JPH0962511A - Search system for state space - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the required storage capacity by providing the high-quality solution on the early stage of search by deciding the transient destination of current state by evaluating the adjacent state of state space corresponding to the sum of a state change cost function value and the evaluation function value of the adjacent state. SOLUTION: The state to be the start point of state space search is initially set to the current state (concerned state) and its evaluation function value is calculated (1). Afterwards, the states transitable from the current state are extracted and their evaluation function values are respectively calculated (4). When any state requiring the change of its evaluation function value exists among the extracted states, that evaluation function value is changed (5). Further, the adjacent extracted state is evaluated corresponding to the sum between the total sum of values of elements of state change cost function value vectors corresponding to the elements having difference between that state and the current state and the evaluation function value of the adjacent state and the state having the smallest evaluation function value is selected as the best adjacent state (the next concerned state) (6). Thus, the adjacent state is evaluated so that the transient destination of current state can be decided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a search method for a state space, which is a set of states representing each combination, and particularly optimization for trying to find the best combination among various combinations. It relates to a method of solving a problem.

[0002]

2. Description of the Related Art FIG. 8 shows, for example, 1991 patent application No. 2688.
52 is a flow chart shown in Japanese Patent No. 52, "Search Method of State Space". In FIG. 8, 31 is a block for initializing the state that is the starting point of the state space search to the “attention state”, 32 is a block for calculating the evaluation function value of the “attention state”, and 33 is for counting the number of searches. A block for initializing the counter, 34 is a block for extracting transitionable states from the “state of interest” and calculating their evaluation function values, and 35 is a block for evaluating the evaluation function values among the states extracted in block 34. If there is an item that needs to be changed, a block for changing the evaluation function value, 36 finds the state having the smallest evaluation function value among the states extracted in block 34, and sets this state as the "next state of interest." It is a block to be selected as ".

Further, 37 is a block for judging whether or not the "next state of interest" has a desired property, 38
Is a block that determines whether or not the "next state of interest" is more desirable than the stored best solution, and if it is desirable, saves this as the best solution. 39 indicates that the "state of interest" is a minimal solution (" A block that determines whether or not there is no evaluation function value smaller than the "state of interest" among the states that can be transitioned from the "state of interest", 40 is the evaluation function value of the "state of interest" It is a block that increases to become larger than the evaluation function value of the "next attention state".

Further, reference numeral 41 is a block for judging whether or not the same state as the "attention state" exists in the state where the evaluation function value has been saved in the past, and 42 indicates the "attention state". A block that changes the evaluation function value of the same state in which the evaluation function value has been increased in the past to the evaluation function value of the "state of interest", 43 is the evaluation function value of the "state of interest" that has been increased in the past Is stored with the evaluation function value, and 44 is a predetermined number of searches LMAX.
Is a block for determining whether or not the value exceeds the value, 45 is a block for switching the "state of interest" and its evaluation function value to the "next state of interest" and its evaluation function value, and 46 is a block for incrementing the counter of the number of searches by 1. .

Next, the operation will be described. First, in block 31, a state that is a starting point for searching the state space is set as a “state of interest”, and in block 32, its evaluation function value is calculated. Next, block 3
In 4, 35 and 36, the state having the smallest evaluation function value among the states that can transition from the “state of interest” is selected as the “next state of interest”. Next, at block 37,
It is determined whether or not the “next attention state” is a solution having the desired property. If the solution has the desired property, the process ends, otherwise the process proceeds to block 38. .

In block 38, it is judged whether or not the "next state of interest" is the best state searched in the past, and if it is the best state, this is stored as the best solution. Next, in block 39, it is judged whether or not the "state of interest" is a minimal solution, and if it is a minimal solution,
Processing transfers to block 40, otherwise processing transfers to block 44.

At block 40, the evaluation function value of the "minimum state of interest" is increased to be larger than the evaluation function value of the "next attention state". Then, in blocks 41, 42 and 43, the increased "state of interest" evaluation function value is saved with that state.

