JPH03186930A - Expert system - Google Patents

Abstract

PURPOSE:To facilitate the improvement of a knowledge base by altering and adding the knowledge base written in a programming language with a natural language. CONSTITUTION:When a user adds a certain rule to the content of the knowledge base 50, the natural language expression of a rule inputted from input means 10-13 is automatically translated in a rule in the program language and transmit ted to the knowledge base 50. Assuming that the knowledge base 50 is written in LISP and when 'the operation of dust exhaustion is acknowledged', the rule of 'a dust exhaustion acknowledgement operation rule class is inferred'. Such a addition by the natural language is executed through the input means. Thus, [dust exhaustion occurring operation 3 ( operation is acknowledged dust exhaustion) ( forward-chain dust exhaustion acknowledgement operation rule class)] is stored in the knowledge base in the form of LISP expression. Thus, the content of the knowledge base can easily be altered and added.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of the Invention The present invention relates to an expert system with an improved user interface that allows natural language input.

[Prior art] In an expert system, even after the system is completed, the developer and, if necessary, the user can change the contents of the knowledge base after examining the output inference results in order to obtain more appropriate inference results. You need to do or add something.

However, knowledge bases are created using convenient programming languages such as LISP, PROLOG, and C for system designers to construct knowledge bases, so if you are not a direct developer, you may not be able to write the contents of the knowledge base. It was extremely difficult to make changes or additions. In addition, attempts have been made to use general-purpose natural language interpretation mechanisms in order to obtain a more comfortable user interface, but such devices do not require lexical analysis means, syntactic analysis means, or even the results of syntactic analysis internally. The converter that converts into formats itself is a much larger system than an expert system, and is not something that can be installed in currently used expert systems.

[Problem to be Solved by the Invention] In view of the above-mentioned actual situation, the problem of the present invention is to enable system developers and system users to easily change the contents of the knowledge base, such as program functions, without imposing a large burden on the system.・Providing the expert system with additional functionality.

[Means for Solving the Problems] According to the present invention, the above problems are achieved by providing a rule-type knowledge base that stores program functions described in rule expressions, and an inference mechanism that performs inference using the knowledge base. A natural language input editor mechanism is built into the equipped expert system, and this input editor is used for input using a dictionary file that stores natural language in correspondence with the program functions used in the knowledge base, and this dictionary file. This problem can be solved by comprising a language conversion unit that converts the natural language expression into a corresponding program function expression and outputs it to the knowledge base.

[Operation] According to this configuration, when a user adds a rule to the content of the knowledge base, the natural language expression of the rule input from the input means is automatically translated into a programming language by the natural language human editor. Since the rules are translated into rules and sent to the knowledge base, you can easily improve the knowledge base even if you are not familiar with the programming language. For example, if this knowledge base is written in LISP, if you now "recognize the operation of a piece of garbage," you can input a rule in natural language that says, ``Infer the operation rule class when recognizing a piece of garbage.'' When done through means, this can be done using a natural language human editor:
-CHAIN Operation rule class when recognizing out of garbage))
It is stored in the knowledge base in the form of a LISP expression.

[Effects] As mentioned above, the expert system according to the present invention can change and modify the knowledge base written in a programming language.
Since additions can be made using natural language, even those who are not familiar with programming languages can easily improve the knowledge base. Other features, operations, and effects of the expert system according to the present invention will become clear from an understanding of the embodiments described in detail below with reference to the drawings.

[Example] Hereinafter, an example of the expert system according to the present invention applied to the control of an incinerator will be described with reference to the drawings.

In FIG. 1, an incinerator 1 is provided with a plurality of types of sensors 2, and each sensor 2 is connected to an instrumentation panel 3. Further, an industrial television camera 4 is arranged inside the incinerator 1, and this industrial television camera 4 is connected to a data processing computer 6 via a first signal line 5, and an instrumentation panel 3 is connected to the computer 6 for data processing. It is connected via a second signal line 7 to a data processing computer 6 .

This data processing computer 6 is connected to an inference computer 9 via a third signal line 8. Further, this inference computer 9 is connected to a keyboard 10, a monitor device 11 as an output device, an audio output device I2, and a printer 13.

As shown in FIG. 2, the inference computer 9 includes a data communication interface 80 connected to the data processing computer, a knowledge base 50 consisting of rules written in the Lisp language, and an inference mechanism 60. , a natural language input editor mechanism 70, and an I10 interface 90 for relaying with external input/output devices 10.11.12.13 such as monitors and printers.

The knowledge base stores the empirical rules of experts regarding the operating technology of the incinerator 1 as a plurality of production rules. This production rule specifies empirical rules in the form of a condition part and an action part. For example, if the set temperature of incinerator 1 is specified in the condition part, and the current temperature of incinerator L is higher than the set temperature, "High temperature" specified for the operating part
The judgment value is the conclusion.

