JPH03250222A - Inference system - Google Patents

Abstract

PURPOSE:To improve efficiency for searching a solution by selectively executing inspection from a hypothesis with high possibility to be established while using knowledge showing the possibility to establish the hypothesis. CONSTITUTION:A hypothesis inspection process 20 collects data 200 required for inspecting the hypothesises in a hypothesis set 61 based on inspecting method knowledge 53. An inspected result evaluation process 30 evaluates the hypothesis based on the inspection method knowledge 53 and the data 200 collected by the hypothesis inspection process 20. When the final result is arrived as the result of the evaluation, an inference result 300 is outputted and when the final result is not arrived yet, it is judged whether the hypothesis set 61 is generated or corrected. Further, a hypothesis set correction process 40 removes the hypothesises in the hypothesis set 61 or moves the hypothesises between the hypothesis set 61 and a standby hypothesis set 62. Thus, by selectively executing the inspection from the hypothesis with the high possibility to be established, the efficiency is improved for searching the solution.

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial Application Field) The present invention relates to an inference method such as a so-called diagnostic expert system.

(Prior Art) In recent years, as computers have become more sophisticated, so-called expert systems have been developed that allow computers to perform some of the tasks that required skill, which were previously handled by experts.

In particular, diagnostic expert systems are equipped with an inference system that, given an initial event, infers the causal event for that initial event, and the performance of this inference system greatly influences the performance of the entire expert system.

By the way, in general, such inference systems generate hypotheses using causal relationship knowledge that contains knowledge showing the causal relationships between events, and then verify the hypotheses by collecting various data about these hypotheses. get the solution.

FIG. 2 shows an example of a diagrammatic representation of the content of the causal relationship knowledge used to generate this hypothesis.

In the figure, for example, event A is shown to have a causal relationship with events E and F.

That is, using this causal relationship knowledge, when event A is given as an initial event, candidate event DSE.

F is proposed as a hypothesis.

However, since all of these hypotheses are treated equally, even hypotheses that are originally unlikely to hold true are tested in the same way as hypotheses that are likely to hold true.

Therefore, in many cases, it takes a long time to arrive at the final solution, resulting in a problem of poor solution search efficiency.

(Problem to be solved by the invention) As mentioned above, in the conventional inference method, all causal events are treated equally as hypotheses, and even hypotheses with a low probability of being established are verified, so it is difficult to solve the problem. The problem was that the search efficiency was low.

The present invention was made to solve such problems,
This provides an inference method that can increase the efficiency of searching for solutions by selectively testing hypotheses.

[Structure of the Invention] (Means for Solving the Problems) The present invention includes a hypothesis generation unit that generates a hypothesis explaining a given phenomenon, and a hypothesis verification unit that verifies the hypothesis generated by the hypothesis generation unit. In the inference system that solves a problem by Alternatively, the hypothesis verification unit includes means for returning a hypothesis that has been put into a temporary standby state in the hypothesis generation unit to a verification target, and a means for deleting a hypothesis that is a target of verification from a temporary standby state. and means for changing the state.

(Function) In the present invention, in addition to so-called causal relationship knowledge, knowledge indicating the possibility of a hypothesis being established, such as the frequency of appearance as a cause, is used to selectively verify hypotheses that are likely to be established.

Therefore, there is no need to unnecessarily verify hypotheses that are unlikely to hold true, and the efficiency of searching for a solution can be increased.

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

FIG. 1 is a diagram illustrating the configuration of an inference system according to an embodiment of the present invention.

As shown in the figure, this inference system includes a hypothesis set generation process 10, a hypothesis verification process 20. It consists of a verification result evaluation process 30 and a hypothesis set modification process 40.

In addition, as knowledge for each process to perform processing, causal relationship knowledge 51, control judgment knowledge 52, verification method knowledge 2
53 are available.

The causal relationship knowledge 51 stores events that have a causal relationship with a certain event in a manner similar to the conventional example shown in FIG. For example, the event DSE, F is a candidate event that has a causal relationship with the event A, and the causal event H1, J
This is an intermediate hypothesis for SK.

Further, the control judgment knowledge 52 includes knowledge indicating the possibility that a hypothesis will be established, such as the frequency of appearance as a cause.

