JPH04346130A - Problem settlement device - Google Patents

Abstract

PURPOSE:To offer the device which advances subsequent inference by using a consequence node obtained by the inference so far although restriction conditions are altered as to a device which performs inference for solving a problem according to assumption wherein it can not be decided whether or not a result is true like scheduling and planning. CONSTITUTION:The device which has a node control part 14, an environment calculation part 16, and a contradiction control part 18 in a noncontradiction control part 12 adds restriction items to a restriction node, an assumption node, and the consequence node. The noncontradiction control part 12 is provided with a restriction condition calculation part 20 which performs arithmetic using the contents of the restriction item at the node indicated by the contents of the premise item at the generated consequence node and sets the result of the arithmetic indicating a link to the restriction node in the restriction item of the consequence node and a restriction condition update part 22 which inhibits inference using the consequence node where the contents showing the link to the altered restriction node are set in the restriction item.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of Industrial Application The present invention relates to a device that makes inferences to solve problems based on hypotheses that cannot necessarily be determined to be true, such as in scheduling and planning.

[0002] When finding a solution by selecting several hypotheses at the same time, contradictions often occur in a particular combination of hypotheses.

[0003] When this contradiction is discovered, the subset that caused the contradiction is identified from the current set of hypotheses.

[0004] In subsequent problem-solving operations, control is performed to prohibit the selection of combinations that would cause contradictions, thereby making it possible to efficiently solve problems while maintaining consistency.

[0005]

[Prior Art] Figure 8 shows this method (ATMS: Assu
emption-based truth
This figure shows a conventional problem-solving device to which the inference system (Nance System) is applied, and the inference unit 10 of the device makes inferences based on the passed nodes, and calculates a set of the conclusion of the inference and the node used to derive the conclusion. A return process is performed on the consistency management unit 12.

The consistency management section 12 is composed of a node management section 14, an environment calculation section 16, and a contradiction management section 18, and passes the node to the inference section 10 and calculates the result of the inference returned by the inference section 10 and the node. Processing to receive the is performed.

FIG. 9A shows the structure of a node (a unit of internal data) used in this conventional example, and as shown in the figure, each node is composed of a data term, an environment term, and a prerequisite term. has been done.

[0008] These nodes are classified into a constraint node in the figure (B), a hypothesis node in the figure (C), and a consequence node in the figure (D), and the data item of the constraint node indicates the constraint conditions of the problem. Contents indicating an empty set are set in the environment term and the prerequisite term, respectively.

[0009] The data item of the hypothesis node is set with the content of the hypothesis given from the outside, and the environment item and the premise item are set with content indicating a link to the node itself.

[0010]Furthermore, content indicating the conclusion of the inference is set in the data term of the consequence node, and content indicating the environment of the inference (indicating a link to the hypothesis node) is used as a premise of the inference in the environment term. Contents indicating a set of links to nodes are set in each prerequisite.

When constraint conditions and hypotheses are given from the outside to the node management unit 14 in FIG. 8, constraint nodes and hypothesis nodes are generated.

Further, when the inference unit 10 returns the conclusion of the inference and the node used to derive the conclusion, a consequence node is generated.

[0013] The environment term of the resultant node is set with content indicating the environment in which the inference was performed, and this content is obtained by the environment calculation unit 16.

Furthermore, when result nodes with the same data term contents occur, the node management unit 14 performs a process of merging them. Therefore, result nodes with multiple environment terms and antecedent terms are often arise.

[0015] These environments become a logical product set of hypotheses,
Having multiple of them means having a logical sum of logical products of hypotheses.

Therefore, logical calculation of the inference environment is performed in the environment calculation section 16, thereby simplifying the complex inference environment.

[0017] Furthermore, the environment term of a consequence node whose premise is a consequence node whose environment term has changed due to fusion is recalculated by the environment calculation unit 16, and this recalculation is successively propagated to other consequence nodes. .

If a contradiction is derived from some nodes, the inference unit 10 returns a conclusion indicating falsehood.

At this time, the environment calculating section 16 calculates the environment that leads to falsehood from the node used as the premise of the inference, and the node management section 14 generates a consequent node whose data item is false. (The consequent node it represents has no antecedents).