Next, in block 44, it is determined whether or not the number of searches exceeds a predetermined number. If it exceeds, the process ends, and if not, the process moves to block 45. The block 45 prepares to move to the next search stage by changing the "state of interest" and its evaluation function value to the "next state of interest" and its evaluation function value. Is incremented by 1 and then the search is continued by returning to block 34.

[0009]

In the conventional search method of the state space, the search is basically performed based on the evaluation function value uniquely determined from the state. Regarding the escape method from the minimal solution, It is realized by deteriorating the evaluation function value only when the state is found to be a minimal solution. However, since the state where the evaluation function value is relatively small is gathered around the minimum solution state, and it is almost the case that a valley-shaped evaluation function value group with the minimum solution as the valley bottom is formed, Until the state moves away from the solution, it is necessary to repeat the process of deteriorating the evaluation function value by sequentially determining the states where the evaluation function values around these minimal solutions have relatively small evaluation function values as the minimal solutions. Therefore, the processing efficiency is low, and there are problems that many unnecessary minimal solutions must be saved.

The present invention has been made in order to solve the above problems, and a state space satisfying a small required memory capacity and a highly efficient search (that a good solution can be found from an early stage of the search). The purpose is to provide a search method of.

[0011]

A state space search method according to the present invention is a state space search method that evaluates a plurality of neighboring states that can transit from the current state and determines a transition destination. Is evaluated by the sum of the difference between the current state and the adjacent state and the state change cost function value determined by the history of past transitions of the present state, and the evaluation function value of the adjacent state,
This is to determine the transition destination of the current state.

In addition, in a state space search method that evaluates a plurality of adjacent states that can transit from the current state and determines a transition destination, a state change cost function value vector having the same length as the state vector representing the state is calculated. Prepare and update the value of each element of the state change cost function value vector by increasing the value of the element of the state change cost function value vector corresponding to the element of the current state vector that has changed with the state transition to the current state. Then, the value of each element of the state change cost function value vector is updated by decreasing or maintaining the value of the element of the state change cost function value vector corresponding to the element of the current state vector that has not changed. Go to the next state, and add the sum of the values of the elements of the state change cost function value vector corresponding to the elements that differ between the current state and the next state, and the next state Was evaluated by the sum of the evaluation function value, it is what determines the transition destination of the current state.

In addition, in a state space search method that evaluates a plurality of adjacent states that can transit from the current state and determines a transition destination, a state change cost function value vector having the same length as the state vector representing the state is calculated. Prepare and compare the sum of the state change cost function value determined by the difference between the current state and the next state and the history of past transitions of the current state and the evaluation function value of the next state with the evaluation function value of the current state By determining whether the current state is a minimal solution, and if the current state is a minimal solution, the state change cost function value vector corresponding to the element of the current state vector that changes with the state transition to the current state. Value of the state change cost function value vector that increases or decreases the value of the element of the state change cost function value vector corresponding to the element of the current state vector that has not changed, When the value of each element is updated and the current state is not the minimum solution, the value of each element of the state change cost function value vector is decreased or maintained, and the value of each element of the state change cost function value vector is changed. The next state is updated and the sum of the values of the elements of the state change cost function value vector corresponding to the element that has a difference between the current state and the next state and the evaluation function value of the next state The sum is evaluated to determine the transition destination of the current state.

Further, a state vector representing a state is divided into several element vectors, a state change cost function value vector having a length equal to the number of the element vectors is prepared, and changes with a state transition to the current state. The value of each element of the state change cost function value vector is updated by increasing the value of the element of the state change cost function value vector corresponding to the element of the present state vector The value of each element of the state change cost function value vector is updated by decreasing or maintaining the value of the element of the state change cost function value vector corresponding to the element, and the adjacent state is changed to the current state and the adjacent state. Evaluated by the sum of the sum of the values of the elements of the state change cost function value vector corresponding to the element vector that differs from the state and the evaluation function value of the adjacent state. And is what determines the transition destination of the current state.

