JPH02214939A - Control method for inference execution system - Google Patents

Abstract

PURPOSE:To shorten the overall executing time by dividing an inference execution system into plural groups in a hierarchical structure and starting the inference of a certain group with only the instruction given from a group of a higher hierarchy. CONSTITUTION:The production systems PS 100, 110, 120, 130, 111 and 112 show those groups which are scattered based on the knowledge, the inference executing function (inference engine) and its meaning and nature of an inference execution system 1. Each PS has a hierarchical structure and is controlled so as to be started by another PS of a higher hierarchy. A common data area 3 is granted to perform the free reference and write via any PS. Thus the inference is regularly carried out toward the PSs of higher hierarchies, and the undesired PSs are not started. As a result, the inference executing speed is improved.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method of controlling an inference execution system such as an expert system, and particularly to a method of controlling a distributed cooperative inference execution system.

[Prior art] As the problems targeted by expert systems become larger and more complex, the number of knowledge and rules increases, which makes design more difficult and reduces the efficiency of inference execution. Problems began to occur. Furthermore, as systems become larger in scale, flexibility, expandability, and maintainability of the system are becoming essential requirements. One way to deal with these situations is to use knowledge and inference execution functions (inference engines).
are divided into multiple groups based on their meanings and properties, each group makes inferences about the given classification, and the overall problem is solved by each group cooperating in a predetermined manner. A method has been proposed. As a representative example, Information Processing Society Journal Vol. 1.27
No. 6 (June 1986) “Distributed production system with communication/synchronization function Po-PS (Procedure 0rie
nted Production System) is known.

[Problems to be Solved by the Invention] However, conventional distributed cooperative systems such as Po-PS have the following problems.

(1) Since information is transmitted between groups in the form of messages, message communication must be performed many times when transmitting information from one group to many groups. Therefore, it takes time, and
It is difficult to design because the procedures are complicated.

(2) When importing external data, the only options are for each group to import the necessary data separately, or for one group to import it all at once and pass it to each group via message communication. It is difficult to design because it takes time and the procedures are complicated.

(3) Since all groups are activated regardless of whether or not it is time to perform inference, there are many unnecessary activations, and therefore processing time is increased.

(4) Since all groups are considered to be at the same level and there is no concept of hierarchy, it is difficult to efficiently configure a large-scale and complex system.

This invention was made to solve the above-mentioned problems, and aims to provide a control method for a distributed cooperative inference system that is easy to construct a practical system and that can execute inference at high speed.

[Means for Solving the Problems] The above problem is a method for controlling an inference execution system that stores correspondence relationships between elements as knowledge in a knowledge file and makes inferences based on the knowledge. divided into multiple groups based on their meanings and properties,
The inference execution method is characterized in that the groups are configured in a hierarchical structure, the inference of a certain group is activated only by an instruction from a group in a higher hierarchy, and a pig area is provided that can be commonly referred to by each group. The problem is solved by the system control method.

Here, the correspondence between elements is the correspondence between cause and effect,
It includes all correspondence relationships, such as correspondence between design conditions and design elements.

[Function] In the above inference execution system, (1) Inference in a certain group is activated only by an instruction from a group in a higher hierarchy, so there is no need to activate or execute groups that do not need to be executed. Overall execution time is reduced.

(2) Since each group is configured in a hierarchical structure, construction of a large-scale and complex system can be easily realized.

(3) By providing a data area that can be commonly referenced by each group, the number of message transmissions between groups is reduced, which makes system construction easier even when data is transmitted in a complex manner, and improves system processing efficiency. It also becomes more expensive.

[Example] Hereinafter, one example of the present invention will be described with reference to FIGS. 1 and 2. FIG. 1 shows an inference execution system controlled by the control method of the present invention. In FIG. 1, 1 is an inference execution system, 2 is a time management timer, 3 is a common data area, 100.110.120.130.111.11
2 shows each group in which the knowledge and inference execution function (inference engine) of the inference execution system 1 is distributed into multiple groups based on their meanings and properties, and is called a production system (hereinafter abbreviated as PS). . Each PS is hierarchical, and in FIG. 1, PSloo is located in the topmost first hierarchy, PSllo 120.130 is located in the second hierarchy, and PSIII 112 is located in the third hierarchy. In this inference execution system, the PS of each layer
is controlled so that it is activated only by PSs in higher layers. The common data area 3 is allowed to be freely referenced and written by any PS.

The operation of this inference execution system 1 will be explained below.

First, PSloo at the highest level is activated. P
Sloo may be activated when there are no more groups to be activated within the inference execution system, or may be activated by time management timer 2, external interrupts, or other tasks outside of the inference execution system 1. good. PSL
oo refers to the common data area 3, performs inference using its own knowledge base, unique inference engine, and unique data area, and as a result writes necessary data to the common data area 3, or It is sent as a message to the lower hierarchy, determines the PS in the second hierarchy that needs to be activated, and activates it. Preferably, the design is such that data that requires complicated transmission is written in the common pig area 3, and data that requires simple transmission is sent as a message to a lower layer. For example, if PSIIO is started, P
Sllo refers to the common data area 3, performs inference using its own knowledge base, unique inference engine, and unique data area, and as a result writes necessary data to the common data area 3, or A third-layer PS that sends a message to a lower layer and needs to be activated
Decide and start. In this way, only the PSs that need to be activated are sequentially activated from the upper layer to perform the necessary inference. Of course, it is also possible for the n-th layer PS to activate the PSs in the (n+2> layer or lower) layers. In this case, depending on the system configuration, the level of each PS is not fixed at the beginning, and the It is also possible to dynamically change the PS level, such as placing all PSs activated from the PS in the (n+1)th hierarchy.
When a plurality of Ss are activated, the PSs in that hierarchy are executed sequentially, and the PSs in the hierarchy below are controlled to start executing after all the PSs in that hierarchy have finished executing.

