JPH10293784A - Scheduling method and device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform flexible solution preparation for the change of schedule conditions and schedule targets by merging a method for generating and adjusting resource allocation for an operation and the method for dissolving the time-based contention of the using resources of the operation in a stochastic search technique. SOLUTION: The stochastic adjustment 0106 of the resource allocation receives data for allocating resources to the respective nodes of the directed network of an operation kind as input and changes the combination of the allocation resources to the more desirable one by using the stochastic search method of a combination problem represented by genetic algorithm. At the time, a solution search parameter 0110 is used as the control data of a stochastic search. The dissolution 0107 of resource contention selects an operation procedure, makes the operation concrete and prepares a solution candidate. Then, the schedules of respective works are independently prepared in the initial stage of preparation and the generated resource contention is dissolved by the change of the operation procedure and the time allocation of the operation.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

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 field of production and transportation by a computer, and more specifically to a method of planning a plan using a computer, a method of displaying a plan result, The present invention relates to a method of correcting a plan result and a method of verifying the plan result.

[0002]

2. Description of the Related Art As to a scheduling method using a stochastic search technique, as described in (1) "Job Shop Scheduling Problem Solving Method" (Japanese Patent Laid-Open No. 5-225203), a number of resources assigned in advance are used. If more than one task is given,
There is known a method of determining the execution order between tasks sharing the same resource as scheduling, and realizing a solution search using a genetic algorithm (GA).

[0003] Regarding a method of preparing a vehicle replacement plan in the transportation field, see (2) "Replacement total system" (Ansai et al .: Railway and Electric Technology, Vol. 6, No. 2, 199).
As described in 5), there is known a method in which the know-how of vehicle replacement plan creation is extracted as an AI rule and reflected in the plan creation.

[0004]

In the prior art (1), the solution search is performed by the GA, taking the time contention of the used resources and assigning the execution order of the work as scheduling. Resource allocation must be provided manually by the planner. Furthermore, as a result of evaluating the created solution, if the solution does not sufficiently reflect the intentions of the plan creator, or if it is impossible to put the order in which it can be executed, it is necessary to allocate resources according to the content. May need to be changed, but in the above-described conventional example, there is a problem that the degree of solution evaluation cannot be reflected in resource allocation. Further, in the above conventional example (1), since the auxiliary processing required to execute the work is not a scheduling target, even if the created schedule is correct in the execution order of the work, the auxiliary processing is not performed. Since the feasibility is not guaranteed, there is also a problem that the feasibility is not practical.

In the above conventional example (2), a vehicle replacement plan at a specific railroad depot is created by search logic incorporating planning conditions and plan targets specific to the depot, and these are therefore required. If it changes, there is a problem in the maintenance and development of the scheduling system that it becomes difficult to search for a feasible solution and cannot be applied to the preparation of a vehicle replacement plan for another depot.

[0006] An object of the present invention is to identify work and auxiliary processing necessary for execution of work as operations, and to solve the production / traffic scheduling problem by referring to the operation procedures required for completing the work and the resources used by each operation. And to determine the processing time (execution order) of each operation, a method of generating and adjusting the resource allocation for the operation and a method of resolving the time conflict between the resources used by the operation in the stochastic search technique It is an object of the present invention to provide a production / traffic scheduling system that can flexibly create solutions for changes in planning conditions and planning goals by fusing.

[0007]

The object of the present invention is to generate, in a step of generating a constraint on an operation procedure from input work data, a constraint on an operation procedure having a directed network structure connecting nodes indicating operation types. In the step of stochastically adjusting the combination of resource allocation of the operation, in the step of generating and adjusting the number of chromosomes set in the solution search parameters, indicating the status of resource allocation of each node of the constraint on the operation procedure for all tasks, In the step of resolving the time conflict between the resources required for the operation and creating a solution candidate for the schedule, an initial solution is created from constraints on the operating procedure and the chromosome, and either one of the operations competing for resources used in the initial solution Is shifted backward in time while satisfying the constraints on work set between work operations, thereby resolving conflicts and creating solution candidates. In the step of calculating the evaluation value by evaluating the solution candidates of the obtained schedule, calculate the evaluation value of each of the solution candidates according to the degree of satisfaction of the solution evaluation items created from the predetermined constraints, In the step of feeding back the evaluation value to the above-mentioned stochastic adjustment step of resource allocation, the calculated evaluation value is given to the chromosome data used in creating the solution candidate, and the optimal solution search processing of the schedule candidate is performed by (1) resource allocation. (2) creating a solution candidate for a schedule by resolving time conflicts between resources required for operation, and (3) calculating an evaluation value by evaluating the solution candidates. (4) repeating the step of feeding back the evaluation value to the stochastic adjustment step of resource allocation, and displaying the solution of the schedule obtained as a result of the optimization. In step operated by an operator to work data, constraints on resources, the solution search parameters, constraints on the order of operations, to modify the solution evaluation items is accomplished by instructing the re-execution of the solution search.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A description will now be given of a vehicle replacement plan creation system in a railway depot as an embodiment of a scheduling method and apparatus according to the present invention.

The present embodiment is a system for supporting a vehicle operation planning operation at a railway depot, in which all vehicles (hereinafter referred to as trains) entering the depot are provided before a predetermined departure time. A transfer schedule of the train set in the vehicle depot is created so that predetermined cleaning, inspection, repair, and other operations can be performed.

FIG. 1 is a diagram showing a basic configuration of the present invention.
Before starting a detailed description of the embodiment, an outline of the present invention will be described with reference to FIG.

The scheduling method and apparatus realized by the present invention are used to create a work schedule. Here, the work is a group of a series of work for achieving a certain purpose, and the start time and the end time of the work are determined in advance. The purpose of the task will vary depending on the specific scheduling problem addressed by embodiments of the present invention. Taking the problem of processing schedule creation of a product in a production factory as an example, one job is to process a raw material through a predetermined processing process to produce a specific product. In this case, the start time of the work corresponds to the time at which the raw materials can be charged, the end time corresponds to the delivery date of the product, and the work corresponds to each step in the processing process. The work data 0101 includes at least information on the start time and end time of each of a plurality of works and information on required work. The scheduling method and apparatus implemented by the present invention processes this work data and creates an executable "operation sequence" for each work that satisfies a plurality of planning conditions and user requirements at a high level. An operation is any processing that is not further divided in a scheduling target. Work at work is also a type of operation. Further, auxiliary processing necessary for performing the work also corresponds to the operation. Regarding the problem of creating the product processing schedule in the above-mentioned production factory, the process of transferring semi-processed products during each process, the process of storing the semi-finished products until the start time of the next process, the setup process for process start, etc. Corresponds to this. The operation is embodied by defining at least three points: a start time of execution, an end time, and resources used. The resources used are the physical, human, and spatial objects required to perform the operation, such as raw materials, machines for processing and transport, people who move the machines, storage spaces and transport routes, etc. Information on resources that can be used for each operation type is defined in resource constraints 0102. Also, for example, the transfer route of a semi-finished product between a certain process may be restricted by the physical arrangement of the machines used in the previous process and the subsequent process. You may need to take advantage of relationships. Such relational information between resources is also included in the restrictions on resources.

The operation sequence of a work is a series of operations required to complete a certain work, and by performing the operations in order along this flow, the work can be completed without any trouble. . Hereafter, the work schedule,
Or, when they are described as solutions to the scheduling problem, they mean a work operation sequence.

[0013] The scheduling method and apparatus realized by the present invention are provided after receiving work data as input.
In step 0103, restrictions 010 on the operation procedure
4 is generated. The constraint on the operation procedure is information that defines the type of operation required to complete a given task and the connection between the plurality of operations. In the case of a simple scheduling problem in which it is not necessary to consider any auxiliary processing, the restriction on the operation procedure is a simple list in which the types of work included in the work are arranged in the order of execution as shown in FIG. . However, the auxiliary processing between operations is different from the operation strictly defined for each operation. For example, the completed product is the same even if the delivery route and method of the semi-finished product between processes are different. In many cases, there are a plurality of processes that provide the same result, and there is often a degree of freedom in their selection. For this reason, restrictions on the operation procedure generally result in a complicated directed network having operation types as nodes as shown in FIG.

On the directed network, one path is obtained by tracing all the work nodes in the execution order, and each node of the path is embodied as an operation so as not to violate the time constraint. It is a schedule. FIG. 3 shows an example of a Gantt chart that is a visual representation of a schedule.
The vertical and horizontal axes of the Gantt chart are the resources and time zones in the scheduling environment, respectively. A thick bar in the table including the horizontal bar 0301 indicates one operation, and the vertical position of the bar indicates the resource used by the operation, and the horizontal position and the length of the bar indicate the execution time zone.

