JP2001005846A - Cyclic sequence scheduling method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a cyclic sequence scheduling method for efficiently preparing a high-quality cyclic sequence schedule. SOLUTION: A cyclic sequence group initializing part 0112 generates the set of cyclic sequences while utilizing an input data base 0101 as the set of resources or jobs. A cyclic sequence group evaluating part 0113 calculates evaluation values while utilizing an evaluation function data base 0104, an evaluation weight data base 0107 and a normalization coefficient data base 0105 concerning all the cyclic sequences in the cyclic sequence set. An updating means determining part 0114 determines an updating means out of the updating means of plural cyclic sequences by deciding search conditions while using an updating strategy data base 0106. A cyclic sequence group updating part 0114 updates the cyclic sequence in one part of the cyclic sequence set to the other cyclic sequence while utilizing the updating means determined by the updating means determining part 0114. A normalization coefficient calculating part 0116 calculates a normalization coefficient effective for a cyclic sequence search part from the input data base 0101 and the evaluation function data base 0104 and stores it in the normalization coefficient data base 0105.

Description

DETAILED DESCRIPTION OF THE INVENTION

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for supporting a scheduling operation in the fields of production, transportation and logistics by a computer, and more specifically, to a method of preparing a plan using a computer and a control of a planning process. The method and the verification method of the planning result.

2. Description of the Related Art In the fields of production, transportation, logistics, etc., certain orders such as assignment order, use order, arrival order, and execution order for given personnel, materials, bases, work, etc. are determined by specific constraints and evaluation indexes. Is determined as an ordering problem or an order scheduling problem. Among them, the problem of finding a so-called cyclic sequence, in which all of the given goods and work are traced back one at a time, such as a package distribution plan in the logistics field and a vehicle operation rotation in the transport field, is called a cyclic sequence. This is called a scheduling problem or a cyclic sequence search problem. Traveling Salseman Plobrem (TSP), a search for the least costly traveling sequence that visits a given city as one of the most classical traveling sequence problems that has been actively studied at the research level for a long time Can be mentioned. To solve the TSP, the genetic solution, which is a type of iteratively improved search algorithm that repeats partial changes to the initial solution created by some means and gradually approaches the optimal solution under the set evaluation function A method of finding a solution using a framework of a genetic algorithm (hereinafter referred to as GA) is known. ((1) Katayama et al.
"A Study on Efficient Hybrid Genetic Algorithm for Traveling Salesman Problem", IEICE Transactions A and Vo
l.J81-A, No.12) In addition to GA, linear search (LP), local search, and improved local search methods such as tabu search method and annealing method are also available. Although there are various methods, a common advantage is that the method is highly responsive to changes in constraints and evaluation indices. When a certain evaluation index becomes unnecessary or a new evaluation index becomes necessary, a new evaluation function can be created simply by deleting or adding the corresponding partial evaluation function, and a plan according to the change can be created. Planning is possible. Even when the trade-off between the indices changes, it can be dealt with just by changing the weight of the partial evaluation function. Since the iteratively improved search algorithm has the advantages described above, it is promising that its application to an actual level planning problem or a planning system in which constraints and evaluation indices are likely to change depending on the planning situation. As such an example, a resource allocation plan creation system for a vehicle allocation plan in a railway field, a vehicle arrangement plan in a depot, and the like is known. ((2) Japanese Patent Application 8-10971
3) In addition, as a method of evaluating the solution in the iteratively improved search algorithm, an output value obtained by applying a partial evaluation function is multiplied by a weight of an evaluation function set in advance to obtain a linear sum. A method of setting the overall evaluation value is known. ((3) Japanese Patent Application No. 9-301979)

An iteratively improved search algorithm such as GA is a very powerful tool for so-called resource allocation problems in the field of combinatorial optimization, such as material and staffing planning. is there. Even a real-level problem that is difficult to formulate due to ambiguous constraints and evaluation indices can be easily applied, and a high-quality plan can be efficiently searched and created. However,
It has recently been pointed out that it is difficult to apply the ordering problem including the cyclic sequence search problem, and it is possible to transform the problem into a more easy problem by using the properties of the target problem, or to use an iterative improvement algorithm. In many cases, this is handled by incorporating individual processing using knowledge specific to the target problem as a base. The above-mentioned known example (1) is also a method in which a property unique to TSP is incorporated in the search processing, and is not a general-purpose search algorithm applicable to other cyclic sequence search problems. In order to apply to other problems, it is necessary to reconstruct not only the evaluation function that is easy to handle but also the inside of the search process with unique knowledge, which requires much time for development and lowers maintenance and expandability. . Another problem with the iteratively improved search algorithm is that it is difficult for the user to directly control the search. The user can pause the search, change various search parameters and evaluation functions, or restart the search after fixing a part of the plan, but the search method itself can be changed. Direct control, for example, when the search falls into a local solution and the possibility of improving the solution is low even if the search is continued further, the search space is forcibly expanded to execute a wide area search. Could not be performed. Furthermore, in the conventional example, the user was able to control the importance to the solution search by changing the weight of the evaluation index. However, since the partial evaluation functions generally have completely different output values, There is also a problem that weights cannot be set on a unified scale, and it is difficult to accurately reflect a user's intention in a solution search. SUMMARY OF THE INVENTION An object of the present invention is to flexibly and accurately reflect the importance of constraints and evaluation indices and a user intention regarding control of a search process for an actual level cyclic sequence search problem, and efficiently create a high-quality plan. It is to provide a scheduling support system that can be created.

The object of the present invention is to utilize a cyclic sequence group initialization unit for receiving input data from an input database and generating a plurality of cyclic sequences, an evaluation function database, a normalization coefficient database, and an evaluation weight database. A cyclic sequence group evaluator that evaluates the cyclic sequence, an updating unit determiner that determines an update unit of the cyclic sequence using an update strategy database, and a cyclic sequence group update unit that generates a new cyclic circuit. A cyclic sequence search unit, and a plan conversion unit that converts the cyclic sequence of the search result obtained by the cyclic sequence search unit into a plan form and stores it in a plan database,
Create a normalization coefficient calculation unit that calculates a normalization coefficient using the evaluation function database and the input database and stores it in the normalization coefficient database, and a user intervention unit that evaluates and modifies each database change and plan. This is achieved by:

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, as one embodiment of the present invention,
A vehicle operation rotation creation support system in a railway depot will be described. The present embodiment is a system for supporting a vehicle operation operation at a railway depot, and determines whether or not a vehicle is to be operated on business on a main line, which operation schedule is in charge of business operation, and repair and inspection. A basic plan (vehicle operation rotation) used when planning a daily operation schedule of a vehicle, such as when and where maintenance work such as cleaning is performed, is created. First, the input data for creating the vehicle operation rotation stored in the input data DB 0101 of FIG. 1 will be described. FIG. 2 is an example of an operation schedule. FIG. 2 shows six types of diamonds, each of which is given a unique number such as numeral 0201. The horizontal axis 0202 represents the operation time zone of the route. The black bar in each diamond is
This means operating on the main line.
03 indicates departure from the depot or station indicated by the label 0204, and triangular symbol 0205 indicates arrival at the depot or station indicated by the label 0206. A label 0207 represents the name of a partial diagram that constitutes the operation diagram. For example, the timetable of No. 2 departs from station A after 6 o'clock in the morning, then operates on the main line until noon and enters base X (partial timetable D
2) After that, return to the main line again in the evening and resume commercial operation,
After one day of operation, the vehicle finally enters the base Y (partial diamond D3). As in this example, some of the schedules that enter the depot once during business operation and then re-enter business operation, and those that operate all day on the main line like the number 1 diamond There is also. In the following description, in order to specify an operation schedule, either a diamond number such as diamond 2 or a combination of partial diamond names such as D2-D3 is used (in FIG. 2, diamond 2 and D2-D3 are the same). Service schedule). In order to implement these service schedules, vehicles for each schedule are not duplicated (one-to-one),
Allocated every day without exception. In addition, because there is special work that must be performed all day at the depot, and it is necessary to prepare standby vehicles to respond to unexpected situations such as vehicle breakdowns, vehicles for (total number of vehicles-total number of trains) The vehicle will not be in business operation, that is, will remain at the depot all day as a spare vehicle to which no schedule is assigned. The allocation to the spare depot is also stored in the input DB 0101 as input data. It is necessary to perform various maintenance work (cleaning, inspection, repair, etc.) on vehicles to maintain safety and comfort. Some of the maintenance work requires removing the vehicle from commercial operation and keeping the train at the depot for a long period of time, while completing the work in a relatively short time of about tens of minutes to several hours between commercial operations Some are relatively simple to do. An example of short-term work can be work called train inspection (or work inspection), which is implemented on a regular basis at regular intervals (currently three days) for each vehicle. Regression conditions are clearly stipulated by law. In order to realize daily transportation operations safely, reliably, and comfortably, each railway company prepares a schedule (maintenance) and maintenance work for each vehicle in advance for several days to several weeks, and that schedule Along with daily vehicle operations. This operation schedule is called a vehicle operation plan. By referring to the vehicle operation plan,
Whether a vehicle will be in commercial operation or on standby on a particular day, which schedule will be in charge for commercial operation, which depot will be in standby, and whether maintenance will be performed, It is possible to obtain all information on vehicle operation such as where, when, and what to receive. FIG. 3 shows a data structure of operation information used in the present embodiment. In FIG. 3, information relating to a certain day's operation of a certain vehicle is stored in a data field as an example. Item 0301 represents a service schedule to be implemented or a spare identification symbol, and in the example, indicates that the second diagram from the top (D2-D3) in the service schedule list in FIG. 2 is assigned. Item 0302 represents the work to be performed. If no work is scheduled to be performed, set a null character string “---” corresponding to an empty work. Item 0303 is the execution time zone of the work. In this embodiment, in order to make the description easy to understand, the work execution time zone is divided into two, that is, daytime and nighttime, and information of either “daytime” or “nighttime” is set in this item. Item 0304 is a place where the work is performed. By the way, the diagram 2 in Fig. 2 has an operation mode in which it enters the base Y once in the daytime, then exits again in the evening and performs commercial operation, and the train inspection incorporated in this operation data can be actually performed You can see that.
If work other than train inspection exists and a plurality of work can be performed, the items 0302, 0303, and 0304 are increased by increasing the number of work according to the number of work. FIG. 4 shows an image format of operation data. The operation image 0401 is a visualization of the operation data in FIG. A label 0403 indicates the content of the work, and indicates that the corresponding work is performed in the time zone where the label is located. The operation image 0402 is a visualization of the preliminary operation. A label 0404 indicates that the corresponding operation is a backup and the location where the line is located. In the example, the label 0404 indicates that the line is to be stayed at the base Y all day. In addition, it can be seen from the work label that long-term work is scheduled to be performed on the day. In planning a rolling stock operation plan, the train inspections mentioned above are usually the most rigorous in terms of regression conditions, so all railway companies take great care to ensure that train inspection intervals do not exceed the number of return days. . Also, the short return period means that
Because there are so many vehicles to be targeted, the amount of work inevitably becomes enormous, and there is no bias in the implementation place, time zone, etc.
It is also an important issue to formulate a plan so that work can be carried out efficiently. For other work, long-term work requires a long return cycle (months to years), so work plans should be prepared in advance, and operation plans can be made relatively easily while prioritizing the work plans. be able to. In addition, the regression conditions such as cleaning work are not strict, and for those that are set as a guideline, once an operation plan has been created, an appropriate plan will be made later as long as the conditions for feasibility are not disrupted. We can cope by assembling. It goes without saying that the most important thing in creating a vehicle operation plan is to improve the safety and efficiency of operation work, but patterning of operation work is also one of the important viewpoints in preparing a plan. If the daily work contents follow a certain pattern, the field workers involved in the operation work do not have to struggle to remember the work contents every day, and mistakes due to mistakes in memory can be greatly reduced. Even if complete patterning is impossible, if most of the work follows the pattern, the contents of the work can be easily understood only by focusing on the differences from the pattern. Furthermore, creating a plan using patterns has the advantage that the burden on the plan creator is far less than the task of creating a plan from scratch. Therefore, many railway companies that have a large number of trains and have large-scale routes that manage vehicles at multiple depots in a distributed manner have (1) created a basic pattern for rolling stock operation plans every time the schedule is revised, and (2) By coordinating, an actual operation plan for a certain period is created.
The basic pattern of the vehicle operation plan is the vehicle operation rotation to be created by the present embodiment. FIG. 5 is a visual image of the vehicle operation rotation. The vehicle operation rotation in the present embodiment is a cyclic allocation order of operations for one vehicle when it is assumed that there is no work schedule other than the train inspection. In FIG. 5, it is assumed that a certain vehicle is assigned operation 0501 on the first day of the planning period. Station A at night for operation 0501
Then, since the morning delivery place of operation 0502 is also station A, it is physically possible to execute operation 0502 the next day along the arrow. Operation 050 and subsequent operations are also performed along the arrow.
3,0504,0d5005,0506,0507,0508, go back again 0501
Can be operated sequentially, and this cyclic operation can be repeated any number of times thereafter. Also,
By operating the vehicles along the arrows in this way, it is possible to perform the train inspection without fail within three days of the prescribed return. As described above, when it is assumed that there is no operation other than the train inspection, a vehicle operation plan that can be executed only by mechanically applying the vehicle operation rotation to the vehicle can be created. The vehicle operation rotation includes all schedules and spares without duplication. From this, the vehicle operation rotation represents the ideal assignment order for one vehicle, and at the same time,
The operation details of the day are also shown at the same time. Therefore, as shown in FIG. 6, if an ideal situation is assumed in which there is no work schedule other than train inspection, such as long-term work, the operation rotation pattern is applied to all vehicles by shifting the allocation start position. Alone can provide a high quality vehicle operation plan. Obviously, there are few situations where there is no long-term work schedule, and it is usually necessary to perform long-term work on some vehicles. In consideration of other work plans, a plan will be made by partially breaking the vehicle operation rotation. FIG. 7 shows the data format of the vehicle operation rotation. It has a one-dimensional array structure, and the array elements are pointers to operation data. The order of arrangement of the elements corresponds to the order of assigning operations.
In the figure, the second pointer 0701 in the array points to operation 0702. The number of array elements is equal to the total number of diamonds including spares, and the diamonds and spares included in each operation are different from each other. The vehicle operation rotation creation support system of the present embodiment accepts input data consisting of operation schedules and spare allocations from the input DB 0101 as shown in FIG. 1, creates a vehicle operation rotation in the format of FIG. Store. In creating an executable vehicle operation rotation, it is necessary to observe the following constraints. (1) The operations of the diamond (spare) connection are arranged in the physically correct order. That is, the morning departure location of the operation B located next to any operation A during the rotation (presence location on the spare line) is always equal to the night storage location of the operation A (presence location on the night) (connection condition). (2) Train inspections can be performed within the specified number of return days specified by law by allocating operations according to the rotation (return conditions). The following are examples of evaluation indices to be considered in order to obtain not only feasible but also high-quality vehicle operation rotation. (1) The total number of train inspections is small (2) There is little variation in spares (3) There is little variation in vacant hours where work is possible (4) There is little deviation in locations and time zones of train inspections (5) Train inspections (6) Avoid allocating train inspections as much as possibleBecause some of the evaluation indicators are in a trade-off relationship with each other, consider the priorities and carefully balance the two to plan. Must be created. In addition, the evaluation index and its weight change variously according to the vehicle operation policy and plan creator of each depot, so that it is required to flexibly respond to such changes and create a vehicle operation rotation. Regarding the connection condition, it is possible to determine whether or not the connection is satisfied from the order of the arrangement of the diamond and the spare included in the rotation. In addition, the regression condition can be satisfied by arranging the diamond and the spare so that the train inspection can be set within three days, and can be regarded as a condition depending on the order of the arrangement of the diamond and the spare. Therefore, the vehicle operation rotation creation problem can be essentially reduced to the order of arranging diamonds and spares, that is, a cyclic sequence search problem for finding a sequence that circulates all diamonds and spares. In this embodiment, the vehicle operation rotation creation problem is regarded as a cyclic sequence search problem, and a solution search is performed to create an executable high-quality vehicle operation rotation. Hereinafter, a specific creation method will be described mainly with reference to FIG. First input D