In addition, in a search method of a state space in which a plurality of neighboring states that can transit from the current state are evaluated and a transition destination is determined, the state vector representing the state is divided into several element vectors, and the element vectors A state change cost function value vector with a length equal to the number of is prepared, and the state change cost function value determined by the difference between the current state and the next state and the history of past transitions of the current state, and the evaluation of the next state By comparing the sum with the function value with the evaluation function value of the current state, it is determined whether the current state is a minimal solution, and if the current state is a minimal solution, change with the state transition to the current state. The element of the state change cost function value vector corresponding to the element of the current state vector that has not changed by increasing the value of the element of the state change cost function value vector corresponding to the element of the present state vector Decrease or maintain the value, update the value of each element of the state change cost function value vector, and if the current state is not a minimal solution, decrease or maintain the value of each element of the state change cost function value vector Then, update the value of each element of the state change cost function value vector,
The adjacent state is evaluated by the sum of the values of the elements of the state change cost function value vector corresponding to the elements having a difference between the current state and the adjacent state, and the sum of the evaluation function values of the adjacent states,
This is to determine the transition destination of the current state.

Further, when the adjacent state is evaluated and the transition destination is determined, it is determined whether or not there is the adjacent state whose evaluation function value is smaller than the current state, and the difference between the current state and the adjacent state and the current state. The sum of the state change cost function value determined by the history of past transitions of the state and the evaluation function value of the adjacent state is determined by comparing it with the evaluation function value of the current state. If a small adjacent state exists, it is selected as the transition destination, and if it does not exist, the evaluation function value is increased for the minimization problem and the evaluation function value is increased for the maximization problem. By degrading the evaluation function value of the current state such as by decreasing it, a neighboring state having a smaller evaluation function value than the current state appears, and this is selected as the transition destination.

Further, when the adjacent state is evaluated and the transition destination is determined, it is determined whether or not there is the adjacent state whose evaluation function value is smaller than the current state, and the difference between the current state and the adjacent state and the current state. The sum of the state change cost function value determined by the history of past transitions of the state and the evaluation function value of the adjacent state is determined by comparing it with the evaluation function value of the current state. If there is a small adjacent state, the state with the smallest evaluation function value is selected as the transition destination, and if it does not exist, the evaluation function value is increased for the minimization problem and the maximum For the optimization problem, by degrading the evaluation function value of the current state such as decreasing the evaluation function value, a neighboring state with a smaller evaluation function value than the current state appears, and this is selected as the transition destination. To do.

[0018]

BEST MODE FOR CARRYING OUT THE INVENTION

Embodiment 1. Hereinafter, Embodiment 1 of the present invention will be described with reference to FIG. FIG. 1 is a flowchart showing a state space search method according to the first embodiment of the present invention. FIG.
In Fig. 1, 1 is a block for initializing a state that is the starting point of the state space search to the current state (state of interest) and calculating its evaluation function value. 2 is a state change cost function value having the same number of elements as the current state. A block that prepares a vector and clears all its elements to 0, a block 3 that initializes a counter for counting the number of searches, and a block 4 that extracts a transitionable state from the current state and evaluates those functions. A block for calculating a value, 5 is a block for changing the evaluation function value if there is a need for changing the evaluation function value among the states extracted in the block 4, and 6 is the next block extracted in the block 4. The state of is evaluated by the sum of the sum of the values of the elements of the state change cost function value vector corresponding to the element that has a difference between that state and the current state, and the sum of the evaluation function value of the adjacent state. A block selecting as the best neighboring state (next focused state) a small state the least evaluation function value at medium.