Furthermore, the environment leading to falsehood is removed from all other resultant nodes by the contradiction manager 18.

As a result, a consequence node with an empty environment term is not passed to the inference unit 10 (the contradiction management unit 18 performs this control), and therefore, inference based on a combination of inconsistent hypotheses is prevented. , consistency is maintained.

[0022]

[Problems to be Solved by the Invention] Conventionally, when the constraint conditions of a problem are changed, all of the resultant nodes that have been generated up to that point are discarded.

[0023] For this reason, the inference has to be restarted from the beginning, and therefore, it takes time to obtain a solution to the problem.

The present invention has been made in view of the above-mentioned conventional circumstances, and its purpose is to proceed with subsequent inferences using the resultant nodes obtained in previous inferences regardless of changes in constraint conditions. The objective is to provide a device that enables this.

[0025]

Means for Solving the Problems In order to achieve the above object, a problem solving apparatus according to the present invention is constructed as shown in FIG.

The device shown in the figure includes an inference unit 10 that performs inference using the passed nodes and returns the conclusion of the inference and the node used to derive the conclusion, and the inference unit 10 that returns the node to which the constraint has been added. and a consistency management unit 12 that receives the inference result and node returned by the inference unit 10.

The consistency management unit 12 includes a node management unit 14, an environment calculation unit 16, a contradiction management unit 18, a constraint calculation unit 20, and a constraint update unit 22.

Among these, the node management unit 14 generates a constraint node whose constraint term indicates a link with its own node and a hypothesis node whose constraint term is an empty set according to instructions input from the outside, A consequence node is generated in which the inference result and node returned by 10 are set in the data term and the premise term, respectively, and the constraint node is updated according to instructions given from the outside.

Furthermore, the environment calculation unit 16 calculates the contents of the environment term in the node indicated by the antecedent term of the generated resultant node, sets the result of the calculation to the environment term of the resultant node, and The content of the term is calculated and simplified, and the content of the environmental term in the consequent node whose premise term is set to the content indicating the consequent node where the content of the environmental term has changed due to fusion is recalculated.

Furthermore, the contradiction management unit 18 deletes the content indicating a false consequence from the environment clause of all the consequence nodes, which is indicated by the content of the environment clause in the consequence node set in the data item, and which leads to the conclusion. As a result, inference using a consequence node whose environment term is empty is prohibited.

[0031] Then, the constraint condition calculation unit 20 performs an operation using the contents of the constraint term in the node indicated by the contents of the antecedent term in the generated resultant node, and calculates the result of the operation indicating the link to the constraint node. The constraint update unit 22 deletes the content indicating the link to the changed constraint node from the constraint term of the consequence node set in the constraint term, and the corresponding content is set in the constraint term of the consequence node. The contents of the environment term are also deleted, and as a result, inference using a consequence node whose environment term is empty is prohibited.

[0032]

[Operation] In the present invention, the environment terms of the constraint node, hypothesis node, and consequence node are expanded and constraint terms are added to them.

When a consequent node is obtained by inference, another node indicated by the content of the antecedent term in the consequent node is searched, and an operation is performed using the content of the constraint term in the searched node.

By performing this operation, the constraint condition (link to the constraint node) on which the inference depends can be calculated, and the result of the operation is set to the constraint term of the same consequence node.

[0035] When a constraint node is updated (constraint conditions are changed), a consequence node whose constraint term has content indicating a link to the same constraint node is searched, and the searched consequence is The content is deleted from the constraint term of the node, and the content of the corresponding environment term is also deleted, and as a result, inference using a consequence node whose environment term is empty is prohibited.

That is, a consequence node that depends only on the changed constraint condition is not used for inference, and a consequence node that does not depend on the constraint condition is used as is for problem-solving inference thereafter.

[0037]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of the problem solving apparatus according to the present invention will be described below with reference to the drawings.

FIG. 1 shows the configuration of an embodiment, and this device includes an inference section 10 and a consistency management section 12.

Furthermore, the consistency management section 12 has a node management section 14, an environment management section 16, a contradiction management section 18, as well as a constraint calculation section 20 and a constraint update section 22.

