JPH0573315A - Large-scale inference device - Google Patents

Abstract

PURPOSE:To accelerate the inference processing speed applying a large-scale inference rule and concept data by storing a pair of inference rule and concept data necessary for the inference in a first storage area and storing the one to be made access at the other inference processing time in a second storage area. CONSTITUTION:In the case of the application for the inference device estimating the cause of a fault in a power system, an inference management means 4 fetches the content of an inference order designation data 13 and outputs a command storing a pair of the inference rule and the concept data relating to a substation in an active storage area 18 to a data transfer means 5 so as to specify a substation where the fault occurs according to the designated order. The data transfer means 5 fetches the inference data from an auxiliary storage means 1 to be stored in an active storage area 18 based on the command. Then, an inference management means 4 outputs a starting command to an inference means 3 after confirming the storage. The inference means 3 uses a pair of the inference rule and the concept data relating to the stored substation, executing the inference processing according to the order.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an inference device used in a production system for solving various problems such as identification of the cause of an accident and failure diagnosis, and in particular inference using a huge amount of inference rules and conceptual data. To a large-scale reasoning device.

[0002]

2. Description of the Related Art Such a production system is
Various problems are solved by using inference rules that represent knowledge of the inference target world as rules and conceptual data that represents the state of the inference target world. Specifically, in general, a matching process for extracting an inference rule in which conceptual data satisfying the condition part (IF part) of the inference rule exists, and an inference rule to be executed from the inference rules obtained by this matching process. Conflict resolution processing for selecting only one and the execution unit (THEN) of the inference rule selected in this conflict resolution processing.
Section) and the inference execution process that repeatedly executes the process defined by the section) until there is no executable inference rule. Therefore, when a huge amount of inference rules and conceptual data are involved,
It takes a lot of time to match the inference rules with the conceptual data, and it takes a considerable amount of time to reach a conclusion.

Therefore, in order to perform the inference at a high speed, conventionally, when the inference rule and the conceptual data are collated, the RETE algorithm for executing the collation at a high speed by having the intermediate result of the previous collation process. Method (RE
TE: A Fast Algorithm for the Many Pattern /
Many Object Pattern Match Problem, Atlanticia
lInteligence vol. 19, No. 1, pp. 17-37 (1982) has been proposed.

Further, the inference rule and the conceptual data are divided respectively to reduce the range of the collation process, and thereby the inference process is performed at high speed (TREAT: A Better Match Al-gorithm for).
Al Production Systems, AAAI-87, pp. 42-47
(1987)) has been proposed.

Further, similar to the above-mentioned TREAT algorithm method, a method for dividing the inference rule and the conceptual data to perform high-speed inference processing (Japanese Patent Laid-Open No. 63-63).
No. 208127) has been proposed.

Further, a method of forming a set for each related inference rule and conceptual data, adding a priority to each set, and executing inference in descending order of priority (Japanese Patent Laid-Open No. 64-88746).
Issue).

[0007]

However, the above-mentioned conventional techniques have the following problems. First, RET
In the E-algorithm method or the TREAT algorithm method, the inference rules are re-matched only for the updated conceptual data, but in normal inference, all inference rules are applied and knowledge processing is rarely performed. In many cases, there is a knowledge utilization order. Further, conventionally, not only is conceptual data updated by the execution unit of the inference rule,
It may be updated by the program. For this reason, since the collation with the inference rule that became unnecessary or the collation processing at unnecessary timing is performed, it takes time to maintain the intermediate result, and it is always subject to the collation processing, which is a wasteful collation. Processing occurs, and it is not possible to achieve sufficiently high speed processing.

Further, in Japanese Patent Laid-Open No. 63-208127, the inference rule and the conceptual data are divided, and the matching process is performed only on the inference rule and the conceptual data necessary for inference. In, a certain high-speed processing is achieved. However, each time the inference is executed, the inference rule and the conceptual data are loaded from the auxiliary storage device to the main storage device, and the division processing is performed on the main storage device, so that unnecessary data is not stored. Extra time is required for loading and dividing processing such as, and the high-speed processing is hindered accordingly.