The input data is obtained from B0101 and passed to the cyclic sequence search unit 0111. The cyclic sequence search unit includes a cyclic sequence group initialization unit 0112, a cyclic sequence group evaluation unit 0113, and an update unit determination unit 01 as internal processing.
14. Having a cyclic sequence group updating unit 0115, solves a cyclic sequence search problem corresponding to input data, and stores the search result in a cyclic sequence DB0103.
Output to Unless otherwise specified, the operation schedule uses the six data shown in FIG. 2 and the spare data is allocated to the garage X and the garage Y one by one. In addition, serial numbers are assigned to the diamonds and the spare in order to handle them together. The same number as the diamond number is given to the diamond, the number 7 is given to the spare of the garage X, and the number 8 is given to the spare of the garage Y. Hereinafter, an outline of the cyclic sequence search unit in the present embodiment will be described using the processing flow of FIG. In the present embodiment, a genetic algorithm (Genetic Algorithm, GA), which is an optimization method that imitates the evolutionary mechanism of an organism, is used as a framework of a search algorithm. G
A perceives the solution of the search problem as one living individual, and the constraints of the problem to be solved and the evaluation index as the individual's survival environment. In the initial state of the search, a population of a specified size is randomly generated, and natural selection, mating,
Genetic operations such as mutation are repeated to generate new individuals one after another. It is considered that the generation of the individual population is updated by one genetic operation, and the individual with the highest fitness in the population obtained after the repetition of the specified generation is output as the final solution. GA is an iteratively improved search method in which partial updating of an initial solution created by some means is repeated, and the approach gradually approaches an optimal solution under a set evaluation function. The present invention, if the internal processing 0112,0113,0114,0115 of the cyclic sequence search unit in FIG. 1 can be provided, the search framework is not limited to GA linear programming (LP), local search method, local search method Other iterative improvements, such as the improved taboo search and annealing methods, can also be used. In the processing flow of FIG. 8, first, in step 0801, the generation number counter is initialized to 1. Then step 080
In step 2, a plurality of individuals representing a cyclic sequence are generated. This processing corresponds to the cyclic sequence group initialization processing 0112 in FIG. Figure 9
Shows the data structure of an individual (cyclic sequence). It has an array structure having data fields equal to the number of data (diamond + total number of spares), and each field stores a data number. Although the individual expression is a simple array in appearance, it is special data that has the following restrictions on the method of storing numerical values. (1) There are no duplicates in the stored values. (2) For all data (diamond or spare), adjacent data (in the case of end data, the data at the other end of the array)
And the connection conditions are satisfied. The connection condition is the above-mentioned condition that “the night storage location and the morning storage location are equal”. The arrangement of diamonds and spares represented by the individual data of FIG. 9 is shown in FIG.
, And all the data satisfy the connection conditions. The connection possibility between the two data arbitrarily extracted from the input data can be determined using the connection constraint graph shown in FIG. The connection constraint graph is a directed graph having (number of data + 1) vertices. A number other than 0 attached to each vertex indicates the data number of the input data. The vertex of number 0 is a dummy vertex, and is used to determine the head data of the array in the individual generation processing described later. The branch connecting the vertices indicates the possibility of connection of the two data. For example, in the connection constraint graph of FIG. 10, the existence of the branch from the vertex 1 to the vertex 8 means that the corresponding operation diagram 1 It means that it can be connected to the spare. Since the actual parking place of the operation schedule 1 is the garage Y and is connectable, the connection restriction graph correctly represents the connection possibility between actual data. It is assumed that the dummy vertex can be connected to all other data. The individual data corresponds to one of the cyclic sequences that trace all the vertices existing on the connection constraint graph once without exception. For example, the individual data shown in FIG. 9 is an example of the cyclic sequence 0 → 5 → 2 → 8 → 3 → 1 → 6 → 4 →
It corresponds to 7 → 0. The vertices connected from vertex 0 represent the head of the individual array, and subsequent numbers represent the data numbers stored in the array.
The individual data of FIG. 9 can be obtained from the eleven cyclic sequences. As shown in FIG. 11, a cyclic sequence that passes through all the vertices on the graph exactly once is called a Hamiltonian cycle in the field of discrete mathematics, and in this embodiment, the Hamiltonian cycle on the directed graph is represented by an individual representation of GA. It can be understood that it is used for. Therefore, in the following explanation, GA individuals,
The cyclic sequence and the Hamiltonian cycle are the same, and the terms are switched according to the context. The description of the population generation processing 0802 in FIG. 8 will be continued. In this process, individual data is generated by randomly selecting a Hamiltonian cycle in a connection constraint graph obtained from input data. The outline of the random generation processing of individual data will be described with reference to the processing flow of FIG. Step 12
In 01, an array A having a size equal to the number n of vertices and an index I for specifying a position on the array A are prepared. Array A
Stores a series of vertices representing a Hamiltonian cycle when this processing is completed. In step 1202, the index I is set to 1 and all vertices are marked as unselected. Next, in step 1203, a dummy vertex 0 is stored at the head of the array A, and the value of the index I is incremented by 1 assuming that vertex 0 has been selected. In step 1204, the (I-1) -th vertex from the beginning of the array A is extracted and set as v1. In step 1205, it is determined whether I is equal to n + 1, that is, vertex v1 is the last vertex of the Hamiltonian cycle, and if the determination result is Yes, the process proceeds to step 1206, where v1 can be connected to the first vertex of array A. It is determined whether there is any, and if the result is Yes, the result of the entire process is set to true and the process ends. If the result is No, the result of the entire process is set to false and the process ends. If the determination result of step 1205 is No, step 12
Go to 09. Here, of the vertices that can be connected from vertex v1,
Collect all the items that have not yet been selected and call the set S. In step 1210, it is determined whether or not S is an empty set. If S is an empty set, the process proceeds to step 1208, where the result of the entire process is set to false, and the process ends. If S is not an empty set, the process proceeds to step 1211. First, one vertex is randomly selected from the set S and set as v2. Further, v2 has been selected, the value of the index I is increased by 1, and the portion (subroutine R) enclosed by the dotted line in the processing flow is recursively executed. The subroutine R recursively calls itself, and as a result, the vertices at each position of the array A are sequentially selected from the top. If the result of this recursive call is false, proceed to step 1214, return v2 to unselected, return to step 1211 to reduce the value of I by 1 and select another vertex. An array if the result is true
Since the vertex selection has been successful up to the end of A, the process proceeds to step 1207, and the process ends with the result of the entire process set to true. If the result of the whole process is true, the vertices of the Hamiltonian cycle created starting from the dummy vertex 0 are stored in the array A from the beginning in the order of the cycles. If a Hamiltonian cycle exists in the given connection constraint graph, one arbitrary Hamiltonian cycle can be generated by the above generation processing. If there is no Hamiltonian cycle, the processing result will return false. In the cyclic sequence search problem, since it is assumed that the connection constraint graph has at least one Hamiltonian cycle, the result does not become false, and the Hamiltonian cycle can always be obtained by executing this processing. Returning to the population generation process 0802 in FIG. 8, the process is continued. One piece of individual data is generated by copying the array A representing the Hamiltonian cycle obtained by this process from the second vertex to the last vertex into the individual data without changing the order. The number of individuals to be generated is specified as a parameter from the search parameter DB in FIG. FIG. 13 shows an example of the data structure of the search parameter. As shown in the figure, the search parameters are (1) the number of individuals, (2)
It has at least four data fields of the number of search generations, (3) natural selection rate, and (4) mutation rate. It is the number of individuals in the field 1301 that defines the number of individuals to be generated in the individual group generation processing. In the figure, generation of 30 individuals is instructed. The meaning and use of the other parameters will be sequentially described below. Returning to FIG. 8, the description of the tour search process will be continued. In step 0802, the population is evaluated. This process corresponds to the cyclic sequence group evaluation process 0113 in FIG. For the vehicle operation rotation obtained from the arrangement of the diamonds and spares represented by the individual, two constraints, connection conditions and regression conditions, and the satisfaction of various preset evaluation indexes are checked. It is assumed that a partial evaluation function is individually provided as a library for each constraint condition and evaluation index. The only condition that the partial evaluation function is a penalty function whose value is 0 when the corresponding constraint condition or evaluation index is completely satisfied, and whose value increases as the degree of satisfaction decreases. For example, the following function can be given to the evaluation index “the total number of train inspections is small”.

