JPH0314130A - Inference controller - Google Patents

Abstract

PURPOSE:To constitute this controller so that an expert system developer can execute an inference of a high level by selecting plural inference systems and combining them freely by leading in inference control information which does not depend on an inference system such as a controller, an event and a queue. CONSTITUTION:The part for executing an inference system and the part for controlling the inference system become separate devices of an inference system executing device 1 and an inference system controller 3, respectively. An inference system control means 34 of the inference system controller 3 determines an event to be executed from a queue storage part 33 in which an execute sequence of an event is stored, in accordance with control information related to the whole inference of a controller storage part 31, and fetches an execution code from an event storage part 32. This execution code is delivered to an inference system executing means 12 and a rule group or a function group is actuated, and generation of a solution is executed. The generated solution is delivered to a user interface device 2 through an inference system control means 34, and brought to output display. In such a way, an expert system developer can execute an inference of a high level by selecting plural inference systems and combining them freely.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to an inference control device for an expert system using knowledge.

Conventional technical expert systems are in the transition period from prototypes to practical systems. In general, expert systems can be broadly divided into diagnostic types, design/planning types, and control types, depending on the target task. Early expert system development was mostly aimed at diagnostic problems. A diagnostic problem is a type of problem in which the search space for solutions is finite and one of the most plausible solutions is determined from observational data. Ta. The inference method used for its development had sufficient descriptive power even if it was a single method. Inference method is a general term for an inference engine that solves problems using knowledge and know-how acquired from experts, and the knowledge representation handled by that inference engine. Typical examples include forward inference method and backward inference method. There are methods etc. A typical example of an expert system that uses forward reasoning is D.
There is ENDRAL. For DENDRAL, see Dendral and Metadendral: There Applications Dimensions, Artificial Intelligence, by Ncannan and Huigenbaum.
Volume 11, 1978 (Buchanan, B. and
Feigenbaum, E.: 5/DENDRAL and Meta-DENDR
AL: The i rAl)pLications
Dimension, Artificial I
Intelligence, Vol. 11.1978). MMCIN is a typical example of an expert system that uses a backward inference method. MMCI
Regarding N, Computer-Based Medical Consultations by Shortlinofe: Mycin American Yersenia 1976 (Short
liffe, E, H,: Computer Ba5ed
Medical Conditions: MY
CIN, American Elsenier 19
76,).

In recent years, the focus of Ex-Heart system development has shifted from the diagnostic type to the design/planning type. A design/planning problem is a type of problem in which an optimal solution is sought while satisfying constraint conditions, and the search space for solutions is infinite. In order to efficiently find a solution to such a design/planning type problem, it is necessary to limit the search space 1j11 by dividing the target problem into several subproblems, and to perform an efficient search. control function is essential.

Problems to be Solved by the Invention However, conventional expert systems have a problem in that they cannot easily perform sophisticated inference control that freely combines the optimal knowledge for each subproblem.

Furthermore, since each conventional expert system can handle only a very limited problem domain and their inference methods are hardly compatible, it is possible to expand the range of problems by coordinating multiple expert systems with completely different inference methods. The problem was that it was not possible to solve the problem.

An object of the present invention is to realize an inference control method that enables an expert system developer to select and freely combine knowledge to make advanced inferences possible, in order to solve such conventional problems.

Means for Solving the Problem In the inference control method of the present invention, the part that executes the inference method and the part that controls the inference method are separate devices called the inference method 7 and -7 formula execution device and the inference method control device, respectively. ○The inference method execution device has a knowledge storage unit that stores expert knowledge, an inference method execution means that executes the inference method to solve the problem, and a situation where the problem solution is stored using the knowledge stored in the inference execution. It consists of a memory section. The inference method execution device has a controller storage section that stores control information related to the entire inference, an event storage section that stores execution codes, a queue storage section that stores the event execution order, and executes an inference method based on the information in each storage section. It consists of inference type control means for controlling the execution unit.

Effect The effect in the above configuration is as follows. In the inference method execution device, the inference method execution means generates a solution for the known facts and hypotheses in the situation storage section using the knowledge stored in the knowledge storage section.

There are also multiple inference method execution means for the knowledge implemented in the inference method execution device. In the inference method control device, the inference method control means determines the event to be executed from the queue storage section storing the event execution order according to the control information regarding the entire inference in the controller storage section, and by executing the event, the inference method control means determines the event to be executed from the queue storage section storing the event execution order. Execution code including information such as whether to execute the device and control information regarding the entire inference is retrieved from the event storage unit. This execution code is passed to the inference method execution means, the Lulu group or the function group is activated, and a solution is generated.