If there is only one job to be scheduled, all the schedules obtained in this way are executable solutions. However, when multiple tasks that share resources and overlap in time are given, even if the schedules of individual tasks are created independently, time conflicts in the resources used occur between operations, and execution is possible. It is difficult to get a solution. To create a feasible solution without resource conflicts,
It is necessary to appropriately select the operation procedure and embody the operation (allocation of resources and time) while associating them with each other. In the present invention, this is realized by the solution search processing of step 0105 in FIG.

This step basically has a loop connection of four internal steps. The stochastic adjustment step (0106) of the resource allocation receives, as an input, data in which resources are allocated to each node of the directed network of the operation type, and performs a stochastic search method of a combination problem represented by a genetic algorithm (GA). To change the combination of allocated resources to a more desirable one. At this time, the solution search parameter 0110
Is used as control data for the probabilistic search. In the resource conflict resolution step (0107), a solution candidate is created by selecting an operation procedure and embodying the operation. The resources used for the operation use those adjusted in the above step. In the initial stage of the creation, the schedule of each task is created independently, and the resource conflict that has occurred is resolved by changing the operation procedure and time allocation of the operation. When an execution order is defined between specific tasks of different tasks, the information is taken in as a constraint 0111 regarding the order of the tasks and used as a resource conflict resolution rule. The solution candidate evaluation step (0108) includes a solution evaluation item 01 for the solution candidate obtained in the conflict resolution step.
12 and the satisfaction degree of the user request is determined, and an evaluation value corresponding to the satisfaction degree is given. The evaluation value is sent to the resource allocation stochastic adjustment step in the evaluation value feedback step (0109), and is used for optimizing the resource allocation. By repeating the above four-step loop process, the combination of the allocated resources is gradually optimized,
Finally, a high quality feasible solution can be generated.
In the solution search step, after this loop process is repeated by the specified number of loops included in the solution search parameter, the finally obtained solution candidate is output as the solution 0113.

At step 0114, the solution is displayed on the display device 011.
6 and accepts an operation by the operator from the input device 0115. The operator outputs work data, resource constraints, solution search parameters, work order constraints, and solution evaluation items to a display device, edits them, and instructs solution search processing again using the edited data. Can be. The solution approved by the operator is registered as a formal plan, converted to a business format, and output from the printer 0117.

This embodiment is realized by embodying all the data, processes and devices shown in FIG. 1 for preparing a vehicle replacement plan. Hereinafter, the details will be sequentially described while clarifying the correspondence with FIG.

First, an outline of a vehicle replacement plan at a vehicle depot, which is a plan target of the present embodiment, will be described with reference to FIGS. 4, 5, and 6. FIG.

FIG. 4 is a part of a form showing a vehicle replacement plan at a virtual railway depot. Each row of the form represents one turn of the train in the depot. Here, the turning line refers to a movement from a track in the depot to another track. For example, in the case of the transfer line 0401, the knitting 02
It indicates that the vehicle moves from 7 to track 05 at 9:30.
Furthermore, from "contents" and "work details", this line
5, it can be seen that the operation is for receiving the operation "cleaning 1". Each transit line is arranged in the order of the start time, and by performing the transit in accordance with this order, all trains entering the base will receive scheduled cleaning, inspection and repair work. It is possible to move one after another to the next line and to leave the depot at the scheduled departure time without colliding with each other while moving or on the train.

FIG. 5 is a layout diagram of the vehicle depot. It has seven lines and one entry / exit line. Track 01
Is divided into two sections, and different formations can be present in each section. There is a special restriction that entry into and out of the 2nd ward is limited to those where the train is not located in the 1st ward.
In the following description, wards are treated as line numbers unless otherwise specified. The knitting can move between another track or outside of the depot via a transit route connected between the track and the entrance / exit line. The depot shown in FIG. 4 has a simple structure in which all the tracks and the outside of the depot can be moved by one turn route for easy understanding, but can be moved only between specific tracks. The present embodiment also corresponds to a structure having a plurality of transfer routes.
The same applies when there are a plurality of entry / exit lines. In general, the number in the depot including the depot shown in FIG.
Can be classified into types. (1) Work line: A line with dedicated facilities that can perform cleaning, inspection, repair, etc. (2) Detention line:
Waiting for the availability of a work line or leaving the work line (3) Pulling work line: a work line that is temporarily present to move between work lines, a work line that can be cleaned, a checkable line, etc. It is further classified according to the type of work.
Since different operations may be possible in the same facility, the possible lines of each operation may overlap. The indwelling track does not require any special equipment, and therefore most of the base stations correspond to this. There is also a depot that uses a part of the work track as a detention track, and in this case, the set of the work track and the set of the detention track overlap. Taking the depot of FIG. 4 as an example, there are two pull-up lines, line 06 and line 07. As can be seen from the layout, the other lines are line 06 and line 0
When going to another track other than 7, be sure to once
You must "pull up" to line 07 and then move to the destination line. In other words, excluding the entry and exit sections, one of the originating line and the destination line is always an upline. Therefore, since the pulling line is used frequently, it is hardly used for other purposes such as detention, and the on-line time is extremely short.

FIG. 6 shows a part of a vehicle replacement schedule obtained by converting the vehicle replacement schedule shown in FIG. 4 into a diagram format. The horizontal axis is the time of the day, and the vertical axis is the line in the depot. Horizontal line 06
A horizontal line such as 01 indicates a line in each line, and a hatched line connecting horizontal lines such as a hatched line 0602 indicates movement between the lines. The appearance of a particular train moving around the depot between entry and exit can be easily determined by following a single line consisting of a diagonal line and a horizontal line that starts at the entry and exit of that train and ends at the entrance and exit. Recognize.
Further, in the vehicle replacement schedule, since the collision at the time of movement and the collision at the time of on-line are visually expressed as a time overlap of the oblique line and the horizontal line, respectively, it is easy to detect them.

To summarize the above, the problem of preparing a vehicle replacement plan is based on the consideration of various constraints on the sequence of movements / rails up to departure of all trains entering a vehicle depot having a specific layout. It can be said that it is to create while. Constraints to be considered can be classified into two types: essential constraints and target constraints. The indispensable constraints are constraints that must be satisfied by all feasible vehicle replacement plans. (1) Keep the departure time of each train. (2) Compete for track numbers when there are trains, and transfer when moving. There is no route conflict (3) There are three points where the scheduled work is on the executable line for a specified time or more and the presence order is the same as the work order. The target constraint is a constraint that is weaker than the essential constraint. Even if the target constraint is not satisfied, as long as the essential constraint is satisfied, the feasibility of the vehicle replacement plan is not impaired. The target constraint can be regarded as a constraint condition relating to the quality of the vehicle replacement plan, and the higher the degree of satisfaction of the target constraint, the higher the quality of the vehicle replacement plan. Further, there are many types of target constraints, and it may not be possible to fully grasp all of them when developing a system. Further, the target constraint dynamically changes due to a change in the depot characteristics such as the depot layout and various facilities in the depot.
Therefore, only the goal constraints considered to be universal to some extent are listed here.

1. 1. The number of turns is small. 2. Each train has a large margin. Keeping the work order The fact that the total number of turns for all the formations is small means that the cost of taking the vehicle replacement as one work in total is correspondingly small. The allowance is a free time from completion of all the scheduled operations to departure. If the allowance is large, it is easy to respond when the scheduled departure time is advanced due to disorder of the train schedule, and it is also possible to use the available time to receive other work A robust plan against dynamic changes in the environment. The work order (3) does not refer to the execution order of a plurality of works scheduled for each knitting, but the execution order set between works of different knittings. , Organization 0
3, in the order of knitting 08 ".

In addition to the above-mentioned target constraints, "the variation of the margin of each knitting is small", "the movement distance of the workers between track numbers is small", "the use of track numbers is not biased", and Keep your priority. "

By applying the present invention, a plan creation system that solves the above-mentioned vehicle replacement plan creation problem can be realized. Hereinafter, details of the implementation of each data and step in FIG. 1 will be sequentially described.

First, the work data shown in FIG. 0101 will be described. One task in the vehicle replacement plan creation problem is a group of scheduled tasks to be performed from entry to exit of one train. FIG. 7 shows a specific example of the work data. In this example, the job data is composed of a plurality of jobs. Among them, the job 01 has an organization number of 03, an entry time of 9:36, a departure time of 11:03, and three jobs in between. Are scheduled, and the time required for each operation is also given in advance. Using this work data as input, a constraint on the operation procedure is generated in step 0103. The operation in this embodiment is classified into one of the following types.