(Equation 1) The evaluation function for calculating the evaluation value is obtained from the evaluation function DB0104 in FIG. FIG. 14 shows an example of the evaluation function data. Item 1401 stores a number for distinguishing the evaluation function. Item 1402 is the name of the constraint or evaluation index. Item 1703 is a logical value indicating whether or not the partial evaluation function is valid. True indicates valid, and False indicates invalid. Item 170
4 is a pointer to a processing routine in the library corresponding to the evaluation function. Each partial evaluation function is not uniform in importance. For example, the partial evaluation function corresponding to the connection condition and the regression condition that must be satisfied is much more important than others. Also, the dispersion of the spare, which is one of the evaluation indices, is an index from the viewpoint that it is preferable that the allocation of the spare is performed at a uniform interval as much as possible, since the frequency of use of the vehicle is uniform, which is preferable. Are not considered to be so important elements, and the importance of the corresponding partial evaluation function is relatively low. The problem of how much importance is given to a partial function has many subjective factors. Therefore, in this embodiment, the weights of the partial functions are given as data that can be externally changed by the system operator, and are stored and managed in the evaluation weight DB of FIG. FIG. 15 shows an example of the evaluation weight data. Item 1501
Stores the number of the partial evaluation function. The item 1502 stores a weight. The numerical value given as the weight is an integer value in a certain range, and one numerical value in the range is given to all partial functions. In the example of FIG. 15, the range is 0 to 100, and the weight of the evaluation function corresponding to the number 1, that is, the connection condition is 100, and the weight of the evaluation function corresponding to the number 8, that is, the assignment of the train inspection to the spare, is 100. It can be read that it is 10. In the evaluation processing, effective partial evaluation functions are sequentially applied to individuals to obtain partial evaluation values, and the values are combined while considering weights to obtain an overall evaluation value. The simplest combination method is a linear weighted sum in which partial evaluation values are multiplied by weights and added. The biggest drawback of this method is that it is difficult to weight partial evaluation functions on a uniform scale. For example, the partial evaluation function (No. 6) corresponding to the evaluation index of the evaluation index `` set the train inspection equally to the place and time zone '' is a ratio of the number of set request numbers and the actual set number for each place and each time zone ( Since the minimum (0, maximum 1) is calculated and the variation (standard deviation) is output, the output value range is 0 to 1. On the other hand, the partial function (No. 4) corresponding to the preliminary variation giving the same weight in the evaluation weight data of FIG. 15 is the variation in the interval between the two preliminary in the total of eight diamonds and the preliminary. Therefore, the output value is 0 (in the case of equal intervals) at the minimum and 3 (in the case where spares are arranged) at the maximum, and the upper limit of the output value is different. The implications are completely different. As described above, if the output value range is different, the degree of influence of the weight differs, so that it is difficult to determine the value according to a certain policy when setting the weight as a parameter. Therefore, in the present invention, when calculating the evaluation values, the output values of all the partial functions are adjusted to a certain range and then multiplied by the weight, thereby realizing the unification of the meaning of the weight. Aligning the output value range is hereinafter referred to as output value normalization. For example, to normalize the output value v of a partial evaluation function ei whose output value range is 0 to xi (i = 1, ..., n) to 0 to X, the coefficient ci = Multiply X / xi. This coefficient ci
Is called a normalization coefficient. The normalization coefficient is the normalization coefficient DB0 in Fig. 1.
Stored in 105. FIG. 16 shows an example of the normalization coefficient data. Item 1601 stores the number of the partial evaluation function. item
1602 stores a normalization coefficient. In the example, the normalization coefficient when the maximum value of the output value is assumed to be 100 is stored.The normalization coefficient of the partial function corresponding to the connection condition of No. 1 is 25.0, and it corresponds to the regression condition of No. 2. It can be seen that the normalization coefficient of the partial function is 50.0. From the above, assuming that the effective partial functions are ei (i = 1, ..., n) and the corresponding weights and normalization coefficients are wi and ci, respectively, the method of calculating the overall evaluation value E of the individual is as follows: It is expressed by an equation.