Further, in JP-A-64-88746, the number of inference targets is limited and the inference processing is reduced, but since the inference rules and concept data of all the sets are stored in the main storage device. , It is necessary to identify and extract a corresponding set of inference rules and conceptual data each time the matching is performed, resulting in unnecessary processing time.

On the other hand, conventionally, in Japanese Patent Laid-Open No. 1-237729, only the inference rules are divided and stored in an auxiliary storage device, and only the inference rules applied when executing the inference are restored in the main storage device. Is proposed. According to this, the collation process of the inference rule can be speeded up, but since the concept data is not divided, all of the concept data are subject to collation. Therefore, unnecessary collation processing occurs and high-speed processing is hindered.

An object of the present invention is to solve the above-mentioned conventional problems. In other words, a large-scale inference apparatus capable of performing inference processing applying a large-scale inference rule and conceptual data at a higher speed. To provide.

[0012]

In order to achieve the above object, the present invention is based on the premise that the inference rule and concept data are divided into a plurality of sets to speed up the inference processing.

However, even if the inference rules and the concept data are divided into a plurality of sets, if all the sets are stored in the main storage means, useless comparison processing occurs in the comparison processing of the inference rules and the concept data. ..

In order to eliminate such waste, only one set of inference rules and concept data necessary for a certain inference is stored in the main storage means, and other sets of inference rules and concept data are saved in the auxiliary storage means. It is possible to keep it.

However, depending on the content of the inference, it may be necessary to frequently access another set of inference rules or conceptual data that has already been used for inference. In this case, the auxiliary storage means is accessed, but as is well known, the access time of the auxiliary storage means is longer than that of the main storage means, which hinders high-speed processing.

Therefore, according to the present invention, the first and second storage areas are separately provided in the main storage means, and a set of inference rules and concept data required for inference is stored in the first storage area and is already used. Among the inference rules and the conceptual data that have been used, those that are expected to be accessed during other inference processing are stored in the second storage area to minimize the target of the matching process and the target of the matching process. However, the reasonably accessed inference rules and conceptual data are stored in the main memory which has a short access time.

Based on such an idea, the large-scale inference apparatus of the present invention comprises: (1) auxiliary storage means for storing inference rules representing knowledge of the target world and concept data representing states of the inference target world; (2) main storage means in which the inference rules and concept data in the auxiliary storage means are stored as necessary,
(3) inference means for performing inference using inference data and concept data in the main storage means, and (4) control of inference rules and transfer of concept data between the auxiliary storage means and the main storage means. In addition, the reasoning management means for controlling the activation of the reasoning means, and (5) the reasoning rule and concept between the auxiliary storage means and the main storage means or in those storage means based on the command of the reasoning management means. (6) The inference rule and the conceptual data stored in the auxiliary storage means are associated with a plurality of inference sections divided from the reasoning execution and are related to each other. Stored in a plurality of groups each including a plurality of inference rules and a plurality of groups including a plurality of conceptual data. (7) The main storage means includes a first storage area and a second storage area. Is divided into regions, 8) The inference means sequentially reads the inference rules stored in the first storage area of the main storage means in response to the start command given from the inference management means, and satisfies the condition part of the read inference rules. Collating whether the conceptual data to be present exists in the first storage area,
The inference rule including the conceptual data satisfying the condition part is extracted, and the process defined in the conclusion part of the extracted inference rule is executed. (9) The inference management means is provided with the given inference. A command for transferring a set of inference rules and a set of conceptual data of an inference section corresponding to a rank command from the auxiliary storage unit to the first storage area of the main storage unit is output to the data transfer unit, and the inference unit is also provided. Based on the process of outputting a start command to the instructor and the conclusion of each inference section output from the inference means in response to the start command, the predetermined inference rule and concept data related to the conclusion are stored in the first storage area. A command to transfer to the second storage area, and the other first command
A command for saving or deleting the inference data and the conceptual data of the region in the auxiliary storage means, and a set of the inference data and the conceptual data of the inference section designated by the conclusion or the inference order command are stored in the auxiliary storage. Since a command to transfer from the means to the first storage area of the main memory means is output to the data transfer means, a process of outputting a start command to the inference means is repeatedly executed. is there.