(1) Work operation (2) Detention operation (3)
Pulling-up operation (4) Movement operation The work operation means the presence of a train line that is longer than the required time on a line where the work scheduled for each train can be performed. The work is further divided into, for example, cleaning, inspection, and repair, and the work operation is also classified according to each. The resource used by the work operation is a line on which each work can be performed. The detention operation means the presence of a line while waiting for an empty work line. The resource used by the detention operation is one of the lines set as the detention line. The pulling-up operation means that the line has been on the track set as the pulling-up line for more than the required time.
The movement operation literally means a movement between the track numbers or between the track number and the outside of the depot via the entry / exit line, and the used resource is a transit route in the depot. The work operation is an indispensable operation for completing the work, and other operations (2)
(3) and (4) correspond to auxiliary processing for performing the work. FIG. 8 shows the data structure of the operation used in this embodiment. Each operation data has an operation number, a job number,
It includes six items of information: operation type, start time, end time, processing time, and resources used. When the operation type is work, the operation type contains the work number of the work. In order to know the detailed information of the work type, the work indicated by the work number is extracted from the work data, and the item of the work content is accessed by the work number. In the following description, the terms entry operation and departure operation may appear, but these mean the movement operation at the time of entry and departure, and unless otherwise specified, the normal movement operation Treat as

An operation sequence that embodies the above operations and arranges them in an appropriate order is a work sequence (a group of works scheduled from entry to departure of the organization). FIG. 9 shows the data structure of the operation sequence according to the present embodiment. Each operation sequence is assigned a different sequence number to distinguish it from the others. The main content of the operation sequence is a reference list for the operation data 0901. Other additional information includes the job number of the job to be scheduled, the number of operations in the operation list, and the margin of the operation sequence.

Using the work data shown in FIG. 7 as an input, in step 0103 in FIG.
FIG. 10A shows a case where the work is one work as an example of a restriction on the operation procedure in the vehicle replacement plan creation problem. On a directed network of this operation type, any one path including a work node from an entry node to an exit node can be executed. The simplest operation procedure in this example is a path consisting of only three nodes: entry, operation, and exit. With respect to this operation procedure, a more complicated operation procedure can be obtained as an on-line operation sequence for detention is inserted between the entry and the operation and between the operation and the exit. Unless the departure time is taken into consideration, there is no limit on the number of on-line operation sequences for insertion that can be inserted, and in that case, there are an infinite number of possible operation procedures. However, in vehicle replacement work, it is desirable that the number of turns is as small as possible from the viewpoint of work efficiency and burden, and it is possible to avoid conflicts between turn routes and track lines only by actually performing detention at most once before and after work. In most cases, FIG.
A finite network structure as shown in FIG. When this network is used, the simplest operation procedure is a path consisting of only the above-mentioned three nodes, the entry node, the operation node, and the exit node. This is a path that includes an operation sequence that performs the on-rail line once. In the following description, the former is referred to as the best pattern of the operation procedure, and the latter is referred to as the worst pattern of the operation procedure. As can be clearly inferred from the figure, all operating procedures have the property that the best pattern is included as a partial network and the worst pattern is included as a partial network. This means that it is possible to compose all operating procedures from the best pattern and the worst pattern. Therefore, the data structure of the constraint on the operation procedure used in the present embodiment is shown in FIG. Each data is given a different number to distinguish it from the others, and includes the work number of the target work as additional information. The main contents are a list of the worst patterns of 1101 and 110
2 is a list of the best patterns, and each list stores data representing an operation type. In addition, in order to store the positions of the working nodes common to both patterns, those pairs are stored as the same resource operation pair. In the following description, the data of this data structure will be referred to as an operation sequence template or simply a template.

In this embodiment, step 0103 in FIG.
The process corresponding to the generation of the constraint in the operation procedure is the operation sequence template generation process. This processing will be described with reference to the flowchart of FIG. First, in step 1201, the number m of templates is initialized. In step 1202, a termination determination is made.
If the number m of templates exceeds the number of jobs in the job data, the process ends. In step 1203, the position w on the worst pattern, the position b on the best pattern, and the number s of operation pairs (work nodes) having common resources are initialized. Step 1
At step 204, the number of the job m is set as the job number of the template. In step 1205, the operation type "entrance" is set at the head position of the worst pattern and the best pattern. In step 1206, a work counter n is initialized. In step 1207, the end of the loop processing is determined. When the work number counter n exceeds the number of works of the work m, the loop processing is ended.
Is incremented by 1 and the process returns to the end determination of the processing in step 1202. To continue the loop processing, step 120
Proceeding to step 8, the operation types “detention”, “move”, “pull up”, and “move” are sequentially set while increasing the position w with respect to the worst pattern. These represent a series of operation sequences for staying on the indwelling track. Next, the worst in step 1209,
For the best two patterns, the operation types “work of work number n of work m”, “move”, and “pull up” are sequentially set while increasing the positions w and b. And step 121
At 0, the pair of positions where the work was set in the immediately preceding step is registered as the same resource operation pair, and s is increased by one. The above is the processing in one loop.
In step 11, the work number counter n is incremented by 1 and step 1207 is executed.
The process returns to the loop process end determination. After exiting the loop processing, the process proceeds to step 1212. In this step, the operation types “detention” and “departure” are set while increasing the position w for the worst pattern. Next, in step 1213, the operation type "departure" is set by reducing the position b by 2 with respect to the best pattern. At the end of this process, one template is completed. Thereafter, the number m of templates is increased by 1, and the process proceeds to the end determination step 1202. Through the above processing, an operation sequence template corresponding to each job in the job data is generated.

The operation sequence template is the main input data for the solution search processing in step 0105 in FIG. Prior to the description of this step, the details of the resource-related constraint of FIG. 1 and the implementation of the solution search parameter of FIG. First, restrictions on resources will be described. The resources used by the operation in the vehicle replacement planning problem are either the track lines in the depot or the transit routes. Information on what kind of operation each resource can be used is defined as a resource attribute table. FIG. 13 shows the data structure. Each row of the table represents a list of available resources for each operation type. Since each resource can be used for multiple purposes, the same resource may be stored redundantly in different rows. However, the resources of the lifting operation and the moving operation are not used for other purposes. Depending on the layout of the depot, there may be a case in which a turn route used for movement between specific tracks is limited, and it is necessary to use the connection information of the turn / turn route to generate an executable schedule. . Therefore, in addition to a resource attribute table, a connection determination table that defines a relationship between resources is adopted as a constraint on resources. FIG. 14 shows the data structure. The vertical axis of the table is a turning route in the depot, and the horizontal axis is the same number line. In each item of the table, a value indicating whether the line of the vertical position is connected to the line of the horizontal position, that is, whether or not the line can be moved through the line of the line is stored. Value TR
The UE indicates that connection is possible, and the value FALSE indicates that connection is not possible.

Next, the solution search parameters will be described.
In the present embodiment, a genetic algorithm (GA), which is one of stochastic search methods, is used as a basic architecture of a solution search process. The GA regards solution candidates of the combination problem as biological individuals, and performs a genetic operation on chromosome information that determines the traits of each individual in a generation alternation cycle of the individual population, thereby searching for a high-quality individual population.
Solution search parameters are used to control this search. FIG. 15 shows the solution search parameters in this embodiment. The parameters used were "number of generations", "size of individual population", "natural selection rate" during genetic operation, "mutation rate", and "extended generations" used when re-running the solution search. . These are all commonly used in GA applications.

The solution search processing in step 0105 in FIG. 1 will be described. As described above, GA is adopted as the basic architecture of the solution search processing. GA is a general term for a framework of a solution search method for a combination problem, and detailed specifications for each are not given in advance. When applying the GA, it is necessary to appropriately realize the components for the target problem. Hereinafter, details of the realization will be sequentially described with reference to the flowchart of FIG.
It shows the correspondence between the steps of "probabilistic adjustment of resource allocation", "resolution of resource conflict", "evaluation of solution candidate", and "feedback of evaluation value", which are the internal steps of the solution search process. . First, in step 1601, FIG.
The resource allocation table shown in (1) is initialized. The resource allocation table is data in which resources are allocated to each element of the worst pattern and the best pattern of all the operation sequence templates to be scheduled. The generation of the resource allocation table corresponds to a process of allocating resources to each node of the directed network representing a restriction on an operation procedure. In the solution search process, in this step, the resource allocation table is initialized by a number equal to the number of individuals in the solution search parameter. In the subsequent processing, this resource allocation table is regarded as a GA chromosome, and its contents are adjusted stochastically within the framework of GA. Next, in step 1602, the generation number counter m is initialized. In step 1603, the end of the process is determined. If the generation counter m exceeds the number of execution generations set by the solution search parameter, the process ends. In step 1604, the contents of the chromosome (resource allocation table) are adjusted stochastically. This step corresponds to the resource allocation stochastic adjustment processing of step 0106 in FIG. This step is a general genetic operation of GA as an internal step, natural selection (1605),
Including three processes of crossover (1606) and mutation (1607). These processes do not necessarily indicate a specific method in the definition of GA, but need to be implemented individually for each GA application target or chromosome structure. The details of the realization will be described in detail after a brief description of the solution search process. Next, in step 1608, the chromosome pointer n is initialized. In step 1609, the loop processing is determined to be completed. If the chromosome pointer n exceeds the number of individuals set in the solution search parameter, the loop processing is completed. In step 1610, the generation number counter m is incremented by one, and the processing proceeds to step 1603. The process returns to the end determination of the process. If the loop processing is to be continued, the process proceeds to step 1611, where the n-th chromosome is used as input data to generate a solution, that is, a work operation sequence. This step corresponds to the resource conflict resolution process of step 0107 in FIG. 1 for selecting and embodying the operation procedure. Next, in step 1612, the planning conditions for the solution generated and the degree of satisfaction of the user request are evaluated.
This step corresponds to the solution candidate evaluation step of step 0108 in FIG. Thereafter, the evaluation value given to each solution in the evaluation step is fed back to the chromosome in step 1613. This is one loop processing, and one solution candidate is generated. Step 161
In step 4, the value of the chromosome pointer n is increased by 1 and step 16
It returns to the end determination of the loop processing of 09.