(Equation 2) In the individual evaluation processing, evaluation function data, evaluation weight data,
After acquiring the normalization coefficient data from the database, the evaluation value of one individual is calculated according to this equation, and the process is repeated until the evaluation values of all the individuals are calculated. The problem here is how to set the normalization coefficient of each partial evaluation function. Since the logical value of the output upper limit value of the partial evaluation function is obtained by logically analyzing the processing of the partial evaluation, it is possible to use it to calculate the logical value of the normalization coefficient from the above method. . However, the more complicated the evaluation process becomes, the more difficult the analysis becomes. If the number of partial evaluation functions increases, the labor of the analysis becomes extremely enormous. Even if the output upper limit values of all partial evaluation functions can be calculated logically, the actual output upper limit value is not necessarily a logical value because the actual output value changes depending on the nature of the input data. The value does not always match the upper limit, and a value much smaller than that often occurs. Therefore, if a logical upper limit value is used for calculating the normalization coefficient, the evaluation function cannot be correctly weighted. Therefore, in the present invention, FIG.
In 1, every time data is input from the input data DB0101 to the cyclic sequence search unit 0111, the normalization coefficient calculation unit 0116 calculates the actual value of the normalization coefficient according to the input data and stores it in the normalization coefficient DB0105. Diagram of how to calculate the normalization coefficient
This will be described with reference to a processing flow of 17. First, in step 1701, evaluation function data is obtained from the evaluation function DB, and an array A having a size equal to the total number (n) of partial evaluation functions is prepared. Furthermore, the values of all elements of A are initialized to -1. At the end of the process, the normalization coefficients are stored in the array A. Next, in step 1702, the loop counter i
Is initialized to 1 and the end of the loop is determined in step 1703. If the value of i is equal to or less than the specified number of times given as a constant, the process proceeds to step 1705. The value of the specified number of times is an arbitrary value of 1 or more, but the larger the value is set, the higher the accuracy of the obtained normalization coefficient becomes. In step 1705, one arbitrary individual is obtained using the random generation flow of the Hamiltonian cycle described in FIG. Next, in step 1705, the value of the evaluation function counter j is initialized to 1. Step 1706
Then, the end of the loop is determined. j is the total number of functions n
If the value is less than, the process proceeds to step 1707, where the evaluation function with the number j is obtained from the evaluation function data and applied to the previously generated individual to calculate the evaluation value x. In step 1708, the j-th value from the left of array A is compared with the value of evaluation value x, and if x is larger, the process proceeds to step 1709, where the j-th value in array A is updated to x, and then step 1710 Proceed to. If x is equal to or smaller than the j-th value of A, the process directly proceeds to step 1710, increments the value of j by 1, and returns to step 1706 again. J in step 1706
If the value exceeds the total number n of functions, the flow advances to step 1711 to increase the value of the loop counter i by 1, and then returns to step 1703. If i exceeds the specified number in step 1703, the process proceeds to step 1712, where it is assumed that the numerical value stored in array A is an approximate value of the output upper limit of each partial evaluation function, and all values are divided by a constant X. The normalization coefficient is calculated, stored in the array A again, and the process ends. The constant X is a value that aligns the upper limit. The array A thus obtained represents the normalization coefficient of the partial evaluation function whose k-th value from the left is number k. To summarize the above processing, (1) Individuals are randomly created from the input data a prescribed number of times, and each time a partial evaluation function is applied to obtain an evaluation value and update the highest value of the partial evaluation function,
(2) The normalization coefficient is calculated using the finally obtained maximum value as an approximate value of the output upper limit value. By calculating the normalization coefficient using the above processing, it is possible to obtain a normalization coefficient with much higher accuracy than when using the logical output upper limit value. Returning to the cyclic sequence search flow of FIG. 8, the description will be continued. After the individual group is evaluated in step 0803, the process proceeds to step 0804, and the value of the generation number counter i is increased by one. Next, in step 0805, the end of the process is determined. If the value of i is smaller than the value defined by the number of generations to be searched for the search parameter, the process proceeds to step 0806 to continue the search and determines the individual crossing means. The crossover means determination processing is processing corresponding to the update means determination unit 0114 in FIG. 1, and is a part that determines a crossover method for generating a new individual in the genetic operation in step 0807 in FIG. In order to explain this processing, it is necessary to explain the crossover processing of individuals, so that the genetic operation will be described first. The genetic operation is processing corresponding to the cyclic sequence group updating unit 0115 in FIG. 1,
In the group of the cyclic series, the contents of the group are changed by deleting and adding the cyclic series. The outline of the gene manipulation processing will be described with reference to the processing flow of FIG. First step 18
At 01, a copy of the population is created and designated as S. Next, in step 1802, the specified number of individuals is deleted from the individual population. The specified number is a constant calculated using the value of the natural selection rate item of the search parameter. In the example of the search parameters in FIG. 13, the natural selection rate is 30, which means that 30% of individuals in the population are deleted. If the group size is 30, the specified number n is 30% of that
9, which deletes nine individuals in the natural selection process.
The method of selecting an individual to be deleted is performed by using the evaluation values, all the individuals are arranged in descending order from the one with the highest evaluation value, and n individuals are deleted from the bottom in descending order of the evaluation value.
In step 1803, the loop counter i is initialized to 1. In step 1804, the loop end is determined, and the counter
If the value of i is equal to or less than the number n of deleted individuals, the process proceeds to step 1805 to continue the process. In step 1805, two individuals are selected from the individual population S from which the population before deletion is copied, and are set as parent individuals A and B, respectively. The selection method may be completely random, or a selection probability proportional to the evaluation value of the individual may be given to increase the selection ratio of a parent individual having a good evaluation value. However, the same individual is not selected as the parent individual A and the parent individual B. Next, in step 1806, crossover is executed with the two individuals A and B selected previously using the crossover means determined in step 0806 in FIG. Crossover is a process of generating a new individual based on a certain individual. The individual on which it is based is called a parent individual, and the resulting individual is called a child individual. Hereinafter, the crossover means will be described. First, the most basic crossover means on which all crossover means are based is arbitrarily selecting either parent individual A or parent individual B,
On the other hand, a child individual that has been partially modified as shown in FIG. 19 is used as a child individual. In FIG. 19, the parent individual is the same as the Hamiltonian cycle shown in FIG. 11, and the positions of the vertices are modified so that the path becomes annular. Replace any branch a of the parent individual with a branch a 'that does not exist in the parent individual,
Similarly, a child individual can be obtained by replacing arbitrary branches b and c with branches b ′ and c ′ that do not exist in the parent individual. Branch a ',
b 'and c' always select those that exist in the connection constraint graph. By this operation, an effective Hamiltonian cycle on the connection constraint graph can always be obtained as a child individual. In the field of combinatorial mathematics, a set of cyclic graphs obtained by replacing arbitrary λ branches on the cyclic graph π with other branches is referred to as λ neighborhood of π. Since the Hamiltonian cycle is also a kind of cyclic graph, the basic crossover means can be regarded as a process of generating one arbitrary circular graph in the vicinity of the three based on the parent individual. However, in order to guarantee that the obtained individual is a valid Hamiltonian cycle which does not cause a connection violation, a newly added branch is necessarily selected from those existing on the connection constraint graph. The basic crossover means is a process of selecting one of the two parent individuals and generating a child individual using it as a model. In the branch selection process, some information relating to the properties of the other parent individual is obtained. A plurality of crossover means are provided by changing the processing so as to narrow the selection range of the branches by utilizing. When a parent individual A is selected as a model of a child individual, and a set of all branches included in the parent individual B but not included in the parent individual A is S, in the present embodiment, the following six types of processing are performed. Give as crossover means. [Means 1] "Basic crossover" As described above [Means 2] "One branch inheritance crossover" One branch is always selected from the set S, and the remaining branches are arbitrarily selected. For example, if the set S = {branch a: vertex 0 → vertex 4, branch b: vertex 6 → vertex 3, branch c: vertex 1 → vertex 8}, one of the new branches to be added to the parent individual A Is always one of the branches a, b, and c, and the remaining two select an arbitrary branch on the connection constraint graph similarly to the basic crossover means to generate a child individual. [Means 3] "One vertex inherited crossover" One branch is selected from those whose start vertices are equal to any of the branches of the set S, and the remaining branches are arbitrarily selected. For example, if the set S is the same as above, one of the new branches to be added to the parent individual A is always a branch a,