[0018]

According to the present invention, since it is configured as described above,
The object of the present invention is achieved by the following actions. In the collation processing of the inference processing performed by the inference means, a set of inference rules and concept data retrieval / collation targets stored in a first storage area (hereinafter referred to as an active storage area) of the main storage means. Since it is limited to inference rules and conceptual data (hereinafter collectively referred to as a set of inference data, as appropriate),
Unnecessary collation processing is suppressed, and necessary collation processing is completed in a short time.

The inference rules and the conceptual data, which are frequently accessed when the conclusion part of the inference rule is executed, but are not necessary for the collation processing, are stored in the second storage area of the main storage means (hereinafter referred to as inactive storage area). Is stored in
The unnecessary collation process is suppressed, and the access time is shortened as compared with the case where the inference data is stored in the auxiliary storage means, so that the inference process is further speeded up.

Further, according to the method in which unnecessary inference data is deleted after the processing of a certain inference section is performed, it is possible to suppress an unnecessary increase in the scale of the auxiliary storage means.

Further, the data transfer means transfers the inference data between the storage means or within the storage means, including the processing of storing only the inference data necessary for the progress of the inference processing in the active storage area of the main storage means. Since it is made possible to perform such processing, it is possible to always perform inference with a constant performance without depending on the operating system.

[0022]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to FIGS. FIG. 1 shows a basic block diagram of the large-scale inference apparatus of this embodiment, and FIGS. 6 and 7 show flowcharts of the inference processing procedure. As shown in FIG. 1, the device configuration includes an auxiliary storage device 1, a main storage device 2, an inference means 3, an inference management means 4, a data transfer means 5, and a user program means 6. Has been done. The auxiliary storage means 1 stores an inference rule group 11, a conceptual data group 12, and inference order designation data 13. The main storage means 2 is
An inactive storage area 17, an active storage area 18,
It is partitioned into working storage area 19.

Next, a case where the detailed configuration of each unit is applied to an inference device that estimates the cause of a failure in the power system will be described as an example. FIG. 2 shows a conceptual configuration diagram of an example of an electric power system that is the world of inference. As shown in the figure, the electric power system includes a plurality of substations (A, B, C). And
Each substation is composed of a plurality of constituent devices (A, B, ...). Furthermore, each component device has a plurality of parts (A,
B, ...). In the case of a power system of this configuration, when a failure occurs at a certain substation, it is assumed that which part of the component equipment constituting the substation is the cause is determined by using inference processing.

Here, a basic execution method for estimating the cause of failure in the substation will be described. First, in order to identify the substation where the fault exists, the inference is executed using only the inference rules that determine which substation in the target substation is faulty and the conceptual data that represents the state of the entire substation. To do. Next, in order to identify the component device in the identified substation that has a failure, the inference rule for determining which component device of the target component device has failed and the status regarding the component device of the substation are shown. Perform inference using conceptual data.
Finally, in order to identify which part of the identified component device has failed, the inference rule is used to determine which part of the target component has failed, and the inference using conceptual data about the component in the component device. To execute. It is assumed that the cause of failure is specified by such inference of three stages.

The inference rule group 11 divides the inference rules representing the rules related to the inference target world according to the step structure (inference division) of the failure cause inference order of the substation described above, and stores each division as a set. Has been done. As an example of the inference rule, a description example of the inference rule for determining the failure of the substation is shown in FIG. As shown in the figure, the structure of each inference rule is as follows: -name part 201: a part that describes the name of the inference rule-condition part 202: part that describes the condition of the conceptual data to which the inference rule applies-inference part 203: Consists of a part that defines an operation to be executed when the condition part is satisfied. Specifically, as described in the condition unit 202, this inference rule is a condition that this rule is applied if the state of the item “power transmission amount” of the conceptual data “substation A” is “decreased”. Is represented. Then, when this condition part is satisfied, as shown in the conclusion part 203, in the column of “failure” in the item “failure” of the conceptual data “failure substation” stored in the concept data group 12, This represents an operation to be performed, that is, "set location A".

