JPH03185522A - Inference engine - Google Patents

Abstract

PURPOSE:To attain more rapid inference by functionally expressing a processing unit to be used for an inference process by the language near the hardware side and describing a part for advancing an inference also by the language of the hardware side. CONSTITUTION:An execution part 5 for combining plural functions each of which expresses the processing of an inference process such as pattern matching is obtained by compiling or assembling a program described by C language (C) or assembly language (assembler). An inference part 6 for accessing a necessary function in accordance with a fact applied to an I/O part 4 and an intermediate processing result and advancing an inference is also described by the C or assembler. Thereby, the execution part 5 is previously compiled or assembled and then integrated in the inference engine. Consequently, respective inference processes at the time of executing the inference can rapidly be processed and the inference itself can be executed at high speed.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to an inference engine in a production system used for an expert system or the like, and particularly to an inference engine that operates at high speed.

[Conventional technology]

FIG. 3 is a block diagram showing a production system using a conventional prolog processing system.
2 is a rule base that is a collection of rules, 2 is a fact base that is a collection of facts, 3 is an inference engine that executes inference, 4 is an input/output unit where the operator inputs facts and goals, 3a is a Pro
The inference section 3b written in log is a Prolog interpreter that interprets and executes a Prolog program.

Next, the operation will be explained. In production systems, the rules stored in rule base 1 are generally
(Premise part) THEN (Conclusion part) format.

Further, the facts input to the input/output unit 4 and the facts obtained during the execution of the inference are stored in the fact base 2. When performing forward inference, for example, the inference engine 3 searches the rules in rule base 1 from the beginning, searches for those whose premises match the facts in fact base 2, and if there is a matching rule, it Add the conclusion part to fact base 2 as a new fact, search rule base 1 again, and when no rule matching the facts is found, use the appropriate one of the obtained facts as the conclusion. do.

When performing the above-described inference, pattern matching between the condition part or the conclusion part of the rule and the facts is repeatedly performed, so the inference engine 3 needs to be a processing system equipped with a pattern matching function. Therefore, processing systems such as Prolog and Li5p, which have powerful pattern matching functions and list processing functions, have been used as the inference engine 3.

For example, if a Prolog processing system is adopted, the inference section 3a, facts, and rules are programmed using Prolog. And as shown in Figure 3.

The Prolog interpreter 3b interprets the inference section 3a and sequentially searches the rule base 1 while performing pattern matching of rules and facts in the rule base 1.

[Problem to be solved by the invention]

Since the conventional inference engine 3 is configured as described above, the inference is proceeded by an interpreter such as the Prolog interpreter 3b, and therefore there is a problem that the inference speed cannot be improved beyond a certain level. In particular, as the number of rules increases and the number of inference functions increases, the reduction in inference speed becomes more pronounced.

This invention was made to solve the above-mentioned problems, and it is possible to significantly improve the inference speed compared to the conventional method, and it also provides users with the ability to
The purpose of this invention is to obtain an inference engine that can behave as if it were performing inference using a processing system such as rolog.

Means for Solving the C111 Problem] The inference engine according to the present invention performs each process in the inference process when performing inference such as pattern matching using the C3
It is written using functions written in a so-called hardware-oriented language, such as Japanese language or assembly language, and has an execution section that is a collection of these functions, and is written in a hardware-oriented language.

It is provided with an inference section that calls a function of the execution section in accordance with the facts given to the input/output section or the facts obtained during the execution of the inference, and causes the function to be executed in a block manner.

[Effect]

Since the execution unit in this invention is incorporated into the inference engine after being compiled or assembled, it processes each inference process at high speed when executing inference, and ultimately executes inference at high speed. .

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings. 1st
In the figure, reference numeral 5 denotes an execution unit that collects processes in the inference process such as pattern matching into a single function, and includes C language (hereinafter referred to as C) and assembly language (
It is a compilation or assembly of a program written in (hereinafter referred to as an assembler). Further, 6 is an inference section that advances inference by calling necessary functions according to the facts given to the input/output section 4 and intermediate processing results, and is also written in C or assembler. Other parts are the same as those shown in FIG. 3 with the same reference numerals.

Next, the operation will be explained. First, a description will be given of how to create the execution section 5 and the inference section 6 when Prolog is adopted as the basic programming language. First, P
The functions, rules, and built-in predicates of rolog are written in subroutine format in C or assembler. Furthermore, these are compiled or assembled (linked if necessary) to form the execution unit 5. On the other hand, access operations to the rule base 1 and fact base 2 are described in subroutine format in C or assembler for each appropriate unit. Then, parts other than the subroutine parts are written in C or assembler.Next, these are compiled or assembled and the inference section 6
shall be. Then, by linking the execution unit 5 and the inference unit 6,
Stored in a storage unit of a computer, etc.

Next, the inference execution operation will be explained with reference to the flowchart shown in FIG. First, when a fact is input to the input/output unit 4 (step 5T1), the inference unit 6 executes this routine by calling a routine for storing it in the fact base 2, and stores the input fact in the fact base 2. (Step 5T2). Then, the pattern matching routine of the execution unit 5 is called to match the rule base 1.
Pattern matching is performed between the premise of each rule and the facts of fact base 2 (Step 5T3). Here, there is a match (Step 5T4), and it is necessary to execute the procedure described in the conclusion. When it occurs (step S
T5), the Prolog interpreter 3b is called and processed (step 5T6). In addition, the conclusion part of the rule can be treated as a new fact in Fact Base 2.
(Step 5T7), a routine for storing in the fact base is called (Step 5T8). After repeating the above process, the inference ends when there are no more rules with a premise that matches the facts.

In this way, the inference engine 3 is implemented in a hardware-oriented language based on the Prolog processing system, so that it appears from the outside that inference is proceeding with the Prolog processing system, and at the same time, the processing speed is can be significantly improved. The degree of improvement in processing speed varies depending on the number of rules and facts, but
An inference system with about 0 rules will be several tens of times faster.

Note that the basic processing system is not limited to Prolog, but may be any other system as long as it is suitable for pattern matching and list processing.

〔Effect of the invention〕

As described above, according to the present invention, the inference engine is configured such that the processing unit used in the inference process is expressed as a function in a language that is closer to hardware, and the part that advances the inference is also written in a language that is closer to hardware. Therefore, it has the effect of being able to perform inference faster without reducing the functionality of conventional inference engines.

[Brief explanation of drawings]

FIG. 1 is a flowchart showing the configuration of a system including an inference engine according to an embodiment of the present invention, FIG. 2 is a flowchart showing the operation of the inference section, and FIG. 3 is a block diagram showing the configuration of a system including a conventional inference engine. It is a diagram. 1 is a rule base, 2 is a fact base, 3 is an inference engine, 4 is an input/output section, 5 is an execution section, and 6 is an inference section. In addition, in the figures, the same reference numerals indicate the same or corresponding parts.

Claims (1)

[Claims] In an inference engine that sequentially searches a rule base containing a set of rules and draws a conclusion according to facts given to the input/output section, each processing unit used in the inference process when searching the rule base is an execution unit that is a collection of functions written in a language that is similar to the hardware; An inference engine characterized by comprising an inference section that calls a function.