A branch whose starting vertex is any of vertices 0, 6, and 1 that are the starting vertices of b and c (regardless of the end point), and the remaining two are any branches on the connection constraint graph similar to the basic crossover means. Select to generate offspring individuals. [Means 4] “Compound Crossover 1” A one-branch inheritance crossover is performed to obtain a child individual, which is further subjected to a basic crossover as a final child individual. [Means 5] “Compound Crossover 2” 1 A child individual is obtained by performing one-vertex inheritance crossover, and a product obtained by performing a basic crossover on the obtained individual is defined as a final child individual. [Means 6] "Compound Crossover 3" One branch inheritance crossover is performed to obtain a child individual, and further one vertex inheritance crossover is performed between the individual and the parent individual B. In addition to the crossover means listed above, two or three branches from the parent individual
What inherits with a book, what inherits two or three vertices,
Alternatively, various crossovers in which the crossover means to be combined and the number thereof are changed can be considered. Since what kind of crossing means is required differs for each problem to be solved, necessary ones are appropriately designed and incorporated based on the above-mentioned crossing means. The relative distance L between two individuals is defined as L = (total number of vertices)-(number of common branches). From this definition, if two individuals are exactly the same, the distance is 0 because the number of common branches is equal to the total number of vertices. The distance is maximized when there are no common branches, and the distance in that case matches the total number of vertices. FIG. 20 is a graph showing the relationship between the total number of child individuals that can be generated from each of the crossover means described above and the relative distance between parent individuals A and B. In this graph, the total number of vertices is taken as n, the horizontal axis is taken as the relative distance, and the vertical axis is taken as the total number of combinations. The function expression is attached. Since the basic crossover generates a child individual from one parent individual, a constant total number of combinations is always maintained regardless of the relative distance between the parents. For the other crossovers, the total number of combinations changes according to the relative distance, and it can be seen from the graph that as the relative distance decreases, the number of possible combinations of the offspring individuals decreases. In addition, compound crossovers 2 and 3 have a relative distance smaller than 2 when the total number of combinations is smaller than the basic crossover, that is, 1
Although it is 0, the relative distance cannot be 1 and the relative distance 0 cannot be assumed because the crossing parent individuals are assumed to be different from each other. Therefore, these two crossings always have a larger total number of combinations than the basic crossover, and the search range is extremely large. Since the crossover means is a process of partially changing the parent individual and generating an individual similar to the parent individual as a child individual, as the generation loop progresses in the search process,
The relative distance between any two individuals in the population is thought to decrease. Therefore, if a crossover means other than the basic crossover is used, the total number of possible combinations of offspring individuals decreases with a decrease in the relative distance. Then, the search range becomes narrower and the search becomes more detailed, so that the possibility of finding an individual having a better evaluation value increases. However, if the speed at which the selection range is narrowed is no longer too large, the search area is searched in a wide area, that is, the so-called wide area search is not sufficiently performed, and the search area falls into a narrow area where no good individual exists, which is insufficient. It is more likely that only a simple search result will be obtained. Optimum values such as the speed of narrowing the selection range, the rate of narrowing (straight line, quadratic curve, etc.) and the initial value of the selection range differ depending on the nature of the problem to be applied. Therefore, in the present invention, a plurality of crossover means can be prepared in advance and selected according to the problem. Also provided is a means for switching the crossover means according to the state of the search during the search. This is the role of the crossover determining unit 0806 in FIG. Returning to FIG. 18, the description of the gene manipulation processing will be continued. In step 1807, it is checked whether or not the same individual as the generated offspring individual already exists in the group. If so, the process returns to step 1806 to repeat the crossover. If the same individual does not exist, step 1808
And add the offspring to the group, and add a loop counter
The value of i is incremented by 1 and the process returns to step 1804. If the value of the counter i exceeds the number n of deleted individuals in step 1804, the process proceeds to step 1809. In step 1809, mutation is performed on the newly generated individual. Mutation is the process of randomly changing a part of an individual independently of other individuals or the search situation, and plays an auxiliary role in expanding the search range when the search falls into a local solution .
Mutation is performed on one individual by the following method. (1) Starting at an arbitrary vertex of an individual, select vertices one by one until the circuit goes around the route. (2) Every time a vertex is selected, a probability determination as to whether or not the mutation is actually performed is performed using the value set for the mutation rate of the search parameter. In the search parameters of FIG. 13, the mutation rate is 0.1, which means 0.1%, that is, one thousandth of the probability. (3) As a result of the probability judgment, when it is determined that mutation is to be performed, the following processing is performed. First, the vertex of interest
Cut two branches with the end point of the branch extending from x and x as the vertex y and the start point of the branch connected to x as the vertex z. On the other hand, for a vertex x 'on the individual, the end point of the branch extending from x' is
Similarly, two branches are cut with the starting point of the branch connected to y 'and x' as the vertex z '. In this state, branches z → x ′, x ′ → y, z ′ → x, x
→ Check whether y 'exists on the connection constraint graph, and if so, add these branches. If not, the process is repeated until a vertex x ′ having a branch is found. Mutation is an operation of exchanging the positions of two vertices x and x 'as long as they satisfy the connection constraint graph, as shown in FIG. 21, and the resulting individual is always a valid Hamiltonian cycle that satisfies the connection conditions . Fig. 8
After the description of the genetic operation processing 0807 of the above, the processing returns to the previous step to explain the crossover means determination processing 0806. In order to dynamically switch and apply a plurality of built-in crossover means (update means) during solution search, the crossover means determination processing 08 shown in FIG.
At 06, the crossover means is selected. In the present invention, in order to specify a condition for updating means switching, search state data indicating the state of the cyclic sequence search at the time of starting the updating means determination processing is defined and used. The search situation data is composed of a plurality of data fields for various elements involved in the search, and at least the number of search steps,
It has independent fields for specifying the evaluation value of the best cyclic sequence obtained up to that point, its improvement speed, and the value range of the density (average relative distance) of the cyclic sequence group.
FIG. 22 shows an example of search status data in this embodiment. The field of the element related to the search is structured in a vertically arranged list, and each field has four slots: an element name label, a lower limit of the value, an upper limit of the value, and a flag for validity determination. By storing specific numerical values in the upper and lower limit value slots, the value range of the corresponding element can be designated. Also, an arbitrary value can be represented by storing a null value “---” instead of a numerical value. When the same value is stored in both the upper and lower limit value slots, it means that a specific value is specified instead of the value range. A logical value true or false is stored in the validity determination flag. Value tru
e means that the corresponding element is used for determining the search situation, and the value false means that the corresponding element is removed and the search situation is investigated. Search situation data A with search step
To determine if matches, first take the valid fields in A independently and check whether the value of the corresponding element in the current search step is within the specified value range, Only if all the results are successful,
It is determined that “the corresponding search step is search situation A”.
Hereinafter, each field of the search status data will be described. A field 2201 relates to the number of generations in the search, and specifies "200 or more generations" in the example. Field 2202 relates to the rate of improvement of the best individual in the population. Here, the improvement speed of the best individual means the rate of change of the evaluation value of the best individual, and can be given by, for example, the following equation.