FIG. 2A shows the node structure of this embodiment, and constraints are added to each node.

The inference unit 10 in FIG. 1 then performs inference using the passed nodes, and returns the conclusion of the inference and the node used to derive the conclusion.

Furthermore, the consistency management section 12 passes the node to which the constraint term has been added to the inference section 10, and receives the result of the inference and the node returned by the inference section 10.

[0043] Node management unit 1 of the consistency management unit 12
4, a constraint node whose constraint term indicates a link with the own node as shown in Figure 2(B) and a hypothesis node whose constraint term is an empty set as shown in Figure 2(C) are input from the outside. The result node is generated according to the instructions (including problem constraints and hypotheses), and the result node shown in FIG. , updates the constraint node according to instructions given from the outside (instructions to change constraint conditions).

Furthermore, the environment calculation unit 16 performs an AND operation on the contents of the environment term in the node indicated by the antecedent term of the generated consequence node, and uses the result of this AND operation (the environment in which the inference was performed) as the result of the conclusion. The content of the environment term in the consequence node is set in the environment term of the node, the content of the environment term in the consequence node is calculated and simplified, and the content indicating the consequence node where the content of the environment term has changed due to fusion is set in the antecedent term. Recalculate.

[0045] Then, the contradiction management unit 18 deletes the content that indicates a false consequence from the environment clause of all consequence nodes when the content indicating the false consequence is indicated by the content of the environment clause in the consequence node set in the data item. , As a result, inference using a consequence node whose environment term is empty is prohibited.

The constraint condition calculation unit 20 performs an operation using the content of the constraint term in the node indicated by the content of the antecedent term in the generated consequence node (the logical product of the constraint terms of all the nodes used as the premise of the inference). ), and sets the result of the operation indicating the link to the constraint node to the constraint term of the resultant node.

The constraint update unit 22 also deletes the content indicating a link to the changed constraint node from the constraint term of the consequence node whose constraint term is set, and
The contents of the corresponding environment term are also deleted, and as a result, inference using a consequence node whose environment term is empty is prohibited.

As a result, this consequence node is no longer passed to the inference unit 10 until it is newly inferred and the environment term is set again.

Note that if the changed constraint logically includes the constraint before the change, the process of changing the constraint with respect to the resultant node whose data item is false is not performed.

FIGS. 3 and 4 show examples of node generation when applied to a certain problem, and nodes N1 to N6 in FIG. 3 and nodes N7 to N12 in FIG. 4 are constraint nodes, respectively.
It is a hypothesis node.

These are generated by the node management unit 14 according to instructions (including problem constraints and hypotheses) given from the outside.

First, the consistency management unit 12
When nodes N6 and N7 are passed from node N1 to node Returns the pair with nodes N4 and N7.

[0053] The node management unit 14 corresponds to both consequences as shown in FIG.
The result nodes N13 and N14 are respectively generated, and the result of the inference and the node that is the premise of the inference are set in the data term and the premise term, respectively.

[0054] The environment calculation unit 16 also calculates the logical product (the environment in which the inference was made) of the environment terms in the nodes that appear in the antecedents of these new consequence nodes, and sets it as the environment term of the new consequence node. .

Furthermore, the constraint condition calculation unit 20 calculates the logical product of the constraint terms in the nodes that appear in the antecedents of the new consequence node, and displays the result of the calculation (the link to the constraint node used to derive the consequence). ) is set as the constraint term of the new consequence node.

Next, when the nodes N1 to N6 and N9 are passed to the inference unit 10, the inference unit 10 calculates the set of the conclusion of 5≦Z and the nodes N1 and N9 used to derive this, X≦5.
Returns the result and the set of nodes N6 and N9 used to derive it.

As a result, the resultant nodes N15 and N1 in FIG.
6 is generated by the consistency management unit 12.

When the next inference is made, FIG.
Consequence nodes N17 and N18 are generated.

[0059] Then, the inference unit 10 has nodes N13 and N1
8, the inference unit 10 discovers a contradiction and generates a consequent node N19 whose data item is false, as shown in FIG.

Further, when nodes N1 to N6 and N12 are passed to the inference section 10, the inference section 10 discovers a new contradiction.

