JP3127634B2 - Constraint relaxation inference method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a constraint relaxation inference method. And a constraint relaxation inference method for performing inference.

[0002]

2. Description of the Related Art In a case of inference using a plurality of constraints, a solution satisfying all of these constraints may not be obtained. In such a case, it is necessary to relax the constraints by some method and find an appropriate solution. Conventionally, as a method of relaxing constraints, a method is defined in which the strength of the constraint that represents the strength of the constraint is defined for all the constraints, and the reasoning is advanced by excluding the application of the constraint from the weaker constraint. (Japanese Patent Application Laid-Open No. 3-67334)
issue). In this method, when a solution is not obtained after considering all the constraints, the application of the constraint having the weakest constraint strength is removed, that is, the inference is performed without considering the constraint as a constraint, and the solution is obtained again. When it is no longer possible, the application of the constraint condition having the next weaker constraint strength is removed, and so on.

[0003]

However, when the constraint strength is defined in the constraint condition as in the above-mentioned conventional method, the constraint condition is ordered from the viewpoint of strong and weak. There is a problem that when there are a plurality of constraint conditions, it is not possible to provide an appropriate constraint strength.

Further, in the method in which the inference is performed by sequentially removing the application from the constraints having the weaker constraint strength and using the constraint strength to proceed with the inference, an order is generated in the manner of removing the constraints, so that only the constraint having the stronger constraint is satisfied. Even when not performed, there is a problem that all the constraint conditions having weaker constraint strengths must be removed, and the accuracy of the solution becomes worse. further,
Since the solution is obtained without considering the removed constraint condition at all, the accuracy of the solution obtained by removing the constraint condition is reduced.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problem. In the case where a constraint condition is relaxed, when there are a plurality of constraint conditions having the same constraint strength, or when the constraint strengths are in an order. Also, an object of the present invention is to provide a constraint relaxation inference method that can obtain a solution with high accuracy by considering all constraint conditions.

[0006]

In order to achieve the above object, the present invention provides a method of expanding a node according to a generation rule to generate a new node, and a method of generating a new node based on a predetermined constraint condition. The degree of satisfying the constraint condition at the node is represented in a plurality of stages from the degree of completely satisfying the constraint condition to the degree of not satisfying the constraint condition at all, and storing the degree of satisfaction as a degree of satisfaction; A node that satisfies the reference satisfaction degree is selected as a node for developing, and when there is no node whose satisfaction degree satisfies the reference satisfaction degree, a value obtained by gradually relaxing the reference satisfaction degree is used as the reference satisfaction degree. And selecting a node whose degree of satisfaction satisfies the reference degree of satisfaction as a node for developing the target node.

[0007]

According to the present invention, new nodes are generated by expanding the nodes in accordance with the generation rules. Each of the newly generated nodes is defined with a sufficiency indicating a degree of satisfying a predetermined constraint condition. . The degree of satisfaction is an index indicating the degree to which the constraint is satisfied. The degree of satisfaction is expressed in a plurality of stages from the degree of completely satisfying the constraint to the degree of not satisfying the constraint at all. It is memorized.

The sufficiency of each node is compared with a reference sufficiency, and a node whose satisfaction satisfies the reference sufficiency is selected as a node for development. On the other hand, if there is no node whose satisfaction degree satisfies the reference satisfaction degree, the value of the reference satisfaction degree is gradually increased.
Relaxed to, by comparing the fullness with relaxed criteria satisfaction degree,
A node that satisfies the relaxed reference satisfaction degree is selected as a node to be developed. Then, by repeating the above, a solution that is a target node is inferred.

[0009]

Embodiments of the present invention will be described below. In this embodiment, the present invention is applied to a product allocating apparatus that allocates a large number of products to a plurality of coating lines. As shown in FIG. 1, the present embodiment includes a condition storage unit 10 storing constraint conditions and a rule storage unit 12 storing generation rules and evaluation rules. The condition storage unit 10 and the rule storage unit 12 are connected to an inference unit 14 that performs inference while expanding a node by searching to generate a search tree. The inference unit 14 is connected to a schedule display unit 16 that displays the inference result as a schedule.