Furthermore, production rules are hierarchically classified by rule class, and a rule class is provided for each task. This task includes operating the control means for operating the incinerator I, and calculating the amount of operation required as a subtask to perform the operation operation. Therefore, the knowledge base collects production rules related to each task and classifies them as rule classes, and classifies the production rules classified into each rule class as lower rule classes for each subtask that forms each task. Production rules classified into each lower-level rule class are hierarchically classified into further lower-level rule classes.

FIG. 3 shows the specific contents of the knowledge base constructed regarding the operating technology of the incinerator 1. In this diagram, it is shown in natural language for easy understanding, but of course in the actual knowledge base, it is written in the procraming language, here L.
It is written in ISP. And as shown in this figure,
The knowledge base consists of rule classes for sensor abnormality judgment, combustion data individual judgment, and comprehensive combustion situation judgment.Each of these rule classes, for example, ``combustion situation general judgment'', can be used to control multiple lower-level judgments such as ``driving operation judgment''. It consists of rule classes.

Furthermore, this ``driving operation judgment'' is composed of lower-level rule classes such as ``garbage-out operation j'', and this ``garbage-out operation'' rule class is further subdivided into ``operation when recognizing out of garbage''.
It consists of the lowest rule class such as "Operation when garbage is running out". This lowest rule class is a production rule such as "operation 1 when garbage is running out".
It is made up of rules.

The control of the incinerator by the expert system having the above configuration will be explained below.

Temperature, pressure, etc. are detected by each sensor 2 as an index indicating the combustion state of the incinerator 1, and the detected measurement values are transmitted to the measurement value data processing computer 6 in a predetermined data format at the instrumentation panel 3.

In addition, an industrial television camera was used to image the inside of the incinerator 1,
An image signal of the captured image is transmitted to a data processing computer 6 via a second signal line 5. Then, the processing computer 6 calculates the amount of steam, etc. from the input various measured value data, and also calculates the state of the flame, the amount of garbage, etc. from the image, which cannot be directly detected by the sensor 2. Collect measurement data. and,
The collected measured value data is transmitted to the inference computer 9 as the main body of the expert system via the third signal line 8, and the inference computer 9 transmits the data to the incinerator 1.
The current incineration state of the incinerator 1 is determined, the future combustion state is inferred, and the operating method of the incinerator 1 corresponding to the inferred combustion state is determined as a result of the inference, and the monitoring device 11, the audio output device 12, and the printer 13 Output to.

The incinerator is controlled based on the inference results output in this way, but it is difficult to completely and logically store the knowledge of experts in the knowledge base when building this system. In addition, current expert systems do not have the learning ability of humans, so they do not necessarily yield optimal results, and the content of the knowledge base must be modified or added by the end user or system constructor. . In the system according to this invention, when changing the knowledge base ◆In addition work, for example, the L
When the program function of the ISP is input in the form of natural language, the natural language human editor mechanism 70 works to create the ISP program function.
It is converted into a program function expression in P and sent to the knowledge base 50.

Next, the natural language input editor mechanism 70 will be described; as shown in FIG. 2, the natural language input editor mechanism 70 basically consists of a dictionary file 71 and a conversion section 72.

As shown in Figure 4, the dictionary file 71 includes all LISP functions used in the knowledge base, that is, the rule base, including the four arithmetic operations, logical operations, comparison operations,
It is classified and stored as user functions, inference control, etc., and each LTS P function has a LISP function name, natural language name, LISP expression, natural language expression, number of arguments,
It has values such as argument content, type of function, whether conditional part is used, and whether conclusion part is used. For example, in Sousaninch, the LISP function name is “$Sousaninch”.
, "Operation recognition" as a natural language name, "As a LISP expression" ($Sousaninch XI) "As a natural language expression"
(Recognized the operation of XI)" and the number of arguments are registered as "■".

The language conversion unit 72 converts the input rules expressed in natural language into rules expressed in LISP language without accessing the dictionary file, and stores the rules in a predetermined position of the knowledge base.

For example, in the rule class hierarchy shown in Fig. 3, the "Out of Garbage ■ Operation j" rule class is located even lower in the "Driving operation judgment" rule class, which is below the "Comprehensive combustion status judgment" rule class. An example of adding a production rule with the following content after "Operation 2 when garbage is running out" will be presented here.

The contents of the additional rules are: [Rule name] Operation 3 when out of garbage occurs [Condition part] Operation of out of garbage has been recognized [Conclusion part]
The operation rule class is inferred when recognizing out of garbage.