In the following, details of the steps of configuring the solution search processing, ie, the step of initializing the resource allocation table (chromosome), the step of stochastically adjusting the chromosome, the step of generating a solution (operation sequence), and the step of evaluating the solution are described. Will be described sequentially.

The process of initializing the resource allocation table (chromosome) will be described with reference to the flowchart of FIG. First, in step 1801, the individual number m is initialized.
At step 1802, it is determined whether the process is completed. If the individual number m exceeds the number of individuals set in the solution search parameters, the process is completed. If the processing is not completed, the flow advances to step 1803 to set the template number n
Is initialized. At step 1804, it is determined whether the loop processing has ended. If the template number n exceeds the number of templates, the individual number m is incremented by 1 at step 1805, and the processing returns to the processing end determination step 1802. If the loop processing is not completed, the flow advances to step 1805 to allocate resources to the worst pattern of template n. Details of this processing will be described with reference to the flowchart of FIG.

First, in step 1901, the position m on the template is initialized. At step 1902, it is determined whether the processing is completed. If the position on the template exceeds the number of operations of the worst pattern, the processing is completed. If the processing is not ended, the process proceeds to step 1903, and the value of the position m of the worst pattern is set in the operation type variable k. In step 1904, it is determined whether or not m is the head of the pattern. If m is the head, the process proceeds to step 1905, where a resource of the operation type k is randomly selected and set to the position m while referring to the resource attribute table. If m is not the head, a resource that can be connected to the resource of the operation type k and the position m−1 is randomly selected and set to the position m while referring to both the resource attribute table and the connection determination table.
In step 1907, m is increased by 1 and the process returns to step 1902 for determining the end of the process.

Returning to FIG. 18, the description of the initialization processing of the resource allocation table will be continued. After step 1805, step 1
At 806, resources are allocated to the best pattern of template n. In this step, only the processing target is changed from the worst pattern to the best pattern.
Exactly the same as 05. Step 1805 and Step 1
Although resources are allocated to both the worst and best patterns by 806, different resources may be allocated to both patterns for the same working node.
Therefore, in step 1807, the resources at the positions where the same work nodes of the worst pattern and the best pattern are stored are made the same. This is one loop processing, and the resource allocation for one template is completed.
In step 1808, the value of the template number n is incremented by 1, and the process returns to step 1804 to determine whether the loop processing has been completed.

Next, the chromosome stochastic adjustment processing in step 1604 in FIG. 16 will be described. This process consists of three internal steps. First, in the first step 1605, a natural chromosome selection process, which is one of the genetic operations of GA, is performed. Using the evaluation value given to the chromosome by the feedback of the evaluation value in step 1613 of FIG. In the initial generation of the search, no evaluation value is given to the chromosome. In this case, processing is performed assuming that all chromosomes have the same evaluation value. The following crossover processing,
In the mutation process, the same process is performed in the initial generation.

In the crossover process in the next step 1605, new chromosomes are generated and supplemented by the number of chromosomes reduced by the natural selection process. The generation of a new chromosome is realized by arbitrarily selecting two chromosomes from a set of chromosomes and synthesizing both elements (this is called a gene). When a chromosome is a resource allocation table, the element is a resource, which is a gene. In the following description, the source chromosome is referred to as a parent chromosome, and the newly generated chromosome is referred to as a child chromosome.

The process of generating the child chromosome C from the arbitrarily selected parent chromosome A and parent chromosome B will be described with reference to the flowchart of FIG. First, in step 2001, the template number m is initialized. In step 2002, it is determined whether the process is completed. If the template number m exceeds the template number, the process is completed. If the processing is not to be ended, step 2003
Then, the worst pattern of the template m of the child chromosome is generated from the parent chromosome A and the parent chromosome B. The details of this step will be described in detail with reference to the flowchart in FIG.

First, in step 2101, the position m on the worst pattern is initialized. In step 2102, an end determination is made. If the position m exceeds the length of the worst pattern, the process ends. If you do not want to end the process,
Proceeding to step 2103, the values of the position m on the worst pattern of parent A and parent B are stored in resource variables a and b, respectively.
In step 2104, it is determined whether or not m is the head of the pattern.
A or b is set at the position m on the worst pattern of the child chromosome. Here, the probability Pc is a value that is dynamically determined so that the possibility that a child chromosome inherits a parent gene having a higher evaluation value reflects the degree of its superiority. The formula for calculating the probability Pc when the parent chromosome A is excellent is shown below. Pc
= (Evaluation value of parent chromosome A) / ((Evaluation value of parent chromosome A)
+ (Evaluation value of parent chromosome B)) After the end of step 2105, the process proceeds to step 2106, where the value of the position m is incremented by 1, and the process returns to the end determination of step 2102. If the position m is not the head, step 21 In this case, the process proceeds to step 2109, where a or b is set to the position m of the worst pattern of the child chromosome with the probability Pc. After the end of step 2109, proceed to step 2106 It is set to the position m of the worst pattern of the body. Step 211
After 1 is completed, the process proceeds to step 2106, where the value of the position m is increased by 1, and the process returns to step 2102 to determine whether the process is completed. If the determination in step 2110 is false, step 2112
Again to determine the connection Then, a is set to the position m of the worst pattern of the child chromosome. After the end of step 2113, the process proceeds to step 2106, where the value of the position m is incremented by 1, and the process returns to the end determination of step 2102. If neither is connectable, the process proceeds to step 2114, where a or b is set to the position m of the worst pattern of the child chromosome with the probability Pc. After the end of step 2114, the process proceeds to step 2106, where the value of the position m is incremented by 1, and the process returns to the end determination of step 2102.

Returning to FIG. 20, description of the process of generating a child chromosome will be continued. After the end of step 2003, the best pattern of template m of the child chromosome is generated from parent chromosome A and parent chromosome B in step 2004. This step is exactly the same as step 2003 except that the processing target is changed from the worst pattern to the best pattern. Steps 2003 and 2004 generate the worst and best patterns of the template m of the offspring chromosome.
In this state, different resources may be allocated to both patterns for the same work node. Therefore, in step 2005, the resources at the positions where the same work nodes of the worst pattern and the best pattern are stored are made the same. In step 2006, the value of the template number m is increased by 1, and the process returns to step 2002 for determining the end of the process.

Returning to FIG. 16, the mutation processing in step 1607 will be described. Mutation is a process of randomly changing the genetic code of a chromosome, thereby preventing a solution search from falling into a local solution. At this time, if the gene is changed completely at random, the possibility that a feasible solution cannot be obtained becomes very high. Therefore, in order to prevent this, it is necessary to select an appropriate resource (gene) while referring to a resource allocation table and a connection determination table, which are restrictions on resources. The details of this step will be described with reference to the flowchart in FIG.

First, in step 2201, the template number m is initialized. In step 2202, it is determined whether the process is completed. If the template number m exceeds the template number, the process is completed. If the processing is not completed, the process proceeds to step 2203, and the worst pattern of the template m is mutated. Figure 2 shows the details of this step.
This will be described in detail with reference to the flowchart of FIG.

First, in step 2301, the position n on the worst pattern is initialized. In step 2302, it is determined whether the processing is completed. If the position n exceeds the length of the worst pattern, the processing is completed. If the process is not to be terminated, the process proceeds to step 2303, and the integer variable v is set from 1 to 10
Values up to 00 are stored at random. Step 2304
Then, the magnitude of v is compared with the mutation rate set by the solution search parameter. If v is larger than the mutation rate, the value of the position n is incremented by 1 in step 2305, and the process returns to the end determination 2301 of the process. If v is smaller than the mutation rate, the process proceeds to step 2306, where the operation type variable k includes the template m
The value of the position n of the worst pattern is set. Step 23
At 07, it is determined whether the position n is the head of the pattern.
If the position n is not the first position, the process proceeds to step 2308, and by referring to both the resource attribute table and the connection determination table, a resource which is of the operation type k and which can be connected to the resource at the position n-1 is selected at random. Set to n. If the position n is the top, the process proceeds to step 2309, where a resource of the operation type k is randomly selected and set to the position n while referring to the resource attribute table. After the end of step 2308 and step 2309, the value of the position n is incremented by 1 in step 2305, and the process returns to the end determination step 2305 of the process.