An example of the operation of the product allocating apparatus is a case where eight kinds of product arcs having different colors, production amounts, and sizes are appropriately allocated to four automated coating lines A to D. This will be explained. Painting line A ~ D, product A ~
Table 1 and Table 2 show the characteristics of the laser.

[0011]

[Table 1]

[0012]

[Table 2]

The inference performed by the inference unit 14 is performed while expanding the nodes to generate a search tree, and the more nodes that go to the lower part of the search tree, the more the assignment is completed. That is, it becomes a solution.

The rule storage unit 12 stores rules for performing inference. Some of the generation rules and evaluation rules used in the above example are shown below.

Generation rules: (1) If there is still room in the painting line and there is a product with 30 or more remaining, the product is allocated to the painting line with room.

(2) If a product of red or purple color remains,
Since these products have limited available coating lines, they are assigned to applicable coating lines.

Evaluation rules: (1) If a product whose color is the same as the reference color of the coating line is allocated, the allocation amount × 1 is added to the score.

(2) If the difference between the maximum and the minimum of the sizes of the products allocated in the same coating line is within 30, 30 is added to the score.

The condition storage unit 10 stores the following constraints. Restrictions: (1) If there is a difference of 60 or more between the maximum and the minimum of the sizes of the products allocated in the same coating line, the satisfaction level is set to 0.2.

(2) If a product whose color is neither the reference color nor the corresponding color is assigned to the coating line, the degree of satisfaction =
Set to 0.

(3) Since the coating lines A and D are small-size coating lines, the sizes of the coating lines A and D are 40 to 6
If the product of 0 is allocated, the satisfaction level is set to 0.8.

(4) Since the coating lines A and D are small-size coating lines, if the products having a size of 70 or more are allocated to the coating lines A and D, the satisfaction level is set to 0.6.

The inference unit 14 performs the inference by relaxing the constraints and backtracking the search tree. This inference will be described with reference to the flowcharts of FIGS.

In the initial state, no products are assigned to all the coating lines. Therefore, step 100
In, the node in the initial state is set as the selected node, and the reference satisfaction degree is set to 1. Satisfaction = 1 means that the constraint condition is 100%
Satisfaction, that is, complete satisfaction, where sufficiency = 0 means no constraint is satisfied, and 1>satisfaction> 0 does not satisfy the constraint at all. Means a state between degrees.

In step 102, the generation rules read from the rule storage unit 12 are applied to the selected nodes in the initial state to develop the nodes, and a plurality of nodes in which the product is allocated to any one painting line are created. In FIG. 2, the generation rule (1) is applied, and 30 or more products C, D, O, and K are assigned to the painting line A having the standard production amount of 50, and 50 products C are assigned to the painting line A. An example is shown in which four nodes are created: a node where 30 products are allocated to painting line A. A node where 40 products are allocated to painting line A. .

In the next step 104, a constraint condition is applied to the created nodes (1) to (4) to determine a "satisfaction degree" representing the satisfaction degree of each node. Note that the initial value of the sufficiency is set to 1, and if no constraint condition is satisfied, the sufficiency of the node is set to the initial value, that is, 1. In the example of FIG. 2, the constraint (3) matches the node and the degree of satisfaction of the node is 0.8, and the constraint (2) and the constraint (3) match and the degree of satisfaction of the node is minimum. It has a value of 0.
Since the node does not satisfy any of the constraints, the degree of satisfaction is 1
Remains.

In step 106, the above evaluation rules are applied to nodes whose satisfaction is determined and the satisfaction is not 0, and a score is given to each node. In the example of FIG.
Since the blue product C, D, and O having the same color as the reference color of the line are assigned to the blue coating line A, the evaluation rule (1) conforms to the nodal point, and the assignment amount × 1 By calculation, each node has 50 points, 3
0 points and 40 points (evaluation points) are given. No score is given to the node since the degree of satisfaction is 0.