Further, 7 is a block for judging whether or not the best adjacent state has a desired property, and 8 is for judging whether or not the best adjacent state is more preferable than the best solution held therein. If desired, this block stores this as the best solution, 9 is the minimum solution of the current state (state change cost function value determined by the difference between the current state and the best adjacent state and the past history of the current state, and the best adjacent state). Of the evaluation function value of the present state is smaller than the sum of the evaluation function value of the present state) and the evaluation function value of the present state. Make the evaluation function value of the current state worse than the evaluation function value of the best adjacent state rather than the sum of the state change cost function value determined by the past history of the current state and the evaluation function value of the best adjacent state. It is a lock.

Further, a block 11 stores the present state and its deteriorated evaluation function value as a state in which the evaluation function value has been increased in the past and its deteriorated evaluation function value (blocks 41 and 42 in FIG. 8). , And 43), 12 is a block that sequentially focuses on the corresponding elements of the current state and the best adjacent state, 13 is a block that determines whether the value of the focused element is equal in the current state and the best adjacent state, and 14 is If the element of the state change cost function value vector corresponding to the focused element is not 0, the value is decreased from that block, and 15 is the block that increases the value of the corresponding state change cost function value vector of the focused element. 1
7 is a block that determines whether the number of searches exceeds a predetermined number LMAX, 18 is a block that switches the current state and its evaluation function value to the best adjacent state and its evaluation function value, and 19 is an increment counter of the number of searches It is a block to be made.

Next, the operation will be described. First, in block 1, the state that is the starting point for searching the state space is set as the current state, and the evaluation function value thereof is calculated. Next, in block 2, all elements of the state change cost function value vector are cleared to 0. Next, in blocks 4, 5 and 6, among the neighboring states that can transit from the current state, the state change cost function value determined by the difference between the current state and the neighboring state and the past history of the current state, and the neighboring state. The state having the best evaluation value is selected as the best adjacent state when evaluated by the sum of the state and the evaluation function value.
Next, in block 7, it is judged whether or not the best adjacent state is a solution having the required property, and if it is the solution having the required property, the processing is terminated, and otherwise, to block 8. Transfer processing.

At block 8, a comparison is made between the best neighbor state and the stored best solution, and if the best neighbor state is more desirable than the stored best solution, then it is stored as the best solution. Next, in block 9, it is determined whether or not the current state is the minimum solution. If the current state is the minimum solution, the process shifts to block 10, and otherwise the process shifts to block 12. .

In block 10, the evaluation function value of the current state, which is the minimum solution, is deteriorated to a value worse than the evaluation function value of the best adjacent state. Next, in block 11, the degraded evaluation value of the current state is saved with that state. (It is determined whether or not the same state as the current state exists in the saved states where the evaluation function value has been increased in the past, and if there is, the evaluation function value is used as the evaluation function for the current state. Change to a value, and if it does not exist, the current state is assumed to be the state where the evaluation function value was deteriorated in the past,
Save with the evaluation function value. )

At block 12, the corresponding elements of the current state and the best adjacent state are sequentially focused on, and at block 13, it is determined whether the value of the focused element is equal in the current state and the best adjacent state. If the element values are equal, the process proceeds to block 14, and if the element of the state change cost function value vector corresponding to the focused element is not 0, the value is decreased. If the element values are not equal, the process proceeds to block 15, and the value of the element of the corresponding state change cost function value vector of the focused element is increased.

Next, in block 17, it is judged whether or not the number of searches exceeds a predetermined number. If it exceeds, the process ends. If not, the process shifts to block 18. In block 18, by changing the current state and its evaluation function value to the best neighbor state and its evaluation function value,
Preparations are made to move to the next search stage, and after the counter of the number of searches is incremented by 1 in the following block 19, the search is continued by returning to block 4.