Returning to FIG. 22, the description of the mutation process will be continued. After step 2203, the process proceeds to step 2204, where the best pattern of the template m is mutated. In this step, only the processing target is changed from the worst pattern to the best pattern, and the processing content is step 2203.
Is exactly the same as Step 2203 and Step 220
4, the worst and best patterns of the template m of the chromosome are mutated, but different resources may be allocated to the same working node for both patterns. Therefore, in step 2205, the resources at the positions where the same work nodes of the worst pattern and the best pattern are stored are made the same. In step 2206, the value of the template number m is incremented by 1, and the process returns to the end determination step 2202 of the processing.

This concludes the description of the chromosome stochastic adjustment step 1604 in the solution search processing of FIG. Next, the solution generation step 1611 will be described.

First, the basic concept of the processing in this step will be described below.

1. The operation sequence of each task is initialized independently while taking a margin as large as possible.

2. If there is an operation pair that causes resource conflict or violation of the order constraint between work operations in the initial solution, the following two processes of (a) change of operation time allocation (shift of operation) and (b) change of operation procedure are appropriate. To solve the problem.
However, the detention operation is excluded from the resource conflict resolution target.

3. When a shift operation is selected (operation ordering) from an operation pair competing for resources, (a) no new order constraint violation occurs between work operations, and (b) the margin of each operation sequence becomes equal. This is done while considering two points. That is, (a)
(B) is used as a resource conflict resolution rule.

4. If resource conflicts that cannot be resolved by (1) to (3) and order constraint violations between tasks remain,
In the subsequent execution generations, the resolution is abandoned in the hope that a resolvable resource allocation combination will be generated by the chromosome stochastic adjustment step 1604 in FIG.

Even when there is a resource conflict regarding the detention operation, no processing is performed in this step from the same idea as described above.

Next, the details of this step will be described with reference to the flowchart of FIG. After using the flowchart to explain the overall processing flow, the configuration steps that need to be explained in more detail will be described.

First, in step 2401, the operation sequences of all tasks are independently initialized. Step 240
In step 2, the order constraint violation is eliminated for the initial solution.
In step 2403, a Gantt chart is constructed from the operations of all the operation sequences. FIG. 25 shows the data structure of the Gantt chart used in this embodiment. The basic elements are references to operational data, which make up a list for each resource used. Each list is always ordered by the start time of the operation. Returning to FIG. 24, the description will be continued. In step 2404, a set of resource conflict operation pairs is created using the Gantt chart. In step 2405, the margin of each operation sequence is calculated. If the margin is a negative number, that is, if there is at least one organization that exceeds the leaving time, the code TRUE is set to the flag OutOver, and if not, the code is set to TRUE. Set FALSE. Step 2406
Then, it is determined whether the number of resource conflicts and the number of work order violations are both 0 or the flag OutOver
Is TRUE, the process ends. If the processing is not to be ended, the process proceeds to step 2407, where the shift start time is compared between the pair of resource conflicting operations having the earliest shift start time and the order constraint violating operation as well. If the shift start time of the resource conflict operation pair is earlier, the resource conflict is resolved in step 2408. If the shift start time of the order constraint violation operation pair is earlier, the order constraint violation is resolved in step 2409. After ending either of the above steps, the process proceeds to step 2410 to store the shifted operation in the operation variable p. In step 2411, the resource conflict that occurred before p in the operation sequence is resolved. Next, the processing of generating the resource conflicting operation pair set in step 2404 is performed again. In step 2412, the satisfaction degree of the order constraint expression is determined, and the process returns to step 2405.

The above is the outline of the process of generating a solution (operation sequence). Next, details of the configuration steps that need detailed explanation will be described in order. In the flowcharts used in the following description, a reference symbol “.” May be used when referring to an internal element of data. For example,
If you write "template m. Best pattern"
Points to the best pattern of template m.

First, the operation sequence initialization processing in step 2401 in FIG. 24 will be described. When each operation sequence is created independently without considering resource conflicts and violation of work order constraints, it is obvious that the operation procedure with the largest margin is the best pattern with the least number of operations. Therefore, in this step, each operation of the best pattern of the operation sequence template is materialized to generate an initial solution. Details of the processing in this step will be described with reference to the flowchart in FIG.

In step 2601, the template number m is initialized. At step 2602, it is determined whether the processing is completed. If the template number m exceeds the template number, the processing is completed. If the process is not to be ended, the process proceeds to step 2603, and the job number of the template m is stored in the job number w. In step 2604, the operation number n is initialized. In step 2605, the operation number n
Is used to determine the end of the loop processing. Operation number n
Exceeds the number of elements of the best pattern of the template m, the generation of the operation sequence for the template m is completed, and the loop processing ends. Thereafter, in step 2615, the operation allocation time is adjusted so that the end time of the last operation in the operation sequence is equal to the end time of the work w. Specifically, this is realized by extending the processing time of the last operation operation of the operation sequence to an appropriate length. Next, in step 2616, the margin of the operation sequence is calculated. Then, in step 2606, the value of the template number m is incremented by 1 and the process returns to step 2602 to determine whether to end the process. If loop processing is to be continued, step 2607
Then, the operation type variable k stores the value of the position n in the best pattern of the template m, and the resource variable v stores the value of the position n in the best pattern of the template m in the resource allocation table. In step 2608, new operation data is generated (stored in the operation variable op) and set as the n-th operation sequence m. In step 2609, w is set to the job number of op, k is set to the operation type, and v is set to the used resource. In step 2610, an op processing time is set. If the operation type is work, the required time is acquired by referring to the work content of the work data, and the value is set as the processing time. In the case of other operations, the processing time is set to a predetermined required time (for example, the movement from line 01 to line 07 is 5 minutes). Step 261
In step 1, it is determined whether or not the operation type k is entry.
Set the start time of. Step 2 if you are not entering
At 613, the end time of the operation immediately before the operation sequence is set as the start time of op. In the general task scheduling problem, there is often no restriction on the relationship between the start time and the end time unless adjacent operations in the same operation sequence have a temporal overlap. However, in the vehicle replacement plan creation problem, if the processing of each operation is not continuous, the obtained solution is not feasible, and thus a time assignment process unique to such a problem is required. Since it is necessary to observe the restriction on the continuity of this operation also in the shift processing of the operation described below, when the shift processing is performed, not only the operation located after the shifted operation but also the operation existing before the shifted operation are performed. The processing time zone changes. In addition, there is a possibility that resource conflicts and order constraint violations may newly occur, so the generation of executable operation sequences is more complicated in vehicle replacement planning than in general work scheduling. Becomes

The description of FIG. 26 will be continued. Step 2612
Alternatively, after ending one of the steps 2613, the end time of the operation op is set in a step 2614. This is one loop process. In step 2617, the value of the operation number n is increased by 1, and the process returns to the loop process end determination step 2605.

The order constraint expression violation elimination processing in step 2402 in FIG. 24 will be described. The order constraint equation is an inequality or an equation composed of a start time and an end time of two work operations or a constant. All of the order constraint equations configured using the work operation A and the work operation B are shown below.

1. A. Start time * B. Start time 2. A. End time * B. Start time 3. A. Start time * B. End time 4. A. End time * B. End time A. 5. Start time # constant A. 6. End time # constant B. 7. Start time # constant B. End time # constant The sign * represents one of the operators <> ≤ ≥ = and the sign # represents one of the operators> ≥ =. In step 2402, an operation pair that violates the set order constraint expression is found, and the violation is eliminated by shifting the operation. FIG. 27 shows the data structure of the order constraint expression used in this embodiment. This data structure includes information on an operator to be used, operation on the left side, operation (or constant) on the right side, and reference to control data. The control data is information used at the time of resolving the order constraint violation, a violation flag indicating whether the order constraint expression is in a violation state, a valid flag indicating whether the order constraint expression is valid, a left side and a right side. It consists of reference to operation data, an operation to shift when a violation is resolved, a shift start time, a shift width, and a shift type.

The types of shift will be described in detail with reference to FIG. First, the shift type Typel is a case where the start time of the shifting operation is included in the order constraint expression.
As an example, an order constraint expression “start time of work operation A <start time of work operation B” is used. In a situation in which the order constraint formula is violated, the shift target is the work operation B, and the shift width w is “start time of B−start time of A”.
Is calculated by The shift method shifts the entire work operation B backward by w or more as shown in the figure. This satisfies the violation of the order constraint expression.