In step 108, it is determined whether or not there is a node whose satisfaction degree satisfies the reference satisfaction degree. In the example of FIG. 2, there is a node satisfying the reference satisfaction level = 1, so the process proceeds to step 110 through step 120, and the node having the highest score among the evaluated nodes is selected as the selected node. In the above example, the node with the highest score is selected as the selected node. In the above example, since the reference satisfaction level = 1, the selected node is the node with the highest score among the nodes satisfying the satisfaction level = 1. If there are a plurality of nodes having the same score, all the nodes having the same score are selected.

Next, returning to step 102, the next branch is extended by applying the above generation rule again to the selected node and expanding the node. In this example, the generation rule (2) conforms, and as shown in FIG. 3, a node where 30 product mosquitoes are allocated to the painting line C. A node where 30 product mosquitoes are allocated to the painting line D. Three nodes of the two nodes are obtained. The three nodes obtained,
In step 104, a constraint condition is applied in the same manner as described above. In the above example, the satisfaction degree is 0.8 because the constraint condition (3) matches the node. Since there is no constraint satisfying the node, the degree of satisfaction remains 1. Then, in step 106, the evaluation rule is applied in the same manner as described above, and a score is given. In this case, the evaluation rule (1) is satisfied, and the scores of the nodes are 50 points, 50 points, and 52 points, respectively.
Points.

Then, step 108, step 110
As described in the above, a node whose satisfaction degree satisfies the reference satisfaction degree, that is, a node having the highest score among the nodes having the satisfaction degree 1 is selected as a selection node, and the process returns to step 102 again to apply the generation rule and to select a branch. Prolong the search.

It is assumed that a search tree as shown in FIG. 4 is obtained by repeating steps 102 to 110.

In FIG. 4, two nodes b and c generated by applying the generation rule to the thick line node a below the search tree have both the sufficiency 0 due to the constraint condition. That is, it is impossible to extend the branch any further and proceed with the search. Therefore, the process proceeds from step 108 to step 112 to perform backtracking. When the process first proceeds to step 112, the first backtrack is performed and the number of backtracks is equal to or less than the limit value (for example, 2).
Proceed to 16 to perform backtracking and search for another branch.
That is, backtracking is performed as shown by the dashed arrow in FIG. 4, and the node X is selected as the selected node.

FIG. 10 shows the details of the backtracking in step 116 of FIG. 9. In step 200, the currently selected node is set to node X. In step 210, there is a node that is arranged side by side with node X. It is determined whether or not there exists a node satisfying the standard satisfaction degree and having an evaluation point lower than the evaluation point of the node X. If the determination in step 210 is negative, in step 220, the parent node of the node X (the node immediately before development) is set to the node X, and the process proceeds to step 21.
The determination of 0 is performed again. On the other hand, if the determination in step 210 is affirmative, in step 230, the nodes that are the side-by-side nodes of the nodes X and satisfy the reference satisfaction degree and have the evaluation points next to the node X are set as new selected nodes. By this backtrack, a node X that satisfies the reference satisfaction degree below the search tree and has an evaluation point next to the node a is selected as the selected node.

However, the search may not be able to proceed even if the first backtrack is performed as described above. In the example of FIG. 5, the node d generated before the node X due to the expansion of the node X becomes the satisfaction degree 0, and the search cannot be advanced. Also in this case, backtracking is performed again as indicated by a broken arrow, and the node Y is selected as a selected node.

However, as shown in FIG. 6, the sufficiency of the point beyond the node Y may become 0, and the search may not be able to proceed. In such a case, backtracking may be performed and another search may be performed again, but the solution may not be reached forever. Therefore, the number of backtracks is limited as described above, for example, the number of backtracks is limited to two. Then, the search cannot be completely advanced in the example of FIG. Therefore, the process proceeds from step 112 to step 114 to relax the restriction condition. That is, the value of the reference satisfaction degree is relaxed, the constraint of the reference satisfaction degree is set to 0.9 ≦ reference satisfaction degree <1, and in step 118, a node satisfying this condition is searched. When there are a plurality of nodes, 1) a node having a high degree of satisfaction 2) a node below the search tree 3) a node having a high evaluation score is preferentially selected. If there are a plurality of nodes having completely the same condition, all of the plurality of nodes may be selected. In the example of FIG. 6, the node Z having the degree of satisfaction of 0.9 is selected as the selected node. Thereafter, the process proceeds to step 108, and the above steps 102 to 110 are repeated to search in the range of 0.9 ≦ satisfaction <1.