(Equation 3) In the example of FIG. 22, the situation where the “improvement speed is 10.0 or less” is specified. The average value of the relative distance between any two individuals in the population will be referred to as population density. The next field 2203 is a part for specifying the value range of the population density. In the example, a situation where the average value of the relative distance is 6.0 or less is specified. The field 2204 is a part for specifying the overall evaluation value of the best individual. In the example, the validity determination flag is false, and is not used for determining the search status. The following fields relate to the output value of the partial evaluation function of the best solution. One field is assigned to every partial function. The example in FIG. 22 shows that "the number of generations is 200 or more, the improvement rate of the best solution is 10.0 or less, and the density of the population is
6.0 or less, connection condition and regression condition violations are resolved, and train inspection settings are not biased. " Summarizing this, `` At the stage when the broad search in the early stage of the search has been settled and the search is being shifted to the local search, a rather good and feasible solution that satisfies some of the essential conditions and some of the evaluation indexes has already been obtained. Is specified. Under these circumstances, in order to obtain a better plan with a high degree of satisfaction of the evaluation index, it is considered effective to perform a more detailed search by narrowing the search range. Conversely, in a situation where a feasible solution that satisfies the two constraints of the connection condition and the regression condition is not found even if the search step proceeds to some extent, if the search range is expanded, the possibility of reaching the feasible solution is often increased. Conceivable. In the present invention,
In order to achieve such adjustment of the search range according to the search situation, update strategy data composed of search situation data and update means is defined and used. The update strategy data is a one-to-one pair of the search situation data and the update means.For example, if the update strategy data X is a pair of the search situation data A and the update means M, X If the situation matches A, switch the updating means to M. ''
Means that. FIG. 23 shows an example of the update strategy data in this embodiment. Item 2301 contains a pointer to the search status,
The item 2302 stores a pointer to the updating means corresponding to the search situation. The required number of update strategy data is stored in the update strategy DB of FIG. In addition to the update strategy data, the update strategy DB also stores default update means to be selected when the current search step does not match any search situation data. In the present embodiment, for example, assuming that the default updating means is the basic crossover and the search status item of the update strategy data in FIG. 23 points to the search status data in FIG. In order to switch to local search, the updating means is switched from the basic crossover to the one-branch inheritance crossover. " The specific contents of the update means determination processing will be described using the processing flow of FIG. First, in step 2401, the value of the loop counter i is initialized to 1. Then step 24
At 02, a loop end determination is made. If the value of i is less than or equal to the total number of update strategy data stored in the update strategy DB, the process proceeds to step 2403 to continue the process. In step 2403, the i-th update strategy is extracted from the database, and in step 2404, it is determined whether the search situation data included in the update strategy matches the current search situation,
If they match, the process proceeds to step 2405, where an update strategy included in the update strategy is selected, and the process ends. Step 2404
In step 2406, if the search status data does not match, the value of the loop counter i is increased by 1 in step 2406 and
Return to If the value of the loop counter i exceeds the total number of update strategies in step 2402, the process proceeds to step 2407, where the default update means registered in the database is selected, and the process ends. Returning to the tour search flow of FIG. 8, the description will be continued. Crossover means selected in step 0806
After the individual is updated (cyclic sequence update) in the genetic operation in step 0807 using (update means), the process returns to step 0803 again to repeat the processing. In the end determination of the process of step 0805, the loop exits the loop when the value of the loop counter i exceeds the number of search generations, and the group having the highest evaluation value among the finally obtained individuals (cyclic group) is determined. The search result is stored in the cyclic sequence DB 0103 of FIG. 1, and the cyclic sequence search processing ends. Returning to FIG. 1, the description of the present invention will be continued. The cyclic sequence obtained by the cyclic sequence search unit is converted into a data format suitable for the target problem by the plan conversion unit 0117, and the result is stored in the plan DB. In the present embodiment, the cyclic sequence is converted into the vehicle operation rotation data format shown in FIG. At this time, the operation data includes data setting items related to the presence / absence of train inspection and the location and time zone of train inspection. Here, the operation data is set so as to follow the evaluation value of the partial evaluation function related to train inspection during the cyclic sequence search. Set the value. The created plan is stored in the user intervention unit 0118.
, The system operator receives a validity check via a display device. When an approval instruction is received from the operator via the input device, the plan is registered as a formal plan, converted into a form format, and output from the printer. In the user intervention section, the system operator partially modifies the plan via the input / output device or changes the data content by operating the evaluation function DB0104, the update strategy DB0106, the evaluation weight DB0107, and the search parameter DB0108. Can be. The data can be freely changed at each stage before the search is started by the cyclic sequence search unit, at any stage during the search, and after the search is completed and the plan is output. If the plan creation result is unsatisfactory because the approval of the system operator cannot be obtained, the plan can be re-created any number of times according to the instruction of the system operator.

