4.

FIG. 6 and FIG. 7 are process flows showing the process flow of the embodiment of the present invention. Hereinafter, the case of FIGS. 4 and 5 will be described as an example based on these drawings.

First, the order file 61 (order data 41) is read from the external storage device 23 and loaded in the memory 22 as the factual knowledge 12 (F01). The manufacturing apparatus data 42 is assumed to already exist. Next, the factual knowledge analysis knowledge 62 is read from the factual knowledge analysis storage unit 14 of the external storage device 23 (F02), and similarly the rule-knowledge knowledge change knowledge 63 is read from the rule-type knowledge change knowledge storage unit 16 (F03). .. Then, it is determined whether or not there is an order not yet allocated (F04), and if satisfied, the following processes of F05 to F18 are performed.

An area for storing a candidate name and an evaluation value is secured in the work table 45 by taking an order for which an allocation flag is "not yet" as an allocation candidate, and the evaluation value is cleared to 0.0 to evaluate as a candidate name. The value is stored (F05). The fact type knowledge analysis mechanism 13 analyzes the distribution state of the delivery dates of the allocation candidates, which changes during the inference, and the result is the analysis result table 4.
It is stored in F3. It is determined whether there is an allocation candidate for which the delivery date has not been evaluated (F07), and the unevaluated allocation candidates are subjected to the processing of F08 to F12. The delivery date of the allocation candidate is acquired in the condition part of the delivery date evaluation rule 51 (F08), and the evaluation value is defined as a variable (F09). The execution part of the delivery date evaluation rule 51 acquires the earliest delivery date and the latest delivery date of the allocation candidate from the analysis result table 43, and changes the delivery date evaluation value calculation formula (F10). An evaluation value is calculated for the delivery date of the allocation candidate by the rule-type knowledge changing mechanism 15 (F11), and the evaluation value variable is added to the evaluation value of the work table 45 and stored (F12).

After the evaluation of the delivery dates of all the allocation candidates is completed, it is judged whether all the allocation candidates are evaluated for the product name (F13).
~ F15 processing is performed. In the condition part of the product name evaluation rule 54, it is determined whether the product name of the allocation candidate matches the manufactured product of the manufacturing apparatus data 42 (F14). If they match, the execution part of the product name evaluation rule 54 sets 0.4. The evaluation value of the work table is added and stored (F15).

After the evaluation of the product names of all the allocation candidates is completed, the work table 45 is sorted in the ascending order of the evaluation values and the allocation candidates are determined (F16). In the order data 41, "already" is set in the allocation flag of the determined allocation candidate, and the start time and end time are stored (F17). The production time of the candidate assigned to the production end time of the manufacturing apparatus data 42 is added, and the product name of the assigned candidate is set to the product name (F18).

When all orders are assigned, the production plan data 6 is provided with information including the determined production time zone of each order.
It is stored as 4 in the external storage device 23 (F19).

In the above, the embodiment has been described by taking as an example the problem of formulating a production plan for producing a plurality of orders with a single manufacturing apparatus, but it goes without saying that the present invention is not limited to this. ..

[0024]

As described above, according to the present invention,
Rule-based knowledge, which deals with factual knowledge that changes dynamically during inference, is analyzed in advance for the situation surrounding the target factual knowledge, and the results of the analysis are dynamically changed to confirm the validity of inference. Improvements, tuning of rule-based knowledge and labor saving of maintenance can be expected. Further, since it is not necessary to define independent rule type knowledge for each situation of fact type knowledge, it is possible to make a reasonable inference with a relatively small amount of rule type knowledge.

[Brief description of drawings]

FIG. 1 is a functional block diagram showing the overall configuration of an embodiment of the present invention.

FIG. 2 is a diagram showing a general hardware configuration of a computer system to which the present invention is applied.

FIG. 3 is a conceptual diagram of a problem used in the present invention.

FIG. 4 is a diagram showing an example of fact type knowledge used in the present invention.

FIG. 5 is a diagram showing an example of rule-based knowledge used in the present invention.

FIG. 6 is a flowchart showing a flow of processing according to an embodiment of the present invention.

7 is a continuation of the flowchart of FIG.

[Explanation of symbols]

 11 Rule-type Knowledge 12 Factual-type Knowledge 13 Factual-type Knowledge Analysis Mechanism 15 Rule-type Knowledge Change Mechanism

Claims (1)

[Claims] 1. An expert system that separates expert knowledge into fact type knowledge and rule type knowledge, and infers using the fact type knowledge and the rule type knowledge, relates to the fact type knowledge targeted by the rule type knowledge. It is characterized by providing a means for analyzing a situation and a means for dynamically changing rule type knowledge based on the analysis result, and changing rule type knowledge that handles fact type knowledge whose state changes during inference in advance. Dynamic rule type knowledge change method.