JPH0652503B2 - Inference processor - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Object of the Invention (Industrial field of application) The present invention relates to an inference processing apparatus that advances inference processing by executing production rules, and in particular rules applicable in the process of inference processing. The present invention relates to an inference processing device that speeds up inference processing by clarifying in advance.

(Prior Art) In a system that executes inference processing by using production rules, a production rule having a condition that matches information about the current state given during the inference processing and a conclusion drawn during the inference process is generated from a large number of production rules. A process of deriving a final conclusion is performed while sequentially extracting and applying these rules. In such processing, it is effective to speed up the inference processing so that the next applicable rule can be extracted at high speed in the inference process.

Conventionally, as a method for quickly extracting a production rule, each condition part and execution part of the production rule are examined, and which of the following conditions is satisfied by the conclusion obtained by the execution of which execution part, that is, which rule is applied? Next, prepare in advance the relationship as to which rule can be applied in the form of a table or network, and at the time of inference, refer to these relationships and select the next applicable rule sequentially. , A method for speeding up inference processing has been proposed (JP-A-60-7203).
And JP-A-62-19940).

However, in these methods, since only applicable relationships between individual rules are described, only the process of selecting the next applicable rule can be speeded up. Therefore, it is not possible to suggest the direction of global reasoning, for example, whether it is possible to lead from one state to another state, or which rule should be applied before reaching a certain conclusion. However, it was difficult to further speed up the inference process by executing useless rules.

(Problems to be Solved by the Invention) As described above, in the conventional inference processing apparatus, the global direction of inference was not performed, so that wasteful execution of production rules hinders further speeding up of inference processing. There was a problem.

The present invention has been made in view of the above problems, and can give a global orientation of inference processing to efficiently execute a production rule, thereby increasing the speed of inference processing. The purpose is to provide a device.

[Structure of the Invention] (Means for Solving the Problem) An inference processing device according to the present invention is configured to execute a condition part and an execution in an inference processing device that sequentially extracts applicable production rules and advances inference processing. A rule storage unit that stores a plurality of production rules composed of pairs of sets, and a setting unit that sets nodes for all possible combinations of variables handled in the production rules stored in the rule storage unit And a connecting means for connecting the nodes by a production rule satisfying the transition condition between the nodes set by the setting means to obtain a directed graph matrix, and a directed graph matrix obtained by the connecting means is decomposed into strongly connected components. Decomposition means and conditional operation for obtaining a condition corresponding to the strongly connected component from the value of the variable of each strongly connected component obtained by this decomposition means Strongly connected with a step and a registration means for obtaining a directed graph matrix using a production rule satisfying the transition condition between strongly connected components and registering it as meta-knowledge indicating which production rule can be applied next to which production rule. The component decomposition part, and each strongly connected component obtained by the strongly connected component decomposition part, with the period of the greatest common divisor of the length of any closed circuit passing through any node in the strongly connected graph, and the production rule After increasing the period by dividing appropriately,
A classification part that divides into a plurality of classes and obtains a production rule that satisfies the variable conditions of these classes and transition conditions of each period as meta-knowledge, and the meta-knowledge obtained by the strongly connected component decomposition part and the classification part is stored. A meta-knowledge storage section
By recognizing the separation between the current state and the target state based on the information indicating the production rule applicable next to the meta knowledge stored in the meta knowledge storage unit, the target state is further approached. Direction of global inference to measure the application of the production rule for, to extract the next applicable production rule stored in the rule storage unit based on this, and an inference unit that executes inference processing. Is provided.

(Operation) According to the present invention, a production rule that satisfies the transition condition between strongly connected components is obtained as meta-knowledge in the strongly connected component decomposition unit, and further obtained by further decomposing each strongly connected component in the classification unit. Production rules that satisfy the transition conditions between classes are required as meta-knowledge. Then, the meta-knowledge is used to determine the general direction of the inference, the order of application of the production rules until the final conclusion is reached is determined based on the meta-knowledge, and the order is stored in the rule storage unit based on the order. Since it is possible to extract the production rules and execute the inference processing, it is possible to prevent unnecessary execution of the production rules and further speed up the inference processing.