Embodiment 2. Next, a second embodiment of the present invention will be described with reference to FIGS. 2, 3, 4, 5, 6, and 7. FIG. 2 is an explanatory diagram of an application result example to a scheduling problem according to the related art, and FIG. 3 is an explanation of an application result example (1) to a scheduling problem of a state space search method according to the second embodiment of the present invention. It is a figure. (Transition of Current State) FIG. 4 shows an example of the result of applying the state space search method according to the second embodiment of the present invention to a scheduling problem (2).
FIG. (Next to Current State 0) FIG. 5 is an explanatory diagram of an application result example (3) to the scheduling problem of the search method of the state space according to the second embodiment of the present invention. (Next to Current State 1) FIG. 6 is an explanatory diagram of an application result example (4) to the scheduling problem of the state space search method according to the second embodiment of the present invention. (Next to Current State 2) FIG. 7 is an explanatory diagram of an application result example (5) to the scheduling problem of the state space search method according to the second embodiment of the present invention. (Currently next to State 3)

Consider a scheduling problem that optimizes the start dates of the four tasks given so that they do not fall outside the 5-day window. Work 1, work 2 and work 3 are works for one day, and work 4 is work for two consecutive days. Although the desired start date for work 1 is the first day, the start date can be moved. However, 2 penalties will be incurred each time the start date of work 1 deviates from the desired start date by one day. Work 2
Although the desired start date is the second day, the start date can be moved. However, 2 penalties will be incurred each time the start date of work 2 deviates from the desired start date by one day. Work 3 has a fixed start date on the 2nd day, and work 4 has a start date of 1
It is fixed on the day. If work 1 and work 2 overlap on the same day, a penalty of 1 will occur and work 1 and work 3 will occur.
If the two overlap on the same day, a penalty of 10 points will be incurred. In addition, even when work 1 and work 4 overlap on the same day, a penalty of 10 points is generated.

In this scheduling problem, the state is defined with the start date of work 1 and the start date of work 2 as elements. The start date of either work 1 or work 2 is 1
The state moved to the next day is set as the adjacent state. Furthermore, the evaluation function value of a state is the sum of the penalties in that state.

FIG. 2 shows the transition of the current state when the search is performed according to FIG. 2, which is a flow chart showing the conventional state space search method, up to the fourth search. The current state 0 is an initial state (a state in which the start date of each work is the desired start date). Moving the start date of work 1 to the second day causes work 1 and work 3 to be executed on the same day, resulting in a penalty of 10 points. On the other hand, moving the start date of work 2 only adds a penalty of 2 points, so in the current state 1, the current state 2, and the current state 3, only the start date of the work 2 is moved. In the current state 4, the start date of work 1 moves for the first time. In the current state 4, the optimal schedule has not yet been obtained.

On the other hand, FIG. 3 is a flowchart showing a state space search method according to the first embodiment of the present invention.
The transition of the current state and the transition of the state change cost function value vector when the search is performed according to are shown up to the fourth search. The state change cost function value vector has two elements, work 1 and work 2, and the current state and the immediately preceding current state are compared, and the element of the state change cost function value vector corresponding to the work having a different start date is the penalty 20. If the value of the element is not 0, the penalty of 20 points is subtracted from the element of the state change cost function value vector corresponding to the same work on the start date, and the value of the element is set to 0.

FIGS. 4, 5, 6 and 7 respectively show all states next to current state 0, current state 1, current state 2 and current state 3 of FIG. 4, FIG. 5, FIG. 67, and FIG. 7, Y is the sum X of the element values of the state change cost function value vector corresponding to the element having a difference between the adjacent state and the current state, and the adjacent state. Is the sum of the evaluation function value of. Of all the states next to the current state,
The value with the smallest Y is the best neighbor state. In FIG.
A state in which the state 2 adjacent to the current state 0 is selected as the best adjacent state, that is, the current state 1 is shown. In FIG. 5, the state 1 adjacent to the current state 1 is selected as the best adjacent state, that is, the current state 2. FIG. 6 shows a state in which the state 3 adjacent to the current state 2 is selected as the best adjacent state, that is, the current state 3. In FIG. 7, the state 2 next to the current state 3
Is selected as the best adjacent state, that is, the current state 4. In addition, the evaluation function value in the state where it is found to be the minimum solution is added with 100 points of penalty.