On the other hand, the conceptual data group 12 divides the conceptual data representing the state of the inference target world according to the step structure (inference division) of the failure cause inference order of the substation described above, and one set for each division. Is stored as. As an example of the conceptual data, a description example showing the state of the substation is shown in FIG.
As shown in the figure, the concept data is defined by: name part 301: portion describing the name of the concept data item name 302 item state (value etc.) 303. Specifically, as shown in the example of FIG. 4, the content of the item “power transmission amount” of the concept data whose name is “Substation A” is in the “decreased” state.

The inactive storage area 17 of the main storage means 2 stores predetermined inference rules and conceptual data which are not necessary for the collation processing but may be needed at the time of executing the conclusion part of the inference rules. .. The inference rules and conceptual data stored in this storage area can be referred to or updated by the user program means 6 or the rule execution section of the inference means 3.

The active storage area 18 of the main storage means 2 is a storage area for storing inference rules and conceptual data used for inference processing. The inference means 3 executes collation processing for the inference rules and the conceptual data stored in the active storage area 18. Also, the active storage area 18
The inference rules and the concept data stored in the user program means 6 are stored in the user program means 6 as in the case of the inactive storage area 17.
Alternatively, the rule execution unit of the inference means 6 can be referred to or updated.

The inference management means 4 interprets the inference order designation data 13 stored in the auxiliary storage means 1, the user program means 6 or the instruction designated by the rule execution section of the inference means 3, and the data transfer means 5 or An operation command is output to each inference means 3. In this embodiment, the inference management means 4
The following nine commands can be specified, and the commands are output in the procedure shown in FIG. 5, for example.

DATA_MOVE_TO_ACT50
1: Instructs the data transfer means 5 to store the specified sequence of conceptual data in the active storage area 18.

DATA_MOVE_TO_NOACT
502: Instructs the data transfer means 5 to store the specified sequence of conceptual data in the inactive storage area 17.

DATA_MOVE_TO_SUB50
3: Instructs the data transfer means 5 to store the specified sequence of conceptual data in the auxiliary storage area.

RULE_MOVE_TO_ACT50
4: Instructs the data transfer means 5 to store the specified sequence of inference rules in the active storage area 18.

RULE_MOVE_TO_NOACT
505: Instructs the data transfer means 5 to store the specified sequence of inference rules in the inactive storage area 17.

RULE_MOVE_TO_SUB50
6: Instructs the data transfer means 5 to store the specified sequence of inference rules in the auxiliary recording means 1.

INFERENCE 507: Inference mechanism 3
Instruct the execution of inference processing.

DELETE_DATA 508: Instructs the data transfer means 5 to delete the specified sequence of conceptual data from the inactive storage area 17, the active storage area 18, and the auxiliary storage means 1.

DELETE_RULE 509: Instructs the data transfer means 5 to delete the specified sequence of inference rules from the inactive storage area 17, the active storage area 18, and the auxiliary storage means 1.

The inference order designation data 13 is data for designating the order of inference rule and conceptual data transfer (saving, restoration) and inference execution between the respective storage means or storage areas. The inference management means 4 interprets the contents of this designation, and instructs the inference means 3 or the data transfer means 5 to operate.

The data transfer means 5 transfers the designated inference rule or concept data between the inactive storage area 17, the active storage area 18 or the predetermined storage area of the auxiliary storage means 1 according to the instruction of the inference management means 4. Evacuate and recover.
First, it is determined whether the object of evacuation / recovery is an inference rule or conceptual data. Next, the inference rule or the conceptual data is inactive storage area 17, active storage area 1
8. Check the position of the auxiliary storage means 1. If the storage area instructed by the inference management means 4 is the same as the checked actual storage area, there is no need for transfer, and the processing of the data transfer means 515 ends. If the storage areas are different, the inference rule or concept data is stored in the storage area designated by the inference management means 4.

According to the instruction from the inference management means 4, the inference means 3 first performs a matching process between the inference rule in the active storage area 17 and the conceptual data. That is, all inference rules for which conceptual data satisfying the condition part of the inference rule exist are found. Next, an inference rule to be executed is selected from the found inference rules according to a preset rule. Then, the operation specified by the execution unit of the selected inference rule is executed. This selection and execution cycle
Repeat until the inference rules found are gone. Further, when performing the matching process for finding all the inference rules for which the conceptual data satisfying the condition part of the inference rule exists, the result of the condition determination performed in the past is stored in the working memory area 19.