If the classification unit appropriately divides the production rule to increase the period of each strongly connected component and then classifies each strongly connected component, classification is possible even in the case where the strongly connected component cannot be classified. Thus, extremely effective meta-knowledge can be obtained in directing the global reasoning.

Embodiment An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing the arrangement of an inference processing apparatus according to this embodiment. That is, this inference processing apparatus is composed of a rule storage unit 11, a strongly connected component decomposition unit 12, a classification unit 13, a meta-knowledge storage unit 14, an inference unit 15, and a state storage unit 16.

The rule storage unit 11 stores a plurality of production rules including a pair of a condition unit and an execution unit.

The strongly connected component decomposition unit 12 extracts the production rule stored in the rule storage unit 11 before the inference processing by the inference unit 15 to generate meta-knowledge. Here, a production rule satisfying the transition conditions between all possible states of each variable handled in the production rule stored in the rule storage unit 11 is obtained as a directed graph matrix, and the directed graph matrix is further connected to the strongly connected component. The production rules that are decomposed and satisfy the variable condition of each strongly connected component and the transition condition between each strongly connected component are obtained as meta-knowledge.

The classification unit 13 calculates the period of each strongly connected component obtained by the strongly connected component decomposition unit 12, and divides the strongly connected component into a plurality of classes based on the value. Then, the production rules satisfying the variable conditions of these classes and the transition conditions between the classes are obtained as meta-knowledge.

The meta-knowledge storage unit 14 stores the meta-knowledge obtained by the strongly connected component decomposition unit 12 and the classification unit 13.

The inference unit 15 sequentially reads out the production rules stored in the rule storage unit 11 and executes inference processing. The inference unit 15 stores the inference process from a given state to another target state as a meta-knowledge memory. A global decision is made based on the meta-knowledge stored in the section 14, and the production rules are executed efficiently.

The state storage unit 16 stores the initial state and the intermediate state of the inference processing in the inference unit 15.

Next, the operation of the inference processing device configured as described above will be described.

It is now assumed that the rule storage unit 11 stores six production rules as shown in FIG. Here, x and y are variables, and a, b and c are states.

First, prior to the inference process, the strongly connected component decomposition unit 12 generates meta-knowledge from the production rule according to the algorithm shown in FIG.

Step S1 First, the condition parts and execution parts of all the production rules stored in the rule storage part 11 are examined, all possible values of each variable handled in the rule are extracted, and all combinations thereof are extracted. , Node numbers are sequentially assigned from 1 '. Here, the maximum number is n. For example, in the rule of FIG. 2, there are three possible states for variables x and y, a, b, and c. Therefore, all combinations of possible states for variables x and y are shown in FIG. There are 9 ways as shown. Therefore, node numbers 1'to 9'are assigned to these nine combinations.

Step S2 Next, a directed graph matrix is generated from the obtained conditions of each node. That is, by connecting the nine nodes shown in FIG. 4 by a rule satisfying the transition condition between these nodes, a directed graph matrix as shown in FIG. 5 is obtained. This matrix describes the rule number i that can transit from the node number j to the node number k.

In order to obtain this matrix, the size is n in step S2.
When a matrix (table) A of × n is prepared and the state of the node number j satisfies the condition part of the i-th rule and the state of the node number k is obtained as the execution result of the rule, (ij of the matrix A is ) The process of registering the rule number i in the component and registering 0 in the other parts is performed for all production rules.

Step S3 Here, the directed graph matrix is decomposed into strongly connected components. Fourth
In the example of the figure, nodes 2'to 8'are grouped as a strongly connected component, as shown in FIG. As a method of strongly connected component decomposition for a directed graph matrix, a simultaneous row and column permutation method is known. In this method, rows are exchanged and columns are exchanged at the same time so that, for example, the upper and lower parts of the matrix are all 0, and the elements gathered on the diagonal of the matrix are strongly connected components.