The generated solution is passed to the user interface device via the inference method control means, and is output and displayed to the user by the output means.

Controllers, events, cues, etc. are inference control information that does not depend on inference methods, so it is easy to freely select and combine knowledge. Moreover, it becomes easy to customize inference control information to perform optimal inference control for each expert system to be developed. These make it possible to solve complex problems such as design and planning problems.

Example Embodiments of the present invention will be described below with reference to the drawings.

9. FIG. 1 is a block diagram showing the overall configuration of an expert system according to an embodiment based on the inference control method according to the present invention. The inference method execution device 1 includes a knowledge storage section 11 that stores knowledge of experts, an inference method execution means 12 that executes an inference method for problem solving using the knowledge stored in the knowledge storage section, and an inference method execution unit 12 that executes an inference method for problem solving using the knowledge stored in the knowledge storage section. It is comprised of a situation storage unit 13 that stores problem solutions using stored knowledge.

The user interface device 2 includes an input means 21 for inputting information from the user side and an output means 22 for outputting information from the system side. The inference method control device 3 includes a controller storage unit 31 and a situation storage unit 13 that store control information regarding the entire inference, and an event memory that stores an execution code including information on whether to execute the control information regarding the entire inference. 32, a queue storage section 33 that stores the execution order of events, and an inference method control means 34 that controls the inference method device based on information in the controller storage section 31, event storage section 32, and queue storage section 33.

10/\-zo The inference method control means 34 of the inference method control device 3 controls the inference method execution device 1 via the inference method execution means 12 of the inference method execution device 1. Further, the problem input means 21 and the solution output 22 of the user interface device 2 are controlled by the inference method control means 34 of the inference method control device 3.

The inference method execution means of the inference method execution device 1 is capable of executing knowledge, and by selecting and freely combining knowledge, it is possible to solve a wide range of problems.

FIG. 2 is a block diagram showing the configuration of the controller storage section 31 and its peripheral devices. Controller storage unit 31
The controller table 201 is composed of a controller table 201 that stores control information regarding the entire inference. Controller y-pull 201j: Controller name field 202 that stores a controller name that uniquely identifies the controller, Usage queue field 203 that stores multiple queues that can be used in inference control, Stores active status cues related to inference control. Active state cue field 204, 11 ・\-
7. Inactive state queue field 205 that stores an inactive state queue that is not involved in inference control; Default queue field 206 that stores a default queue that is used when no queue is explicitly specified in inference execution;
It consists of a start event field 207 that stores the event that is executed first at the start of execution.

FIG. 3 is a block diagram showing the configuration of the event storage section 32 and its peripheral devices. The event storage unit 32 includes the device of the situation storage unit 13 and an event table 301 that stores an execution code including information such as whether to execute control information regarding the entire inference. The event table 301 includes an event name field 302 that stores an event name that uniquely identifies an event, an action status field 303 that stores the status of the status storage unit 13 that takes effect when the event is executed, and rules that should be activated in functional units. Start Lulu group field 3 that stores the modularized rule group in
04, an activation function group field 305 that stores the function group to be activated, and a priority field 306 that stores the priority at the time of event execution.

FIG. 4 is a block diagram showing the configuration of the queue storage section 33 and its peripheral devices. The queue storage unit 33 includes a queue table 401 that holds events waiting to be executed and also stores ↑j-information for determining registration/retrieval of the event group. Cue table 401
For example, a queue name field 402 that stores a queue name that uniquely identifies a queue, an event retrieval field 403 that indicates a method for retrieving an event from a queue, and an event registration field 4 that indicates a method for registering an event in a queue.
04, an execution waiting event field 405 that stores an execution waiting event.

FIG. 5 is a processing flowchart of the inference method control means 34. Below, follow the flowchart in Figure 5 from Figures 2 to 4.
Controller storage unit 31 while referring to the diagrammatic configuration.
, the processing of the inference method control means 34 that controls inference using the control information of the event storage section 32 and the queue storage section 33 will be explained.

13.\-/゛When the inference method control means 34 starts operating (step 5
00), first perform preprocessing such as initializing the inference environment and registering an execution start event in the queue (step 505).
. That is, the waiting event field 405 of the queue full 401 is initialized, and the controller table 2 is
The event stored in the start event field 207 of 01 is added to the queue table 4.01(7) with the queue name indicated in the default queue field 206.
It is stored in the pending event field 405 of the record.