FIG. 3 shows the transition of the current state searched as described above. By adding and evaluating the sum of the values of the elements of the state change cost function value vector corresponding to the elements that have a difference between the adjacent state and the current state, it is possible to suppress continuous movement of the same work start date. It
For this reason, it is possible to avoid the concentration of the search as the movement of the start date of the work 2 as in the search by the conventional technique, and the optimum schedule has already been obtained in the current state 4.

The embodiment of the invention described above is
Although the description has been made for the combination minimization problem, the present invention can be applied to the combination maximization problem as well.

[0034]

As described above, according to the present invention, since a search is performed in a direction avoiding a similar state, a good solution can be found at an early stage of the search, and the number of searches needs to be reduced. This has the effect of reducing the storage capacity. Also,
Under the condition that the problem space becomes a strongly connected graph, a sufficient search has an effect of reaching the optimum solution.

[Brief description of drawings]

FIG. 1 is a flowchart showing a state space search method according to a first embodiment of the present invention.

FIG. 2 is an explanatory diagram of an application result example to a scheduling problem by a conventional state space search method.

FIG. 3 is an explanatory diagram of an application result example (1) of the state space search method according to the second embodiment of the present invention to a scheduling problem. (Current state transition)

FIG. 4 is an explanatory diagram of an application result example (2) to the scheduling problem of the state space search method according to the second embodiment of the present invention. (The state next to the current state 0)

FIG. 5 is an explanatory diagram of an application result example (3) to the scheduling problem of the state space search method according to the second embodiment of the present invention. (Currently next to State 1)

FIG. 6 is an explanatory diagram of an application result example (4) to the scheduling problem of the search method of the state space according to the second embodiment of the present invention. (The state next to the current state 2)

FIG. 7 is an explanatory diagram of an application result example (5) to the scheduling problem of the search method of the state space according to the second embodiment of the present invention. (Currently next to State 3)

FIG. 8 is a flowchart showing a conventional state space search method.

Front page continuation (72) Inventor Koichi Hirose 2-3-3 Marunouchi, Chiyoda-ku, Tokyo Sanryo Electric Co., Ltd. (72) Inventor Takahiro Ishiguro 2-3-2 Marunouchi, Chiyoda-ku, Tokyo Sanryo Denki Incorporated (72) Inventor Nobuhiko Itaya 2-3-3, Marunouchi, Chiyoda-ku, Tokyo Sanryo Electric Co., Ltd.

Claims (7)