Step S4 For each strongly connected component thus obtained, the logical sum of the conditions of the values of the variables corresponding to the node numbers included therein is taken to obtain the condition corresponding to that strongly connected component. For example, since the strongly connected component shown in FIG. 6 includes nodes 2'-8 ', the condition of the strongly connected component is (x = b) v (x ≠ y).

Step S5 Next, a directed graph matrix is obtained using a production rule that satisfies the transition condition between strongly connected components. That is, a table having a size of m × m is first prepared, where m is the number of strongly connected components obtained, and if the node a belongs to the cth strongly connected component and the node b belongs to the dth strongly connected component, the table (C
The rule number registered in the (ab) component of the strongly connected component decomposition matrix obtained earlier is registered in the d) component. This operation is performed for each matrix component of the strongly connected component decomposition matrix obtained in step S3. This graph matrix is stored in the meta-knowledge storage unit 14 as meta-knowledge.

Next, the classification unit 13 calculates the calculated cycle of the strongly connected component. Here, the period of the strongly connected component is the greatest common divisor of the length of any closed circuit passing through any node in the strongly connected graph. This value is the same for every node in the graph. If this period is a number of 2 or more, the strongly connected component can be further divided.

Now, when this cycle is calculated for the strongly connected components in FIG. 6, the length of the closed path 3 '→ 7' → 6 '→ 3' passing through the node 3'is "3", and the closed path 3 '→ 5' → 2 '. → 4 '→ 6' → 3 '
The period is "1" because the length of "" is "5". Therefore, as it is, the strongly connected component cannot be decomposed any more. Therefore, we try to increase the cycle by dividing some suitable production rules. Here, for example, as shown in FIG. 7, rule 6 is divided into rule 6-1 and rule 6-2.

FIG. 8 is a graphical representation of a strongly connected component composed of a set of new production rules obtained in this way. By decomposing the rule 6, a node 10 'is newly added. As a result, the closed circuit 3 '→ 1 passing through the node 3'
The length of 0 ′ → 5 ′ → 2 ′ → 4 ′ → 6 ′ → 3 ′ becomes “6”, and the period of this directed graph becomes “3”. Therefore, this strongly connected component is classified into a class I consisting of nodes 5 ', 6', 8 ', a class II consisting of nodes 2', 3 ', and a class III consisting of nodes 4', 7 ', 10'. can do. The relationship between these types and rules is shown in FIG.

FIG. 10 shows a strongly connected graph of cycle n with cycle m (where n ≧
An algorithm for classifying nodes while expanding the graph of m) is shown. In this algorithm, the classification number is sequentially assigned to each node in the graph while sequentially tracing the nodes starting from the node (S11).
-S14, S16-S20). Then, in this process, if different classification numbers are given to the same node, a formal node is added to the route to the node (S15).
Then, the number of added nodes for each node is totaled (S18), and the one with the smallest value is taken as the solution (S2).
1).

With this classification, the meta-knowledge storage unit 1 uses the production rules satisfying the transition condition between the classes in FIG. 9 as meta-knowledge.
Store in 4.

As a result of the above processing, the direction of global inference processing can be known by using the obtained graph as meta-knowledge.

In the process of inference, the inference unit 15 performs a global orientation of inference based on these meta-knowledges, and based on this,
The production rules of the rule storage unit 11 are sequentially executed while storing the intermediate result of the inference in the state storage unit 16.

For example, assuming that the initial state of inference is x = a, y = a and it is desired to know whether x = c, y = a can be derived as an inference result from this state, the strongly connected component decomposition unit 12
By using the meta-knowledge obtained by, it is possible to determine a global inference policy such that either rule 3 or rule 4 in FIG. 6 should be applied first.