This is a natural language expression, and is actually a rule in the form of the following LISP expression: ((Operation 3 when garbage is running out) → ($FORWARIl-CHAIN Operation rule class when garbage is recognized)) Here, $FORWARD-CHA IN and $FORWARD-CHA IN are LISP functions, and the garbage outage and garbage outage recognition operation rule class are their arguments, respectively.

In this embodiment, the process of adding rules is menu-based. Step 1: Selection of rule classes and rules When the knowledge base modification program is started, the topmost layer of rule classes, as shown in Figure 5a, is added. The rule class will be displayed and you will be waiting for the input of the rule class you want to modify. If "3" is selected here, as shown in FIG. 5b, the rule class located at the next level after the "Comprehensive Combustion Status Judgment" rule class is displayed, and the system waits for input again. If you select "2" here, the next layer of rule classes will be displayed again, and by repeating the process of selecting the desired rule class, you will reach the menu that displays the production rules located at the bottom layer. 5th
"2" is selected here in order to select Add as shown in Figure c and then instruct the location of the rule to be added.

Through this series of operations, the position of adding the next new rule to be input is determined after the "Operation 2 when running out of trash occurs" rule in the "Operation during running out of trash" rule class. This position is stored in the pointer means of the natural language knowledge 3 editor 70.

Step 2: Enter the rule contents according to the menu screen displayed in order, first input the rule name (Figure 5d), then input the condition part (Figure 5e).
figure). As mentioned above, the new rule to be added uses a LISP function called ``Sousaninch'', so ``4'' is selected here. Regarding the selection items shown in FIG. 5e, the type of function is displayed using the dictionary structure already explained in FIG.

The next menu displays all applicable user functions (Figure 5f). In other words, among the functions stored in the dictionary file, only those that are user functions and have the conditional part used -Y are displayed in natural language color. Here, operation recognition is selected. Next, an input menu for the arguments of the LISP function for operation recognition is displayed by referring to the number of arguments and the contents of the arguments stored in the dictionary file (FIG. 5g).

Now that the input of the condition part has been completed, the data input so far can be confirmed using the confirmation menu shown in FIG. 5H.

In this work, the condition part is displayed on the screen in the pattern of the "natural language expression" of the dictionary, and at the same time the rLI of the dictionary file is displayed on the screen.
The conditional part is converted to ($Sousaninch garbage cut) using "SP expression".

The conversion process is performed by the language conversion unit 72.

In other words, at this stage, "((operation 3 when out of garbage occurs ($ out of garbage))) →" is stored in the buffer of the natural language conversion unit 72.

Step 3: Inputting the conclusion part When entering the conclusion part, the menu shown in Figure 51 is displayed, but the rules to be added are FOR,
In order to use the inference-activated LISP functions WAR and D-CHAIN, "3° inference control" is selected here.

Next, a function having inference control as its type is retrieved from the dictionary file and its natural language menu is displayed as shown in FIG. 5j, so select "2. Start inference". Next, the argument manual menu for the inference-activated LISP function is displayed as shown in the fifth figure by referring to the "number of arguments" and "argument contents" in the dictionary.

Below, select hierarchically and finally select the "Operation when recognizing out of trash" rule class. This completes the human effort in the conclusion section, and finally the fifth section. Confirm the input using the confirmation menu shown in figure f' and exit.

At this stage, the buffer of the natural language conversion unit 72 is
ORWARD-CHAIN Operation rule class when recognizing out of garbage)) has been added, and the overall operation 3 ((operation 3 when out of garbage occurs) ($SOUNA NINCH OUT OF TARGET) → ($FORWARD-CHAIN Operation rule class when recognizing out of garbage) ).This converted L
The ISP function, ie the rule to be added, is stored at a predetermined location in the knowledge base 50 indicated by the pointer means. After the expert system is restarted, the added rules will become available.

6

[Brief explanation of drawings]

The drawings show one embodiment of the expert system according to the present invention. Fig. 1 is a block diagram of incinerator equipment using the expert system, Fig. 2 is a block diagram of the expert system, and Fig. 3 is a concrete example of the knowledge base. An explanatory diagram showing the contents, Figure 4a is an explanatory diagram showing part of the reserved functions of the dictionary file, Figure 4b is an explanatory diagram showing part of the user functions of the dictionary file, and Figures 5a to 51 are new rules added. FIG. (50)...Knowledge base, (60)...Inference mechanism,
(70)...natural language input editor mechanism, (71)...
...Dictionary file, (72)...Language conversion unit.

Claims (1)

[Claims] a rule-type knowledge base (50) that stores program functions described in rule expressions; and the knowledge base (50).
and a natural language input editor mechanism (70), the editor mechanism storing program functions used in the knowledge base and natural language in correspondence with each other. A language conversion unit that converts an input natural language expression into a corresponding program function expression using this dictionary file (71) and outputs it to a knowledge base. expert system.