JPH03125225A - Ai tool - Google Patents

Abstract

PURPOSE:To simply develop a knowledge base by providing a means which constructs an inference module from each knowledge and an inferring method, a means which constructs an inference algorithm from the inference module, and a means which carries out interactively and with trial and error those two mentioned means. CONSTITUTION:A building block BB production system 1 which constructs the BB is provided together with a BB inspection system 2 which inspects the BB, a BB data base 3 which registers the BB, a bulder 4 which combines interactively the BBs and also inspects the result of combination of BBs, and a total control system 5. A BB serves as a least unit for execution of the infer ence and describes an inference method and the address of a knowledge base used for the inference. Then a list of events are shown on a screen and a user selects these events. Thus it is possible to interactively decide an inference method and to simply develop a knowledge base.

Description

[Detailed description of the invention] [Industrial application field] The present invention relates to AI tools.

[Conventional technology]

With conventional AI tools, the entire knowledge base is described as a result of describing individual pieces of knowledge.

[Problem to be solved by the invention]

Conventional methods include (1) a contradiction between the idea of expanding the system little by little through prototyping and the fact that a good system cannot be created unless the overall design of the knowledge base is well-designed;

(2) It is not possible to componentize the knowledge base.

There was a problem.

An object of the present invention is to provide an AI tool that can solve the above conventional problems.

[Means to solve the problem]

The AI tool according to the present invention has a means for constructing an inference module from individual knowledge and inference methods, a means for constructing an inference algorithm from the inference module, and a means for executing the two means interactively through trial and error. It is characterized by comprising the following.

That is, in the present invention, the construction of a knowledge base is considered in two stages, first creating modules called building blocks, and then systemizing the idea of building the whole by combining pilting blocks. There is. Furthermore, in order to do this efficiently, we have made it possible to verify and combine individual building blocks interactively.

[Effect]

According to the present invention, (1) the knowledge base design can be modularized;
It is now possible to prototype module units while keeping an overall perspective.

(2) Since the knowledge base is modular, there is a high possibility that it can be used as a component when building another knowledge base.

〔Example〕

FIG. 1 is a block diagram showing the configuration of a prototyping-oriented AI tool according to an embodiment of the present invention.

The prototyping-oriented AI tool shown in FIG. a builder 4 that combines BBs and verifies the results of the combination;
It consists of a control system 5 that performs overall control.

BB is an object as shown in FIG. As can be seen from this, the BB is the minimum unit for performing inference, and describes the inference method and the address of the knowledge base used for inference. This BB only has knowledge and inference methods, and the data and problems used for inference can be kept by oneself, or can be received as messages from other BBs, so the block It is highly versatile. Therefore, the behavior of a certain BB can be defined by itself or by another BB. Below, details of the elements of BB will be explained.

First, it is a knowledge base that can describe rules, frames, and procedures. These cannot be mixed and written. Therefore, if the types of knowledge bases are different, they must be described as separate BBs. Multiple knowledge bases of the same type can be specified, and their usage can be defined in the knowledge base control slot. When defined as sequential type, the inference methods written in the inference method are applied to the knowledge base in the order defined in the knowledge base slot.

If parallel type is specified, inference is applied all at once. The inferences that are applied include forward inference, backward inference, hypothetical inference, frame search, and procedural evaluation. Each inference method is standard. In the question slot,
For example, there is such a thing as backward inference, such as whether an event can be verified or not.

Building block construction system l is a system for developing BB. As shown in Figure 3, this system can be used automatically by filling in the knowledge base and inference method described above in the sheet prepared on the computer (Figure 4). The format is such that it can be handled by users, and it is further registered in the building block database 3. Additionally, if the BB specified by the user already exists in the building block database, it will be treated as a request for correction, and the content read from the building block database will be displayed on the sheet in advance, and the user will be asked to correct it. let

The building block database 3 is a database that has block names as search keys, as shown in FIG.

The building block verification system 2 is an environment in which BB verification is performed. A user enters this system and B
When the B name is specified, BB is automatically executed (Figure 6). It is up to the user to decide whether this result is correct. Furthermore, for item 1 given in a message or the like, the user is requested to generate a pseudo-generation of missing data during execution.

For example, let's say that the building block is something that takes in signals from a plant and diagnoses an abnormality.

At this time, since the verification system does not have that signal, the user is asked to simulate the analysis result of that signal (for example, create an event that the temperature is rising from the temperature time series).

Specifically, the system displays a list of such events on the screen and allows the user to make a selection.

This system involves performing individual inferences step by step,
It has the ability to trace message evaluation and inference processes.

After the verification of each BB is completed, the builder 4 is used to assemble the entire BB.

Builder 4 has a structure as shown in FIG. The builder 4 plays the role of interactively executing the BB, and it consists of an interactive building block simulator 8 that controls the interaction, a message stocker 9 that stores the messages assigned to each BB, and a message stocker 9 that executes the BB. It consists of a building block interpreter 10. The specific behavior of the builder is as follows. First, the user starts the interactive building block simulator 8,
Decide which BB to execute. Then, the simulator 8 searches for the block in question from the database 3 and retrieves it.

Here, BB describes how the block behaves when a signal (message) is sent to it from the outside (another block). The message contains transmission information for the block to the receiving DB. Next, the simulator 8 refers to the message stocker 9, searches for the message assigned to the block (that is, the message whose receiving BB is the BB), and extracts the transmission information. The message stocker consists of the message name, sending BB, receiving BB, message content, and tag as shown in FIG. If there are multiple messages addressed to the BB, the one with the largest tag is adopted.

Once this much data is collected, it is passed to the building block interpreter roJ. The interpreter applies the conveyed information to the contents of the block (this is similar to applying data to a program) and executes the block. The execution results of a block are reflected in other blocks through messages. That is, the block sends the execution result as a message and sends the simulator IC 12f.
vinegar. Shiminire, -ta receives this, sending BB, receiving B
B. Classify messages, determine tags, and store in message stocker.

If the operation does not go well, the user can temporarily rewrite the contents of the message stocker from the simulator 10.
It is also possible to simulate.

By repeating this process, the builder simulates the message exchange in a pseudo manner.

If the message is modified in the interactive block building simulator, and the results are satisfactory, the user returns to the building block construction system to modify and verify the building block.

The control system 5 interactively controls the operations of the series of systems described above.

〔Effect of the invention〕

The effects of the present invention are as follows.

(1) The knowledge base construction method is hierarchical, creating a knowledge base with good maintainability.

(2) Since it is possible to decide on inference strategies etc. interactively,
Knowledge base development becomes easier.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of an embodiment of the present invention;
Figure 2 is a diagram showing the configuration of BI3 in Figure 1, Figure 3 is the same (block diagram showing the BB construction system, Figure 4 is a diagram showing the building block construction system sheet type input screen image, Figure 5 is a diagram showing the same building block construction system sheet type input screen image, The figure also shows the building block database, Fig. 6 shows the execution process of the same <BB, Fig. 7 is a block diagram showing the configuration of the builder in Fig. 1, and Fig. 8 It is a diagram showing the configuration of a message stocker. ! Building block construction system, 2...
Building block verification system, 3... Building block database, 4... Builder 5...
Overall control system.

Claims (1)

[Claims] The invention comprises means for constructing an inference module from individual knowledge and inference methods, means for constructing an inference algorithm from the inference module, and means for executing the two means interactively through trial and error. Featured AI
tool.