The verification method knowledge 53 includes knowledge necessary to verify whether a hypothesis exists, such as what data is necessary to confirm the existence of a certain event.

Next, each process of this inference system will be explained in more detail.

The hypothesis set generation process 10 is a process of re-creating the hypothesis set 61, and generates a hypothesis set 61 and a standby hypothesis set 62 by referring to the causal relationship knowledge 51 and control judgment knowledge 52 for the input initial event 100. do. For example, when event A is given as an initial event, the hypothesis set generation process [O] generates candidate events based on the causal relationship knowledge 51, identifies E and F, further generates the event using the control judgment knowledge 52, and generates ESF into the hypothesis set. 61 and a standby hypothesis set 62. This means that if event F occurs and the probability is lower than H, it will occur, E will be placed in the hypothesis set, event F will be placed in the standby hypothesis set, and the verification of event F will be put on hold.

The hypothesis verification process 20 collects data 200 necessary for verifying the hypotheses in the hypothesis set 61 based on the verification method knowledge 53. For example, if there is an event and E in the hypothesis set, the data necessary to confirm the existence of the event, the data necessary to deny the existence of the event, and the data necessary to confirm the existence of the event E. Either the necessary data or the data necessary to deny the existence of event E is collected.

The verification result evaluation process 30 evaluates the hypothesis based on the verification method knowledge 53 and the data 200 collected by the hypothesis verification process 20. As a result of the evaluation, if the final result has been reached, the inference result 300 is output, and if the final result has not been reached, it is determined whether to generate or modify the hypothesis set 61. For example, if the verified hypothesis is an intermediate hypothesis such as events D1E and F, the next hypothesis set 61 is generated by the hypothesis set generation process 10 for this intermediate hypothesis. Further, if the hypothesis is denied as a result of the verification, the hypothesis set 61 is modified by the hypothesis set modification process 40.

The hypothesis set modification process 40 removes hypotheses within the hypothesis set 61 and moves hypotheses between the hypothesis set 61 and the standby hypothesis set 62. This removes the hypothesis whose existence is denied by the verification result evaluation process 30 from the hypothesis set 61. Also,
When the existence of a hypothesis in the hypothesis set 61, for example, an event H, is denied, a standby hypothesis set 6 is created based on the control judgment knowledge 52.
A hypothesis in a standby state in No. 2, for example, event F, is moved to hypothesis set 61.

In this inference system, inference is performed by repeating the process described above.

Therefore, hypotheses are classified into a set of hypotheses to be verified immediately and a set of standby hypotheses that are in a state of waiting for verification, based on control judgment knowledge, and hypotheses with a low probability of being established are only recognized when a hypothesis with a high probability is denied. Since the verification is performed, the efficiency of searching for a solution can be improved.

[Effect of the invention] The present invention uses knowledge indicating the possibility that a hypothesis will be established,
Since hypotheses with a high probability of being true are selectively verified, it is not necessary to unnecessarily verify hypotheses with a low probability of being true, and the efficiency of searching for a solution can be increased.

[Brief explanation of drawings]

FIG. 1 is a diagram showing the configuration of an inference system according to an embodiment of the present invention, and FIG. 2 is a diagram illustrating causal relationship knowledge. 10... Hypothesis set generation process, 20... Hypothesis verification process, 30... Verification result evaluation process, 40...
・Hypothesis set modification process, 51... Causal relationship knowledge, 5
2... Control judgment knowledge, 53... Verification method knowledge, 61
...Hypothesis set, 62...Standby hypothesis set, 100...
- Initial event, 200...data, 300...inference result.

Claims (1)

[Claims] (1) In an inference system that solves a problem using a hypothesis generation unit that generates a hypothesis that explains a given phenomenon, and a hypothesis verification unit that verifies the hypothesis generated by the hypothesis generation unit, the hypothesis generation unit , a means for limiting the hypotheses to be verified by the hypothesis verification unit, and placing the remaining hypotheses in a temporary standby state for verification or deleting them from the verification targets; An inference method characterized in that the section has means for returning a hypothesis placed in a temporary standby state by the hypothesis generating section to a verification target, and means for changing a hypothesis that is a verification target to a temporary standby state. .