In this way, the inference is executed in an orderly manner from the PS in the upper hierarchy to the PS in the lower hierarchy, and PSs that do not need to be activated are not activated, so that inference can be executed at high speed. In addition, the receiving and passing of data between each of the 18
Complex data can be processed via the common data area 3, and only simple data can be processed in the form of messages between 18 units, so the system is easy to construct and has high processing efficiency.

FIG. 2 is a diagram showing a part of a specific example (measures against abnormal rise in furnace temperature) in which the inference execution system controlled by the present invention is used in a control system of a waste incinerator. The parts shown in FIG. 2 of the control system of the waste incinerator include a scheduler 11, a preprocessing task 12, a furnace temperature increase task group 13,
It is composed of post-processing tasks 14 and has a common data area 3 that can be referenced and written in common by these tasks. The scheduler 11 is positioned at the top first layer and controls each task.

The scheduler 11 first starts the preprocessing task 12. The preprocessing task 12 inputs process data of the waste incinerator and stores it in the common data area 3. After that, the scheduler 11 starts the furnace temperature increase task group 13. The furnace temperature rise task group 13 predicts a rise in the furnace temperature by inference, determines the optimal countermeasure by inference based on the result, and when it is determined that the countermeasure has been taken and an abnormal rise in the furnace temperature can be avoided. , has the function of determining by inference what procedure should be used to return to the normal control state.

The furnace temperature rise task group 13 includes a furnace temperature rise prediction task 15 that predicts an abnormal rise in the furnace temperature and a return to normal, a furnace temperature rise countermeasure task 16 that determines the optimal countermeasure for abnormal furnace temperature 4, and a furnace temperature rise countermeasure task 16 that predicts the furnace temperature to return to normal. It consists of a furnace temperature rise recovery task 17 that determines the procedure for returning to the normal control state when returning to the normal control state.The furnace temperature rise prediction task 15 is at the second level, and the furnace temperature rise countermeasure task 16 and furnace temperature rise recovery task 17 It is positioned in the third layer.

First, the scheduler 11 executes the furnace temperature rise prediction task 1.
5 is activated. The furnace temperature rise prediction task 15 applies the process data stored in the common data area 3 to the furnace temperature prediction rules stored in the knowledge base to predict the occurrence of an abnormal rise in the furnace temperature and the return to a normal state. do. and,
When an abnormal rise in the furnace temperature is predicted, the furnace temperature rise countermeasure task 16 is started, and when it is determined that the abnormal rise in the furnace temperature can be avoided, the furnace temperature rise recovery task 17 is started. When the furnace temperature rise countermeasure task 16 is activated, the furnace temperature rise countermeasure task 16 applies the process data stored in the common data area 3 to the countermeasure determination rules stored in the knowledge base and determines the optimal countermeasure. common data area 3
Write to the inference result area inside. Furnace temperature rise recovery task 17
When the furnace temperature rise recovery task 17 is activated, the furnace temperature rise recovery task 17 applies the process data stored in the common data area 3 to the countermeasure determination rules stored in the knowledge base, determines the optimal recovery procedure, and returns the process data to the common data area 3. Write in the inference result area of 3.

Finally, when the scheduler 11 starts the post-processing task 14, the post-processing task 14 outputs the data in the inference result area in the common data area 3.

The furnace temperature increase countermeasure task 16 and the furnace temperature increase task 17 are as follows:
Both cannot be started at the same time. Further, when neither an abnormal rise in the furnace temperature nor a return to normal is detected by the reasoning of the furnace temperature rise prediction task 15, the furnace temperature rise countermeasure task 16 also performs the following.
The furnace temperature rise recovery task 17 is also not activated.

In this way, unnecessary startup execution can be prevented, and complex transmissions can be performed via the common data area 3, making system construction easier and increasing system processing efficiency.

In this embodiment, each group has been described as being composed of a production system, but in the present invention, each group is not limited to a production system, and may be a collection of other knowledge expressions.

[Effects of the Invention] As explained above, according to the control method of the present invention, the inference execution system is divided into a plurality of cruises based on their meanings and properties and configured in a hierarchical structure, and the inference activation of a certain group is Since this is done only by instructions from groups in the upper hierarchy, there is no need to start or execute groups that do not need to be executed, reducing the overall execution time. Furthermore, since each group is configured in a hierarchical structure, it is possible to easily construct a large-scale and complex system. Furthermore, since we have created a data area that each group can reference in common, the number of message transmissions between groups is reduced, making system construction easier even when data is transmitted in a complex manner, and improving system processing efficiency. It gets expensive.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an inference execution system controlled by the control method of the present invention, and FIG. 2 is a part of a specific example in which the inference execution system controlled by the present invention is used in a control system for a waste incinerator FIG. 3 is a diagram showing countermeasures against abnormal rise in furnace temperature. 1. Inference execution system, 3...-Common data area, 1
00.110, 120.1.30, 111, 112...
- A group of inference execution systems (production systems).

Claims (1)

[Claims] A control method for an inference execution system that stores correspondence relationships between elements as knowledge in a knowledge base file and performs inference based on the knowledge, the inference execution system being divided into a plurality of groups based on their meanings and properties. The groups are configured in a hierarchical structure, and the activation of the inference of a certain group is as follows:
A control method for an inference execution system, characterized in that control is performed only in response to instructions from a group in a higher hierarchy, and a data area is provided that can be commonly referenced by each group.