Using the inference apparatus of the embodiment thus configured, the operation procedure of the inference processing of the cause of the power system accident shown in FIG. 2 will be described below with reference to the flowcharts shown in FIGS. 6 and 7. explain.

First, the inference rule and the conceptual data to be applied as the initial processing are stored in the auxiliary storage means 1 in advance. That is, the respective states relating to the substation, the component equipment, and the parts are divided into concept data regarding the substation, concept data regarding the component equipment, and concept data regarding the parts, and stored as the concept data group 12 of the auxiliary storage unit 1. Further, the rules for identifying the substation, component equipment, or part in which the failure has occurred are stored in the inference rule group 11 in the form of inference rules.

Next, when a start command is given to the inference apparatus of this embodiment, the inference management means 4 fetches the contents of the inference order designation data 13 in step S1 of FIG. In order to identify the existing substation, in step S2, a command to store a set of inference rules and conceptual data regarding the substation in the active storage area 18 is output to the data transfer means 5. In response to this, in step S3, the data transfer means 5 retrieves the inference data consisting of the set of inference rules and conceptual data designated according to the instruction from the auxiliary storage means 1 and stores it in the active storage area 18. Then, in step S4,
The inference management means 4 outputs a start command to the inference means 3 after confirming that the designated inference data is stored in the active storage area 18. In response to this, the inference means 3 executes the inference process in step S5 in accordance with the procedure shown in FIG. 7 using the inference rule and the conceptual data set regarding the substation stored in the active storage area 18.

In the inference processing, as shown in FIG. 7, first, in step S51, it is determined whether or not there is conceptual data that satisfies the condition part of the inference rule, the active storage area 18 is accessed, and collation processing is executed. Extract an inference rule in which conceptual data satisfying the condition part exists. Then, the determination result is stored in the working storage area 19 (step S52).
Next, it is judged whether or not all the inference rules and the conceptual data have been collated (step S53), and if completed, the inference rules extracted according to the rules are selected one by one and defined in the conclusion section. The performed operation is executed (step S54). For example, when the result that the substation C is out of order is obtained, “substation C” is recorded in the conceptual data of the failed substation as described above. Then, it is determined whether or not the inference execution has been completed for all the extracted inference rules, and if so, it is reported to the inference management means 4 (step S55). The inference management means 4 having received this report identifies the component-level failure of the substation C at the component device level based on the inference order designation data in step S6 of FIG. A command for fetching the data set from the auxiliary storage means 1 and storing it in the active storage area 18 is output to the data management means 5. At that time, since the inference rules and the conceptual data regarding the substation C stored in the active storage area 18 may be referred to or updated during execution of the failure-specific inference at the component device level, the inactive storage area 17 may be updated.
The command to transfer to is output. Further, since the inference rule and the conceptual data regarding the substations other than the substation C are not necessary for the inference at the component device level of the substation C, the instruction to save or delete the auxiliary storage means 1 is output.
Based on this, the data management means 5 causes the inference management means 4 to
The inference data is transferred in response to the command (step S7). The inference management means 4 confirms that the saving and recovery of the predetermined inference data is completed, and issues a start command to the inference means 3 (step S8). Based on this, the inference means 3
Executes the same inference processing as described above according to the procedure shown in FIG. 7 (step S9). In other words, the inference is executed using the inference data that is newly restored in the active storage area 18 and that is made up of the set of inference rules at the component device level and the conceptual data. At this time, since the inference data on the substation C related to the previous inference execution process is stored in the inactive storage area 17, the inference rules and the conceptual data on the substation C are displayed while executing the inference on the constituent devices. It is possible to refer to or update at high speed. Then, based on the inference result regarding the identified component device, the inference for identifying the component is executed in the same procedure as described above to identify the faulty component of the power system equipment in FIG. it can.

Although the above embodiment has been described as being applied to an apparatus for diagnosing the cause of a failure in a power system, as another application example of the present invention, a large amount of data parts such as a failure diagnosis of nuclear power plant equipment is used. It can be applied as it is to failure diagnosis in the field to be handled.