Next, the shift type Type 2 is a case where the end time of the shifting operation is included in the order constraint expression.
As an example, an order constraint expression “start time of work operation A <end time of work operation B” is used. In a situation in which the order constraint formula is violated, the work operation B is to be shifted.
Similarly to Type 1, the order constraint violation can be eliminated by shifting the entire work operation B backward, but the order constraint violation can also be eliminated by extending the end time of B instead. Further, in this case, it is not necessary to change the allocation time for the operation in front of B in the operation sequence in order to maintain the continuity of the operation. Therefore, in this case, "B
The violation of the order constraint is eliminated by extending the end time of the work operation B by the width w calculated by “the end time of −the start time of A”.

All order constraint expressions can be determined as violations or no violations by comparing the values of the elements of the expressions according to operators. When the order constraint expression is in a violation state, the shift target operation, the shift width, the shift start time, and the shift type for the violation operation pair can be uniquely determined without complicated processing.

The order constraint expression violation elimination process will be described with reference to the flowchart of FIG. First, in step 2901, the degree of satisfaction is determined for an order constraint expression whose valid flag is TRUE. For the order constraint expression in violation state, the operation to be shifted for the violation operation pair,
The shift width, shift start time, and shift type are determined, and their values are set in each item of the control data. Step 29
In step 02, the end of the process is determined. If the number of order constraint expressions in the violating state is 0, the process ends. If the processing is not to be ended, the process proceeds to step 2409, and the violation of the order constraint expression with the earliest shift start time is resolved. Thereafter, the flow returns to step 2901 to determine the degree of satisfaction of the order constraint expression. By sequentially resolving the order constraint violation with the earliest shift start time, an operation sequence set free from order constraint violation can be finally generated.

The earliest order constraint violation resolution processing will be described in detail with reference to the flowchart of FIG. First, in step 3001, control data of an order constraint expression for eliminating a violation is stored in the control data variable dat. Step 300
In 2, the shift target operation of the control data dat is stored in the operation variable op, and the shift width of the control data dat is stored in the integer variable width. In step 3003, the type of shift is determined. If the shift type of the control data dat is typel, in step 3004, the shift operation op and the shift width width are used to shift the operation and adjust the idle time zone, and the process ends. The processing that does not come to an end is ended. With the above processing, the violation elimination processing corresponding to each of the two shift types described with reference to FIG. 28 is executed. Details of the operation shift and the idle time adjustment process in step 2904 will be described with reference to the flowchart in FIG.

Here, the shift operation is op and the shift width is wi.
dth. First, in step 3101, the operation op is shifted by the width width in the sequence. An outline of this processing will be described with reference to FIG. Assuming that the operation 3201 is an operation to be shifted, the operation sequence is first traced back from this operation, and the operation returns to the operation immediately before the operation or the detention operation. In FIG. 32, the operation 3202 corresponds thereto. Next, the partial sequence after the operation 3202 is shifted backward by the width width. That is, the value “width” is added to the start time and end time of all the operations included in the partial sequence. The process of following the operation is required to keep the moving operation and pull-up operation processing times constant. If you do not do this,
There is a case where both processing times are extended in the adjustment processing of the idle time zone after the shift operation.

By the above processing, the departure time exceeds the scheduled time. So that the departure operation is in time for the scheduled time,
The processing time of the work operation or detention operation immediately before departure is reduced as much as possible. The limit of the shortening is the time required for the work in the case of the work operation, and the minimum time of staying of the detention line in the case of the detention operation. If the scheduled time is exceeded even if the limit is exceeded, the shortening process is terminated there.

The above is the outline of the operation shift processing. With this processing, an idle time is generated between the leading operation of the partial sequence shifted backward and the operation immediately before the leading operation. In order to generate an executable vehicle replacement plan, it is necessary to observe restrictions on the continuity of operations, and therefore, a process of filling this idle time by some method is required. The following two methods can be considered. (1) Extend the end time of the immediately preceding operation. In the case of FIG. 32, the end time of the operation 3204 is extended to be equal to the start time of the immediately following operation. (2) Insert a series of operation for turning to the indwelling number line. However, since there is a restriction that the number of lines on the detention line between operations is at most once, this method cannot be used when a line transfer operation sequence to the detention line has already been inserted.

Regarding (1), the processing contents are obvious. The outline of the process (2) for inserting the line-turning operation sequence into the indwelling track will be described with reference to FIG. First, all the moving operation and pull-up operation (partial operation sequence 3301 in FIG. 33) between the work operations before and after the idle time zone are deleted.
Next, the worst pattern is extracted from the operation sequence template and the resource allocation table, which are the templates of the operation sequence. From each worst pattern, a partial pattern corresponding to a vacant time zone is extracted. This partial pattern is the on-line operation sequence for the detention line. In FIG. 33, the partial pattern 3302 and the partial pattern 3303 correspond thereto. The partial operation sequence is embodied from these partial patterns and inserted into the idle time zone.

The process of inserting a line turning operation sequence into the indwelling number line is as follows.
This can be regarded as a process of changing the operation procedure of the job. As will be described later in detail, by properly performing this processing, it is possible to prevent the occurrence of the order constraint violation and the new resource conflict.

When the technique (1) is applied to the adjustment of the idle time zone, since the operation procedure does not change, it is possible to minimize the occurrence of extra turns. However, there is a high possibility that the work operation or the detention operation whose processing time has been extended causes resource contention. In addition, there is a possibility that an order constraint violation may occur in a work operation. On the other hand, when the method (2) is applied, there is a demerit that the number of turns is increased. However, since the allocation time of the immediately preceding operation does not change, resource competition and order constraint violation for the previous operation do not newly occur. . As will be described later in detail, since the resource conflict resolution step in the present embodiment does not target the detention operation, there is a possibility that a new resource conflict may occur even if a line transfer operation sequence is inserted into the detention line. The operation is
Only the inserted move operation and pull-up operation. Since the processing time of these operations is constant and very short, the frequency of occurrence of the conflict is extremely low as compared with the case where the end time of the work operation or the detention operation is extended. The use of the two methods of the idle time zone will be clarified in the description of the shift of the operation and the adjustment processing of the idle time zone with reference to FIG.

After the step 3101 is completed, the process proceeds to a step 3102, where the idle time zone adjustment processing (1) is performed. Next, in step 3103, the type of the operation op shifted is determined, and if op is not a work operation, the process ends. This means that if the moving operation or the pulling-up operation is a shift target, a line transfer operation sequence to the detention line is not inserted immediately before. If op is a work operation, it is determined in step 3105 whether a resource conflict (excluding the detention operation) or an order constraint violation has newly occurred immediately before op. If not, the process ends. If it has occurred, the processing time extended in step 3106 is returned to its original state, and the adjustment processing (2) of the idle time zone is performed in step 3107, and the processing ends.

Returning to FIG. 24, the description of the main constituent steps in the process of generating a solution (operation sequence) will be continued. Next, the process of generating the resource conflict operation pair set in step 2404 will be described. First, FIG. 34 shows the data structure of a resource conflict operation pair used in this embodiment.

Each resource conflict operation pair is given a different operation pair number to distinguish it from the others. The contents of the data include a reference to two operation data constituting an operation pair, an order between operations to be added by shift processing at the time of conflict resolution, a shift start time, and a shift width. When the order between operations is 1 → 2, it means that the operation 2 is shifted backward at the time of conflict resolution, and conversely, when the order between operations is 2 → 1, it means that the operation 1 is shifted backward. The shift start time and the shift width are determined according to the order given between operations.

Next, the details of the process of generating the resource conflicting operation pair set will be described with reference to the flowchart of FIG. In step 3501, a resource variable m is initialized. Step 3
502 determines the end of the process, and ends the process if the resource variable m exceeds the total number of resources. If the processing is not completed, the process proceeds to step 3503, where all operation pairs using the resource m are extracted using the Gantt chart and subjected to partial processing 3504. For details of this processing, first, in step 3505, the type of each operation of the operation pair is determined. If either type of operation is detention, the partial processing ends. If neither type is detention, the process proceeds to step 3506, and it is determined whether or not the operations op1 and op2 forming the operation pair have a temporal overlap. If there is no temporal overlap, the partial processing ends. If there is a temporal overlap, the operation pair (o
A resource conflict operation pair set is created from p1, op2). Thereafter, in step 3508, the resource conflict operation pair is added to the set, and the partial processing ends. Partial processing 3 for all operation pairs
After performing step 504, the value of the resource variable is increased by 1 in step 3509, and the process returns to the end determination step 3502. The process of creating the resource conflicting operation pair set in step 3507 will be described in detail. Resource conflicts can be resolved by shifting backward by the appropriate width for either operation in the operation pair. Therefore, it is possible to assign either of the two orders to the operation pairs. In this embodiment, the order is assigned to the operation pairs based on the following basic policy. (1) Add an operation that does not cause a new order constraint violation. (2) If order constraint violation does not occur in either order, it is not necessary to insert a line transfer operation sequence to the indwelling number line in the case of performing the shift processing (that is, whether or not the sequence has already been inserted, (The resource conflict and the order constraint violation do not occur even if the end time of the operation is extended.)
(3) If the order constraint is violated in either order, the order of shifting the operation with a large margin is set. In this case, the order constraint is clearly violated regardless of which order the shift processing is performed. (4) If there is no need to insert a line transfer operation sequence to the stranded number in either order, or if it is necessary to insert a line transfer operation sequence to the stranded number in either order, perform an operation with a large margin Put the order of shifting. If resource conflicts are resolved by ordering the competing operation pairs under this basic policy, it is expected that a feasible solution that has a uniform margin of each operation sequence and satisfies the order constraint as much as possible can be expected.