[Claims] 1. A method for searching a state space, which evaluates a plurality of neighboring states that can transit from the current state and determines a transition destination, wherein the neighboring state is defined as a difference between the current state and the neighboring state and the current state. Evaluated by the sum of the state change cost function value determined by the history of past transitions and the evaluation function value of the adjacent state,
A state space search method characterized by determining the transition destination of the current state. 2. A state change cost function value vector having the same length as a state vector representing a state in a search method of a state space that evaluates a plurality of adjacent states that can transit from the current state and determines a transition destination. Prepare and update the value of each element of the state change cost function value vector by increasing the value of the element of the state change cost function value vector corresponding to the element of the current state vector that has changed with the state transition to the current state. Then, the value of each element of the state change cost function value vector is updated by decreasing or maintaining the value of the element of the state change cost function value vector corresponding to the element of the current state vector that has not changed. Go to the next state and evaluate the next state by summing the values of the elements of the state change cost function value vector corresponding to the elements that differ between the current state and the next state. A state space search method characterized by determining the transition destination of the current state by evaluating the sum with the function value. 3. A state change cost function value vector having the same length as a state vector representing a state in a search method of a state space for evaluating a plurality of adjacent states that can transit from the current state and determining a transition destination. Prepare and compare the sum of the state change cost function value determined by the difference between the current state and the next state and the history of past transitions of the current state and the evaluation function value of the next state with the evaluation function value of the current state By determining whether the current state is a minimal solution, and if the current state is a minimal solution, the state change cost function value vector corresponding to the element of the current state vector that changes with the state transition to the current state. The value of the element of the state change cost function value vector is increased or decreased, and the value of the element of the state change cost function value vector corresponding to the element of the current state vector that has not changed is decreased or maintained. If the current state is not a minimal solution, the value of each element of the state change cost function value vector is reduced or maintained, and the value of each element of the state change cost function value vector is updated. Then, the next state is added to the sum of the values of the elements of the state change cost function value vector corresponding to the element having a difference between the current state and the next state and the evaluation function value of the next state. A state space search method characterized by deciding the transition destination of the current state by evaluating. 4. A state vector representing a state is divided into a number of element vectors, a state change cost function value vector having a length equal to the number of the element vectors is prepared, and changes with a state transition to the current state. The value of each element of the state change cost function value vector is updated by increasing the value of the element of the state change cost function value vector corresponding to the element of the present state vector The value of each element of the state change cost function value vector is updated by decreasing or maintaining the value of the element of the state change cost function value vector corresponding to the element, and the adjacent state is changed to the current state and the adjacent state. Evaluated by the sum of the values of the elements of the state change cost function value vector corresponding to the element vector with a difference between the state and the sum of the evaluation function value of the adjacent state, A state space search method characterized by determining the transition destination of the current state. 5. In a search method of a state space for evaluating a plurality of adjacent states that can transit from the current state and determining a transition destination, the state vector representing the state is divided into several element vectors, and the element vectors are divided into several element vectors. A state change cost function value vector with a length equal to the number of is prepared, and the state change cost function value determined by the difference between the current state and the next state and the history of past transitions of the current state, and the evaluation of the next state By comparing the sum with the function value with the evaluation function value of the current state, it is determined whether or not the current state is a minimal solution, and if the current state is a minimal solution, change with the state transition to the current state. The value of the element of the state change cost function value vector corresponding to the element of the current state vector is increased, and the value of the element of the state change cost function value vector corresponding to the element of the current state vector that has not changed is changed. Decrease or maintain, update the value of each element of the state change cost function value vector, if the current state is not a minimal solution, reduce or maintain the value of each element of the state change cost function value vector, The value of each element of the state change cost function value vector is updated, and the value of the element of the state change cost function value vector corresponding to the element with the difference between the current state and the adjacent state A state space search method characterized by determining the transition destination of the current state by evaluating the sum of the sum and the evaluation function value of the adjacent state. 6. When evaluating a neighboring state and determining a transition destination, it is determined whether or not there is a neighboring state having an evaluation function value smaller than that of the current state, as well as the difference between the current state and the neighboring state and the current state. The sum of the state change cost function value determined by the history of past transitions of the state and the evaluation function value of the adjacent state is determined by comparing it with the evaluation function value of the current state. If a small adjacent state exists, it is selected as the transition destination, and if it does not exist, the evaluation function value is increased for the minimization problem and the evaluation function value is increased for the maximization problem. By degrading the evaluation function value of the current state such as decreasing, a neighboring state having a smaller evaluation function value than the current state appears, and this is selected as a transition destination. Search for the state space described in any of 5 Search method. 7. When evaluating a next state and determining a transition destination, it is determined whether or not there is a next state having an evaluation function value smaller than that of the current state, as well as the difference between the current state and the next state and the current state. The sum of the state change cost function value determined by the history of past transitions of the state and the evaluation function value of the adjacent state is determined by comparing it with the evaluation function value of the current state. If there is a small neighboring state, select the state with the smallest evaluation function value as the transition destination,
If it does not exist, by increasing the evaluation function value for the minimization problem and decreasing the evaluation function value for the maximization problem, the evaluation function value in the current state is deteriorated. The state space search method according to any one of claims 1 to 5, wherein a neighboring state having an evaluation function value smaller than that of the current state is made to appear, and this state is selected as a transition destination.