The search may be continued as it is, and the situation shown in FIG. In this case, the satisfaction degree of the generated node W does not satisfy the condition of 0.9 ≦ satisfaction degree <1. In addition, there is no node that satisfies the condition of 0.9 ≦ satisfaction degree <1 among the remaining nodes even when backtracking is attempted. In this case, the constraint condition is relaxed again, and the reference satisfaction degree is set to 0.8 ≦ the reference satisfaction degree <0.9, and the search is advanced in this range. Then, the above steps are repeated until it is determined in step 120 that the allocation is completed.
The node V as a solution which is the target node shown in FIG.
6 is displayed.

Although the present invention has been described above with reference to an example in which the present invention is applied to a product allocating apparatus, the present invention is not limited to this, and is applicable to artificial intelligence and various systems utilizing artificial intelligence. You can do it.

[0038]

As described above, according to the present invention, by defining the degree of sufficiency in a constraint, a constraint that must be strictly obeyed from a constraint that must be absolutely observable is preferably set. Can be expressed flexibly, so that even if there are multiple constraints with similar strength, it is possible to give an appropriate degree of satisfaction, and always satisfy all constraints without removing constraints. Since a node whose degree satisfies the reference satisfaction degree is selected, an effect that a highly accurate solution can be obtained is obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram of a product allocating apparatus according to an embodiment.

FIG. 2 is a diagram showing a state in which a selected node in an initial state is expanded to generate a node ~.

FIG. 3 is a diagram showing a state in which a selected node is expanded to generate a node ~.

FIG. 4 is a diagram showing a state in which backtracking is performed from a node a to a node X.

FIG. 5 is a diagram showing a state in which backtracking is performed from a node X to a node Y.

FIG. 6 is a diagram showing a state where backtracking is performed from a node Y to a node Z;

FIG. 7 is a diagram illustrating a state where a node Z is obtained by expanding a node Z;

FIG. 8 is a diagram showing a state where a target node V is obtained.

FIG. 9 is a flowchart showing an inference routine in an inference unit.

FIG. 10 is a flowchart showing details of step 116 in FIG. 9;

[Explanation of symbols]

 Reference Signs List 10 Condition storage unit 12 Rule storage unit 14 Inference unit 16 Schedule display unit

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-4-186434 (JP, A) JP-A-4-125734 (JP, A) JP-A-4-118720 (JP, A) JP-A-1- 177163 (JP, A) JP-A-5-346856 (JP, A) IEICE Technical Report, Vol. 90, NO. 390 (AI 90-72 to 75), published by the Institute of Electronics, Information and Communication Engineers (January 1991), 17-23 (Patent Office CSDB Document No .: CCNT199900975003) (58) Fields investigated (Int. Cl. ⁷ , DB name) G06F 9/44 G06F 17/60 G06F 19/00 B23Q 41/08 G05B 13/02 JICST file (JOIS) CSDB (Japan Patent Office)

Claims (1)

(57) [Claims] A step of expanding a node in accordance with a generation rule to generate a new node; A step of dividing the degree of complete satisfaction from the degree of not satisfying the constraint condition at all into a plurality of stages and storing the degree of satisfaction as a degree of satisfaction; and as a node for developing a node at which the degree of satisfaction satisfies the reference degree of satisfaction. And selecting the reference satisfaction degree when there is no node satisfying the reference satisfaction degree.
A step of selecting a node satisfying the reference satisfaction degree as a reference satisfaction degree using the value relaxed in a stepwise manner as a node for developing a node satisfying the reference satisfaction degree.