The details of the processing embodying the basic policy are shown in FIG.
This will be described with reference to the flowchart of FIG. First, in step 3601, a new resource conflict operation pair is generated and stored in the resource conflict operation pair variable pair. Step 360
In the case of 2, for the operation pair (op1, op2) to be generated, o
The shift width when p1 is shifted and the shift width when op2 is shifted are calculated. In step 3603, it is determined whether either op1 or op2 is an entry or exit operation. If either of the operations is an entry or exit operation, the process proceeds to step 3604, and the order in which the operation other than the entry or exit operation is shifted to pair. Set in order. After that, another value is appropriately set in the pair and the process ends.
If neither is an entry or exit operation, it is determined in step 3605 whether op1 and op2 are both work operations. If at least one of the operations is not a work operation, the process proceeds to step 3610. If both are operation operations, the process proceeds to step 3606, and it is determined whether shifting of both op1 and op2 causes a new order constraint violation between op1 and op2. If neither of them is newly generated, the process proceeds to step 3610. If at least one of them newly causes an order constraint expression violation, it is determined in step 3607 whether or not either shift causes an order constraint expression violation between op1 and op2. If both occur, the order of shifting the operation having a large margin in step 3608 is set to pair. Set in order. If only one of them causes an order constraint violation, step 360
9 to determine the order in which shifts that do not cause order constraint violations are performed. Set in order. After the end of step 3608 or step 3609, another value is appropriately set in the pair, and the process ends. In step 3610,
Only one of op1 and op2 determines whether extending the processing time of the immediately preceding work or detention operation after the shift by the shift width will cause a new resource conflict or a violation of the order constraint. If this is true, in step 3611 the order in which shifts that do not cause conflicts and violations are performed is determined as pair. Set in order. In the case of false, the order of shifting the operation with a large margin is pair. Set in order. After the end of step 3611 or 3612, another value is appropriately set in the pair and the process ends.

Returning to FIG. 24, description of the main constituent steps in the process of generating a solution (operation sequence) will be continued. Regarding the earliest resource conflict resolution processing in step 2408,
This will be described with reference to the flowchart in FIG.

First, in step 3701, a resource conflicting operation pair having the earliest shift start time is stored in the resource conflicting operation pair variable pair. Next, in step 3702, the operation variable op
The operation to fix the operation to be shifted to the operation variable keep is extracted from the pair as the integer variable width and stored. In step 3703, it is determined whether keep is the last work operation in the operation sequence, and whether a line transfer operation sequence to the indwelling number is inserted behind it. If this is true, keep is a work operation that also serves as a standby before leaving, so that the processing time can be significantly reduced. Therefore, in step 3704, k
Reduce processing time of eep as much as possible, and keep in step 3705
After that, the sequence of operation for turning to the indwelling track is inserted, and the process is terminated. The insertion method is the same as step 3107 in FIG. If the determination is false, the shift operation and the adjustment of the idle time zone in step 3004 described in the processing for eliminating the order constraint violation in FIG. Returning to FIG. 24, the description of the main configuration steps in the solution (operation sequence) generation processing will be continued. The earliest order constraint expression violation solving process in step 2409 has been described in the description of the order constraint expression violation solving process in FIG. The process of resolving resource conflicts occurring before the shift operation in step 2411 will be described with reference to the flowchart in FIG.

First, in step 3801, a set of resource conflicting operation pairs for all operations preceding the shifted operation p is generated. Details of this step will be described with reference to the flowchart in FIG.

First, in step 3901, the head operation of the operation sequence including p in the operation variable tmp is stored.
Next, in step 3902, the end of the process is determined.
If tmp and p are the same operation, the process ends. If the process is not completed, the process proceeds to step 3903 to determine whether the type of tmp is detention. If the type of tmp is detention, in step 3909 tmp is added to tmp in the operation sequence.
Is stored, and the process returns to the end determination processing of step 3902. Step 3 if the type of tmp is not detention
Proceeding to 904, partial processing 3905 is performed for all operations using the same resources as tmp using the Gantt chart. The partial process first determines in step 3906 whether there is a temporal overlap between tmp and the operation. If there is no temporal overlap, the partial processing ends. If there is a temporal overlap, a resource conflicting operation pair set is generated in step 3907 in the order of shifting tmp. In step 3908, the resource conflict operation pair is added to the set, and the partial processing ends. After the completion of the partial processing, at step 3909, tmp
The operation next to tmp in the operation sequence is stored in, and the process returns to the end determination processing of step 3902.

Returning to FIG. 38, the description of the processing for resolving resource conflicts that occurred before the shifted operation will be continued. Step 38
01, the process advances to step 3802 to determine the end of the process. If the number of resource conflicting operation pair sets is 0, the process ends. If the number of sets is not 0, the fastest resource conflict resolution processing of step 2409 is performed.
Thereafter, the flow returns to step 3801. This is the end of the detailed description of the solution (operation sequence) generation process in the solution search process of FIG. Next, the solution evaluation step in the solution search processing will be described. In this step, it is checked to what extent the generated solution satisfies the evaluation items of the solution set in advance, and an evaluation value is given by the penalty method based on the number of violations of the item and a penalty according to the content of the violation. FIG. 40 shows an example of the solution evaluation items used in this embodiment. Items can be easily deleted or added to the solution evaluation items, and the penalty for violation can be easily adjusted. Details of the solution evaluation processing will be described with reference to the flowchart in FIG.

First, in step 4101, the item number m
Is initialized. In step 4102, it is determined whether the process is completed. If the item number m exceeds the number of evaluation items of the solution, the process ends. If the processing is not to be ended, the process proceeds to step 4103, and the specified maximum score is stored in the score p.
In step 4104, the degree of satisfaction of the solution for the evaluation item m is determined, and the number of violations of the item is stored. In step 4105, the score calculated by the formula “number of violations × penalty of evaluation item m” is subtracted from score p. In step 4106, the value of the item number m is incremented by 1, and the process returns to the end determination step 4102.

The detailed description of the solution search processing in FIG. 16 has been completed. The description of the embodiment of the data, processing, and apparatus shown in the basic configuration (FIG. 1) of the present invention will be continued. The solution 0113 in FIG. 1 is an operation sequence for all tasks (compositions) obtained by the solution search process described based on the flowchart in FIG. The data structure is shown in FIG.

In this embodiment, the solution is displayed on the display device 0116 and the operation by the operator is received from the input device 0115 in the display of the solution and the operation process of the operator in step 0114 of FIG. The operator outputs the work data of FIG. 7, the resource attribute table of FIG. 13, the connection determination table of FIG. 14, the solution search parameter of FIG. 15, the order constraint equation of FIG. 27, and the solution evaluation item of FIG. These can be edited, and the execution of the solution search process of FIG. 16 can be again instructed using the edited data. The solution approved by the operator is registered as a formal vehicle replacement plan, converted into a business format, and output from the printer 0117.

[0087]

【The invention's effect】 [Brief description of the drawings]

FIG. 1 shows a basic configuration of the present invention.

FIG. 2 is an explanatory diagram of a constraint on an operation procedure.

FIG. 3 is an example of a Gantt chart.

FIG. 4 is an example of a vehicle replacement schedule.

FIG. 5 is an example of a depot layout.

FIG. 6 is an example of a vehicle replacement diagram.

FIG. 7 is a data structure of work data in the present embodiment.

FIG. 8 is a data structure of operation data in the present embodiment.

FIG. 9 is a data structure of an operation sequence in the embodiment.

FIG. 10 is an explanatory diagram of restrictions on an operation procedure in the present embodiment.

FIG. 11 is a data structure of an operation sequence template.

FIG. 12 is an operation sequence template generation processing flow.

FIG. 13 is a data structure of a resource attribute table.

FIG. 14 is a data structure of a connection determination table.

FIG. 15 is an explanatory diagram of a solution search parameter in the present embodiment.

FIG. 16 is a solution search processing flow in the present embodiment.

FIG. 17 is a data structure of a resource allocation table.

FIG. 18 is a flowchart of processing for initializing a resource allocation table.

FIG. 19 is a flowchart of processing for allocating resources to the worst pattern of the operation sequence template.

FIG. 20 is a flowchart of a generation process of a child chromosome.

FIG. 21 is a flowchart of processing for generating a worst path of an operation sequence template of a child chromosome from a parent chromosome.