In this case, the resultant node N19 is _N19
As shown in Figure 5, the result node _ is changed by simplifying the environment term.
Changes to _N19.

Furthermore, since a contradiction in the environment (N7) is confirmed, this is deleted from the environment terms of the resultant nodes N13 and N14.

As a result, both consequence nodes N13 and N14 are not passed to the inference unit 10 until a value is entered into the environment term again.

Thereafter, similar processing is performed, and resultant nodes N20 to N23 in FIG. 5 are generated.

Here, as shown in FIG. 6, the constraint nodes N2, N
When the constraint condition (content of the data term) indicated by 6 is changed, the resultant node that includes it in the constraint term (N16, N
18, N21, N22) are deleted (due to the deletion of the environment term).

However, the resultant node N19 representing false is:
The new constraint (_N6) is the same as the old constraint (N6)
) logically includes the constraint condition (N6
) is included in the environment term, it does not change.

Therefore, as shown in FIG. 7, the resultant node N1
3, N14, N15, N17, ___N19, N20, N
23 remains, and after changing the constraints mentioned above,
Those nodes N13, N14, N15, N17, ___N
19, N20, and N23 are used to solve the problem.

Therefore, according to this embodiment, the inference is not restarted from the beginning, and the problem can be solved efficiently.

[0069]

[Effects of the Invention] As explained above, according to the present invention, even if the constraint conditions of a problem are changed, many of the assets obtained up to that point can be retained, and this can be used to efficiently solve future problems. It becomes possible to do so.

[Brief explanation of drawings]

FIG. 1 is an explanatory diagram of the configuration of an embodiment.

FIG. 2 is an explanatory diagram of the data structure of a node in the embodiment.

FIG. 3 is an explanatory diagram of the contents of a constraint node used in the embodiment.

FIG. 4 is a content explanatory diagram of a hypothesis node used in the embodiment.

FIG. 5 is an explanatory diagram of the contents of a consequence node used in the embodiment.

FIG. 6 is an explanatory diagram of the contents of a constraint node changed in the embodiment.

FIG. 7 is a diagram illustrating the contents of a consequence node that remains after changing constraints in the embodiment.

FIG. 8 is a configuration explanatory diagram of a conventional example.

FIG. 9 is an explanatory diagram of the data structure of a node in a conventional example.

[Explanation of symbols]

10 Reasoning part 12 Consistency management department 14 Node management department 16 Environmental calculation department 18 Contradiction Management Department 20 Constraint condition calculation section 22 Constraint condition change section

Claims (1)

[Claims] [Claim 1] Perform inference using the passed node,
An inference unit (10) that returns the conclusion of the inference and the node used to derive the conclusion, passes the node to which the constraint has been added to the inference unit (10), and returns the conclusion of the inference that the inference unit (10) has returned. and a consistency management unit (12) that receives a node and a constraint node whose content indicates a link with its own node and a hypothesis where the constraint term is an empty set. Generate a node according to an instruction input from the outside, and generate a consequence node in which the conclusion of the inference returned by the inference unit (10) and the node are set in the data term and the premise term, respectively,
A node management unit (14) that updates the constraint node according to instructions given from the outside, and calculates the contents of the environment term in the node indicated by the antecedent term of the generated resultant node, and applies the result of the operation to the resultant node. The content of the environment term in the consequence node is calculated and simplified, and the content of the environment term in the consequence node set as the antecedent is rewritten with the content indicating the consequence node where the content of the environment term has changed due to fusion. an environment calculation unit (16) that performs the calculation, and content indicating a false consequence is indicated by the content of the environment term in the consequence node set in the data item, and content that leads to the conclusion is deleted from the environment terms of all consequence nodes; As a result, a contradiction management unit (18) prohibits inference using a consequent node whose environment term is empty, and an operation using the contents of a constraint term in a node indicated by the contents of antecedent terms in the generated consequent node. a constraint condition calculation unit (20) that sets the result of the operation indicating the link to the constraint node in the constraint term of the resultant node; and the content indicating the link to the changed constraint node is set in the constraint term. a constraint update unit (22) that prohibits inference using a consequence node;
A problem-solving device characterized by having.