According to the present invention, efficient creation of a high-quality cyclic sequence schedule can be realized.

[Brief description of the drawings]

FIG. 1 is a system configuration diagram of an embodiment.

FIG. 2 is an explanatory diagram of a scheduled operation schedule.

FIG. 3 is a data structure diagram of operation information.

FIG. 4 is an explanatory diagram of operation information.

FIG. 5 is an explanatory diagram of a vehicle operation rotation.

FIG. 6 is an explanatory diagram of a vehicle operation plan creation method.

FIG. 7 is a data structure diagram of a vehicle operation rotation.

FIG. 8 is a flowchart showing a flow of processing of a cyclic sequence search unit.

FIG. 9 is a data structure diagram of a cyclic sequence.

FIG. 10 is an explanatory diagram of a connection constraint graph.

FIG. 11 is an explanatory diagram of a Hamiltonian cycle on a connection constraint graph.

FIG. 12 is a flowchart illustrating a flow of a Hamiltonian circuit generation process.

FIG. 13 is a data structure diagram of a search parameter.

FIG. 14 is a data structure diagram of an evaluation function.

FIG. 15 is a data structure diagram of evaluation weights.

FIG. 16 is a data structure diagram of a normalization coefficient.

FIG. 17 is a flowchart illustrating a flow of processing of a normalization coefficient calculation unit.

FIG. 18 is a flowchart illustrating a flow of a genetic operation process.

FIG. 19 is an explanatory diagram of a basic crossover means.

FIG. 20 is a diagram showing the relationship between the total number of combinations of individuals that can be generated by the crossover means and the relative distance of parent individuals.

FIG. 21 is an explanatory diagram of a mutation.

FIG. 22 is a diagram illustrating an example of search situation data.

FIG. 23 is a diagram illustrating an example of update strategy data.

FIG. 24 is a flowchart showing a flow of processing of an updating means determining unit.

[Explanation of symbols]

0101… Input DB, 0102… Plan DB, 0103… Circulation Series DB, 01
04… Evaluation function DB, 0105… Normalization coefficient DB, 0106… Update strategy
DB, 0107: evaluation weight DB, 0108: search parameter, 0111 ...
Cyclic sequence search unit, 0112: Cyclic sequence group search unit, 0113: Cyclic sequence group evaluation unit, 0114: Update means determination unit, 0115: Cyclic sequence group update unit, 0116: Normalization coefficient calculation unit, 0117: Plan conversion unit 0118: User intervention unit, 0121: Input device, 0122: Display device, 0123: Printer.

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5B046 AA00 AA04 BA08 GA01 GA02 KA05 5B049 BB07 BB31 CC32 CC40 DD00 EE03 EE05 EE31 FF02 FF03 FF04

Claims (1)

[Claims] 1. A method for supporting, by a computer, the creation of a schedule representing a cyclic sequence of a plurality of resources or jobs in the field of production, transportation and logistics. A) A set of cyclic sequences from input data which is a set of resources or jobs. A first step of generating, and B) for all cyclic sequences in the created cyclic sequence set, a second step of calculating an evaluation value using an evaluation function, an evaluation weight, and a normalization coefficient, C)
A third step of determining a search condition by using an update strategy from among a plurality of cyclic sequence update means and determining an update means; andD) using some of the cyclic sequences in the cyclic sequence set using the update means. A cyclic sequence search step including a fourth step of updating the sequence to another cyclic sequence, and E) a fifth step of calculating a normalization coefficient effective for the cyclic sequence search step from the input data and the evaluation function. And a cyclic sequence scheduling method.