FIG. 22 is a flowchart of a mutation process.

FIG. 23 is a flowchart of a mutation process for the worst path of the operation sequence template.

FIG. 24 is a flowchart of a process for generating an operation sequence.

FIG. 25 is a data structure of a Gantt chart in the present embodiment.

FIG. 26 is a flowchart of an operation sequence initialization process.

FIG. 27 is a data structure of an order constraint expression in the embodiment.

FIG. 28 is an explanatory diagram of an order constraint violation operation pair.

FIG. 29 is a flowchart of a process for resolving an order constraint violation.

FIG. 30 is a flowchart of the earliest order constraint violation resolution processing flow.

FIG. 31 shows an operation shift and an idle time adjustment process.

FIG. 32 is an explanatory diagram of an operation shift process.

FIG. 33 is an explanatory diagram of a process of inserting a line-turning operation sequence into a detention line.

FIG. 34 is a data structure of a resource conflict operation pair.

FIG. 35 is a flowchart of a resource conflicting operation pair set generation process.

FIG. 36 is a flowchart of a resource conflicting operation pair generation process.

FIG. 37 is a flowchart of the earliest resource conflict resolution processing flow.

FIG. 38 is a flowchart of a process for resolving resource conflicts occurring before a certain operation.

FIG. 39 is a flowchart of a generation process of a resource conflicting operation pair for all processes before a certain operation.

FIG. 40 is an explanatory diagram of a solution evaluation item in the present embodiment.

FIG. 41 is a flowchart of a solution evaluation process.

[Explanation of symbols]

0101: work data, 0102: constraints on resources,
0103: Restriction on operation procedure, 0104: Restriction on operation procedure, 0105: Solution search processing, 0106
... probabilistic adjustment of resource allocation, 0107 ... resolution of resource conflict,
0108: evaluation of solution candidates; 0109: feedback of evaluation values; 0111: restrictions on the order of work;
0112: Solution evaluation item, 0113: Solution, 0114: Solution display and operator operation, 0115: Input device, 011
6 display device, 0117 printer.

Claims (16)

[Claims] An operation set O is a semi-ordered set having an order relation ≦ O∪W≡V from an essential operation set W and an auxiliary operation set V, and an element op is a resource that is an element of a resource set R. r
And a start time st which is an element of the time set T and an end time ft which is an element of the time set T as attributes, and the job Ji (i
= 1, 2,...) Are all ordered subsets in which J i ⊂ W, all element attributes are unassigned, and J ₁ ∩ J ₂ ∩ =
φ, and the schedule Si of the job Ji is Ji⊂Si⊂O
, S ₁ ∩S ₂ ∩... = Φ, and all given work schedules Sall 1S ₁ ∪S ₂ ∪ ·· (1) From the input work data, a constraint on an operation procedure indicating an operation that can be used by each work and an order between the operations is defined as: (2) probabilistically adjusting the combination of resource allocations for operations; and (3) solving time schedules for resources required for operations and solving schedules for all tasks. Creating a candidate; (4) evaluating a solution candidate obtained by performing step (3) to calculate an evaluation value; and (5)
(E) feeding back the evaluation value obtained by performing step (4) to the above-mentioned stochastic adjustment step of resource allocation; and (6) performing the above-mentioned steps (2) to (5) to search for an optimal solution for a solution candidate. And (7) displaying a solution representing the schedule of all jobs obtained as a result of the optimization and editing the solution by an operator. 2. The method according to claim 1, wherein the step (2) relates to a chromosome indicating a resource allocation status of each node of the constraint relating to the operation procedure for all jobs, and a resource indicating a resource attribute and a relationship between resources. 2. The scheduling method according to claim 1, wherein the number set in the solution search parameter is created while satisfying the constraint. 3. The method according to claim 1, wherein in the step (2), a genetic operation including at least natural selection processing, crossover processing, and mutation processing is performed on the chromosome based on the evaluation value given to the chromosome. 2. The scheduling method according to claim 1, wherein new chromosome data in which the resource allocation status of each node of the constraint on the procedure is different is created. 4. The crossover process according to claim 3, wherein the crossover process selects one of two randomly selected chromosomes from the resource allocation status of the same node and sets the selected node to a new chromosome at the same node. The scheduling method according to claim 1, wherein 5. The method according to claim 4, wherein when selecting one of the resource allocation statuses of the same node of two randomly selected chromosomes, the resource allocation status satisfying the resource-related restriction is selected. However, if both satisfy the resource constraints, the possibility of inheriting the resource allocation status of chromosome data with a high evaluation value is dynamically adjusted to reflect the relative height of the evaluation value. 2. The scheduling method according to claim 1, wherein the selection is performed according to the probability determined in (1). 6. The method according to claim 3, wherein the mutation treatment comprises:
2. The scheduling method according to claim 1, wherein the resource allocation status of some nodes of the chromosome is changed probabilistically as long as the resource does not violate a resource constraint. 7. The method according to claim 1, wherein step (3) comprises:
(A) For each job, create an operation sequence indicating operation procedures, resources used by each operation, and processing time of each operation,
A sub-step of initializing the schedule of each job as a schedule of each job by performing a process of satisfying the constraint on the order of the operations for all operation sequences; (b) (a)
2. The method according to claim 1, further comprising the sub-step of selecting an operation pair that causes a time conflict of the used resources from the solution candidates of all the work schedules obtained by executing the task, and resolving the resource conflict. Scheduling method. 8. The method according to claim 7, wherein the sub-step (a)
2. The scheduling method according to claim 1, wherein the operating method selects an operation procedure of each task from constraints on the operation procedure, and allocates resources used in each operation based on a resource allocation status obtained from a chromosome. 9. The method according to claim 7, wherein:
The start time and end time of one of the operations included in the order constraint expression are defined as an order constraint expression, which is an inequality or an equality equation composed of a start time of the work operation, an end time of the work operation, and a constant, as a constraint on the work order. 2. The scheduling method according to claim 1, wherein the constraint on the order of the work is satisfied by shifting the time backward. 10. The method according to claim 7, wherein the sub-step (b) is performed.
Resolves the time conflict of the used resources by selecting one of the two operations for the operation pair causing the time conflict of the used resources and shifting the start time and the end time of the selected operation backward in time. 2. The scheduling method according to claim 1, wherein: 11. An idle time generated between an operation whose start time and end time are shifted backward in time and an operation included in a schedule of the same work existing ahead of said operation in time according to claim 10. By (1) extending the end time of the operation,
2. The scheduling method according to claim 1, wherein the adjustment is performed by any of (2) a partial change of an operation procedure. 12. The method according to claim 10, wherein, when one of the two operations is selected, the start time and the end time are shifted backward in time to select an operation that does not cause a new order constraint expression violation. The scheduling method according to claim 1, wherein 13. The method according to claim 1, wherein step (4) comprises:
2. The scheduling method according to claim 1, wherein an evaluation value of each of the solution candidates is calculated according to a degree of satisfaction of a solution evaluation item created from a predetermined constraint condition. 14. The method according to claim 1, wherein step (5) comprises:
2. The scheduling method according to claim 1, wherein the evaluation value calculated in the step (4) is provided to a chromosome used when creating a solution candidate. 15. The method according to claim 1, wherein step (5) comprises:
2. The operator according to claim 1, wherein the operator modifies the work data, resource constraints, solution search parameters, work order constraints, and solution evaluation items using the input device, and instructs re-execution of the solution search. Scheduling method. 16. An operation set O is a partially ordered set having an order relation ≦ O ≦ W∪V from a required operation set W and an auxiliary operation set V, and an element op is a resource that is an element of the resource set R. r and start time st and time set T which are elements of time set T
Has the end time ft, which is an element of
(I = 1,2, ···) is the total order subsets is Ji⊂W, in the attributes of all elements unallocated, J ₁ ∩J ₂ ∩ ··
== φ, and the schedule Si of the job Ji is Ji⊂Si
In the all ordered subset of ⊂O, attributes of all elements have been assigned, S ₁ ∩S ₂ ∩... = Φ, and all given work schedules Sall≡S ₁ ∪S ₂ ∪ Is a scheduling device that uses an information processing device that creates ... (1) From input work data, a constraint on an operation procedure indicating an operation that can be used by each job and an order between the operations is set as an operation. Means for generating a directed network as a node; (2) means for stochastically adjusting the combination of resource allocation for operations; and (3) schedule for all tasks by eliminating temporal conflicts between resources required for operations. Means for creating a solution candidate for
(4) means for evaluating the solution candidates obtained by executing the means (3) to calculate an evaluation value; and (5) calculating the evaluation value obtained by the execution of the means (4) based on the probability of the resource allocation. Means for feeding back to the adjusting means; (6) means for performing the optimal solution search processing of the solution candidates by repeating the above means (2) to (5); and (7) all means obtained as a result of the optimization. Means for displaying a solution representing a work schedule and allowing an operator to edit the solution.