After the preprocessing (505) is completed, a queue for extracting an event to be executed is selected (step 510). Select a queue that has events and has a high selection priority. All cues can be in an active or inactive state, but only the active cue is involved in inference control. The status of this queue is 4i1. It is possible to dynamically change the program from the program side while the logic is running.

After selecting the queue, the event to be executed is taken out of the queue (step 515). That is, cutie full 4
In U1, a record having the selected queue in the queue name field 402 is found, and one event is extracted from the waiting event field 4-05 using the method shown in the event extraction field 403.

After retrieving the event, the event is executed (step 520). That is, in the event table 301, the extracted event is entered in the event name field 30.
Find the record with 2 and enter the effect status field 303.
The rule shown in the activation rule field 304 and the function shown in the activation function field 305 are activated for the work table shown in the activation rule field 305.A new event is created by executing this event. may be issued and stored in the pending event field /I 05 K of the queue table 401.

After the event is executed, check whether there is an event to be newly executed in any queue (step 525).
, if the queue exists, processing starts from queue selection in step 510. If it does not exist, post-processing to calculate execution time, statistical information, etc. (step 530) is performed, and the entire inference is terminated (step 535).

By making the part that executes the inference method and the part that controls the inference method into separate devices, the inference method execution device 1 and the inference method control device 3, respectively, completely different knowledge can be freely selected and combined. It's easy to do. Further, the inference method control device 3 includes a controller storage section 31, an event storage section 32, a queue storage section 33 that stores inference control information, and an inference method control means that controls the inference method execution device 1 based on the control information. By separating 34, it is possible to easily customize only the control information for each Exbat system, and it becomes possible to realize optimal inference control.

Effects of the Invention According to the present invention, by introducing inference control information that does not depend on inference methods such as controllers, events, and keys, an expert system developer can select knowledge,
By freely combining them, an inference control system that enables advanced inference can be realized.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the overall configuration of an expert system that is an example of the present inference control method, FIG. 2 is a block diagram showing the configuration of a controller storage unit and its peripheral devices, and FIG. FIG. 4 is a block diagram showing the configuration of the storage unit and its peripheral devices, FIG. 4 is a block diagram showing the configuration of the queue storage unit and its peripheral devices, and FIG. 5 is a flow chart showing the flow of processing in the inference method control means. be. ■... Inference method execution device, 2... User interface use device, 3... Inference method control device, 11... Knowledge storage unit, 12... Inference method execution means, 13... Situation record 17. \-7 Boundary part, 21... Input means, : 31... Controller storage section, 32... Event storage section, 33... Cue storage section, 34... Inference method control means.

Claims (5)

[Claims] (1) A knowledge storage unit that stores expert knowledge, an inference method execution means that executes multiple inference methods for problem solving using the knowledge stored in the knowledge storage unit, and known facts in the inference execution. an inference method execution device consisting of a situation storage unit that stores information and intermediate results; a controller storage unit that stores control information regarding the entire inference; and a controller storage unit that stores information about which inference method is to be executed for which situation in the situation storage unit. an event storage section that stores execution codes including an event storage section, a queue storage section that stores an event execution order, and inference that controls the inference method execution device based on information stored in the controller storage section, the event storage section, and the queue storage section. An inference control device comprising an inference control device comprising a method control means. 2. The inference control device according to claim 1, further comprising a user interface device comprising input means for inputting information to the inference control device and output means for outputting information from the inference control device. (3) The controller storage unit includes a controller table, and this controller table has a field for storing a controller name that uniquely identifies a controller, a field for storing a plurality of queues used in inference control, and a field for storing a plurality of queues used in inference control. A field that stores an active state cue, a field that stores an inactive state cue that is not involved in inference control, a field that stores a default cue that is used when no queue is explicitly specified during inference execution, and a field that stores an inactive state cue that is not involved in inference control. 2. The inference control device according to claim 1, further comprising a field for storing an event to be executed first. (4) The event storage unit is equipped with an event table, and this event table has a field for storing an event name that uniquely identifies an event, a field for storing a situation that affects when an event is executed, and a function for rules to be activated. A field for storing a group of modularized rules, a field for storing a group of functions to be activated,
2. The inference control device according to claim 1, further comprising a field for storing a priority at the time of event execution. (5) The queue storage unit includes a queue table, and this queue table includes a field for storing a queue name that uniquely identifies a queue, a field for indicating a method for retrieving an event from the queue, and a field for indicating a method for registering an event in the queue. 2. The inference control device according to claim 1, comprising: a field; and a field for storing an event waiting to be executed.