[0047]

As described above, according to the present invention,
In the collation processing of the inference processing performed by the inference means, the object of retrieval / collation of the inference rules and the conceptual data is limited to the set of inference rules and conceptual data stored in the first storage area of the main storage means. The unnecessary matching process is suppressed, and the necessary matching process is completed in a short time.

Further, since the inference rule and the concept data which are frequently accessed when the conclusion part of the inference rule is executed but are not necessary for the collation process are stored in the second storage area of the main storage means, the unnecessary collation process is performed. Is suppressed and the access time is shortened as compared with the case where the inference data is stored in the auxiliary storage means, so that the inference process is further speeded up.

Further, according to the method in which unnecessary inference data is deleted after the processing of a certain inference section is performed, it is possible to suppress an unnecessary increase in the scale of the auxiliary storage means.

In addition, the data transfer means transfers the inference data between the storage means or within the storage means, including the processing of storing only the inference data necessary for the progress of the inference processing in the active storage area of the main storage means. Since it is made possible to perform such processing, it is possible to always perform inference with a constant performance without depending on the operating system.

[Brief description of drawings]

FIG. 1 is a basic configuration diagram of a large-scale inference apparatus according to an embodiment of the present invention.

FIG. 2 is a conceptual configuration diagram of an example of a power system of an inference target world.

FIG. 3 is a diagram showing an example of an inference rule.

FIG. 4 is a diagram showing an example of conceptual data.

FIG. 5 is a diagram showing an example of inference order designation data.

FIG. 6 is a flowchart showing an inference processing procedure according to an embodiment.

FIG. 7 is a flowchart showing a detailed processing procedure of a part of the flowchart of FIG. 6;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Auxiliary storage means 2 Main storage means 3 Inference means 4 Data transfer means 5 Inference management means 6 User program means 11 Inference rule group 12 Conceptual data group 13 Inference order designation data 17 Inactive storage area 18 Active storage area 19 Working storage area

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Ken Honda Honda 5-2-1 Omika-cho, Hitachi-shi, Ibaraki Hitachi Process Computer Engineering Co., Ltd. (72) Toshihiko Nakano 5-chome, Omika-cho, Hitachi-shi, Ibaraki 2-1 Hitachi Ltd. Omika factory

Claims (1)

[Claims] 1. Auxiliary storage means for storing inference rules representing knowledge of the inference target world and concept data representing states of the inference target world, and inference rules and concept data in the auxiliary storage means as required. Main storage means to be stored, inference means for performing inference using inference data and concept data in the main storage means, transfer of inference rules and concept data between the auxiliary storage means and the main storage means And inference rules and concept data between the auxiliary storage means and the main storage means or in those storage means based on a command of the inference management means. And inference rules and concept data stored in the auxiliary storage means are associated with a plurality of inference divisions divided from the reasoning execution, respectively. Stored in a plurality of sets including a plurality of inference rules and a plurality of sets including a plurality of conceptual data, and the main storage means is divided into a first storage area and a second storage area. The inference means sequentially reads the inference rules stored in the first storage area of the main storage means in response to the activation command given from the inference management means, and the condition section of the read inference rules. Whether or not the conceptual data satisfying the condition is present in the first storage area, the inference rule in which the conceptual data satisfying the condition part is present is extracted, and is defined in the conclusion part of the extracted inference rule. The inference management means executes a set of inference rules and a set of conceptual data of an inference section corresponding to a given inference order command from the auxiliary storage means to the first storage means of the main storage means. Transfer to storage Associated with the conclusion based on the processing of outputting the command to the data transfer means and outputting the start instruction to the inference means, and the conclusion of each inference section output from the inference means in response to the start instruction. A predetermined inference rule and a command to transfer the conceptual data from the first storage area to the second storage area;
A command for saving or deleting other inference data and conceptual data in the first area in the auxiliary storage means, and a set of inference data and conceptual data in the inference segment designated by the conclusion or the inference order command. And a command for transferring the set from the auxiliary storage unit to the first storage area of the main storage unit, and a process for repeatedly executing a process for outputting a start command to the inference unit after outputting to the data transfer unit. A large-scale reasoning device.