Further, if the meta-knowledge obtained by the classification unit 13 is used, for example, when the current inference state belongs to class I in FIG. 9, there are only two applicable rules, rule 1 and rule 5, If the target state belongs to class III,
It can be seen that the number of times the rule is applied up to that is n = 3 × i + 2 (i = 0, 1, ...). Further, for example, the checking of the ending condition of backward inference is usually performed every time the rule is applied, but to which kind of node the node corresponding to the initial state of inference and the node corresponding to the state satisfying the ending condition belong respectively. If it is known, the end condition need only be checked every n times the rule is applied with the cycle being n.

When the classification unit 13 that performs the process of adding a node as described above is used, it is desirable to limit the number of nodes to be added to some extent. In other words, this process aims at giving a concise structure to the inference process as a whole by making a part of the expression redundant, but if the number of nodes to be added is too large, the size of the graph becomes large. Therefore, the redundancy is predominant and the efficiency of the whole reasoning is rather reduced. For example, when the number of added nodes exceeds the number of original nodes, this case is considered to correspond to this case. Therefore, it is preferable to select a combination in which the cycle can be increased by adding the number of nodes corresponding to several percent or less of the number of nodes in the original graph. For example, it is effective to extend a graph with a cycle n (usually n = 1) to each graph with a cycle m (n <m ≦ N) and select the one with the smallest number of added nodes.

[Effects of the Invention] As described above, according to the present invention, it is possible to determine the direction of global inference using the meta-knowledge obtained in the strongly connected component decomposition unit and the classification unit, which is wasteful. It is possible to prevent the rule from being executed and speed up the inference process.

[Brief description of drawings]

FIG. 1 is a block diagram of an inference processing device according to an embodiment of the present invention, FIG. 2 is a diagram showing the contents of a rule storage unit in the device, and FIG. 3 is a processing procedure of a strongly connected component decomposition unit in the device. FIG. 4, FIG. 4 and FIG. 5 are diagrams for explaining the processing of the decomposition unit, FIG. 6 is a diagram showing a strongly connected graph obtained by the decomposition unit, and FIGS. FIG. 10 is a flowchart for explaining the contents of processing in the classification unit, and FIG. 10 is a flowchart showing the processing procedure of the classification unit. 11 ... Rule storage unit, 12 ... Strongly connected component decomposition unit, 1
3 ... Classification unit, 14 ... Meta-knowledge storage unit, 15 ... Inference unit, 16 ... State storage unit.

Claims (1)

[Claims] 1. An inference processing apparatus for sequentially extracting applicable production rules to advance inference processing, and a rule storage unit for storing a plurality of production rules each including a pair of a condition section and an execution section. Setting means for setting nodes for all possible combinations of variables handled in the production rules stored in the rule storage section, and production rules satisfying the transition conditions between nodes set by the setting means By means of combining the nodes to obtain a directed graph matrix, decomposing means for decomposing the directed graph matrix obtained by this combining into strongly connected components, and variables for each strongly connected component obtained by this decomposing means. Condition calculation means for obtaining the condition corresponding to the strongly connected component from the value, and the production route satisfying the transition condition between the strongly connected components. Is used to obtain a directed graph matrix, and a strongly connected component decomposition unit having registration means for registering as meta-knowledge indicating which production rule can be applied next to which production rule, and the strongly connected component decomposition unit. Each strongly connected component has a period which is the greatest common divisor of the length of any closed circuit passing through any node in the strongly connected graph, and the period is increased by appropriately dividing the production rule. And a meta-knowledge obtained by the strongly connected component decomposition unit and the classification unit, which stores the production rules that satisfy the variable conditions and transition conditions of each period as meta-knowledge. Based on the meta-knowledge storage unit and the information indicating the production rule that can be applied next to the meta-knowledge stored in this meta-knowledge storage unit. Recognize the gap between the current state and the desired state, and then conduct a global inference to measure the application of production rules to approach the desired state. An inference processing device, comprising: an inference unit that extracts the next applicable production rule stored in the storage unit and executes inference processing.