Information processing device and information processing method
The information processing device and method optimize high-importance objective functions in large-scale combinatorial problems using a two-step approach with mixed integer programming and Ising machines, addressing the limitations of existing Ising machine methods.
Patent Information
- Application Number
- JP2023081155
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-17
- Publication Date
- 2024-11-28
AI Technical Summary
Existing methods using Ising machines struggle with large-scale combinatorial optimization problems, particularly in resource-constrained scheduling, as they fail to strictly minimize the most important objective function (makespan) and do not effectively consider additional objective functions.
An information processing device and method that prioritizes optimization of high-importance objective functions, reduces variable definition through provisional solutions, and employs a two-step optimization process using mixed integer programming and Ising machines to address multiple objective functions.
Effectively minimizes critical indicators like makespan while considering other optimization factors, ensuring reliable and efficient solutions for large-scale combinatorial problems.
Smart Images

Figure 2024165214000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to an information processing device and an information processing method. [Background technology]
[0002] In combinatorial optimization problems, there are cases where an optimal solution must be found under multiple objective functions. In such multi-objective optimization, the importance of each objective function differs, and they can be classified into particularly important objective functions and objective functions with relatively low importance. Therefore, it is important to reliably optimize the particularly important objective functions while also optimizing in a way that reflects the influence of objective functions with relatively low importance. However, this is difficult to achieve with large-scale combinatorial optimization problems.
[0003] For example, in resource-constrained scheduling problems, minimizing makespan (work period) is often set as the most important objective function, but in addition, reducing the resources used and leveling the amount of resources used may also be additional objective functions. In such cases, optimization that strictly minimizes makespan while also taking into account other objective functions is required. In particular, when the target problem is large-scale and there is an objective function other than makespan that is expressed as a quadratic expression, it is difficult to optimize using a linear programming solver by transforming the quadratic expression into a linear expression due to the large scale, and it is desirable to utilize a method suitable for optimizing quadratic expressions, such as an Ising machine.
[0004] Regarding the processing of the above-mentioned combinatorial optimization problem, Patent Document 1 states that "We provide a method for obtaining an approximate solution to a makespan minimum solution with high accuracy in a short time for a resource-constrained scheduling problem." and "An optimization device that iteratively obtains a constraint violating solution, a constraint satisfying solution, and an approximate solution includes an optimization processing unit that specifies an optimization target end time between a first end time, which is the end time of a previous constraint violating solution, and a second end time, which is the end time of a previous constraint satisfying solution, and obtains an approximate solution by a metaheuristic method, an optimization result determination unit that determines whether the approximate solution violates or satisfies the constraint, and a solution updating unit that updates the first end time with the end time of the approximate solution if the approximate solution violates the constraint based on the determination result by the optimization result determination unit, and updates the second end time with the end time of the approximate solution if the approximate solution satisfies the constraint, and is characterized in that the approximate solution is iteratively obtained while updating the first end time and the second end time."
[0005] Non-Patent Document 1 describes the solution of a job shop scheduling problem using an Ising machine. [Prior art documents] [Patent documents]
[0006] [Patent Document 1] JP 2021-190019 A [Non-patent literature]
[0007] [Non-Patent Document 1] D. Shimada, T. Shibuya and T. Shibasaki, A Decomposition Method for Makespan Minimization in Job-Shop Scheduling Problem Using Ising Machine, 2021 IEEE 8th International Conference on Industrial Engineering and Applications (ICIEA), Chengdu, China, 2021, pp. 307-314 Summary of the Invention [Problem to be solved by the invention]
[0008] In Patent Document 1, a metaheuristic method such as simulated annealing using an Ising machine is used to minimize the makespan by repeatedly finding constraint-violating solutions and constraint-satisfying solutions while changing the upper limit of the makespan in a resource-constrained scheduling problem.
[0009] However, the number of iterations required to obtain the minimum makespan is unknown and may be excessively large in some cases, and because it uses a metaheuristic approach, there is no guarantee that the makespan of the obtained solution is strictly minimum.
[0010] In addition, in Non-Patent Document 1, in resource constrained scheduling, the sum of the end times of each task is minimized using an Ising machine instead of the makespan, and an approximate solution to the makespan minimization problem is searched for.
[0011] However, even if the sum of the task finish times is the same, the makespan can differ between a case where the completion time of one task is much slower than the others and a case where all tasks are completed at roughly the same time. As can be seen, this method does not necessarily minimize the makespan.
[0012] As described above, in the Ising machine, there is a problem in strictly minimizing the makespan, which is the most important factor, and the optimization of other additional objective functions cannot be taken into consideration.
[0013] The present invention has been made in consideration of the above background, and aims to provide an information processing apparatus and an information processing method that are capable of solving large-scale combinatorial optimization problems consisting of multiple objective functions, not limited to the resource-constrained scheduling problem exemplified above. [Means for solving the problem]
[0014] In order to solve the above problems, the present invention provides an information processing device that uses a computer to process a combinatorial optimization problem consisting of a plurality of objective function terms, the information processing device comprising: a first stage including a combinatorial optimization unit that performs an optimization process for objective function terms with high importance among the plurality of objective function terms; and a variable definition unit that reduces the number of variables to be defined by referring to a tentative solution obtained by the combinatorial optimization unit; and a second stage including a combinatorial optimization unit that performs an optimization process for objective function terms taking into account objective function terms with relatively low importance; and an optimization result judgment unit that judges whether to terminate the optimization process based on a specified judgment criterion; and when re-executing the optimization process, the information processing device updates the conditions for defining the variables and repeatedly executes the optimization process of the second stage.
[0015] The present invention is also described as "an information processing method for processing a combinatorial optimization problem consisting of a plurality of objective function terms using a computer, the information processing method comprising the steps of: in a first step, executing an optimization process for objective function terms having a high importance among the plurality of objective function terms; and performing a process of reducing the number of variables to be defined by referring to a tentative solution obtained by the optimization process; in a second step, executing an optimization process for objective function terms taking into account objective function terms having a relatively low importance; and performing a process of determining whether to terminate the optimization process based on a specified determination criterion; and, if the optimization process is to be re-executed, updating the conditions for defining the variables and repeatedly executing the optimization process of the second step." Effect of the Invention
[0016] According to the present invention, even for large-scale combinatorial optimization problems, important optimization indices can be rigorously optimized while other optimization indices can be taken into account. Problems, configurations, and effects other than those described above will become apparent from the following description of the embodiments of the invention. [Brief description of the drawings]
[0017] [Figure 1] FIG. 1 is a block diagram showing an example of a schematic hardware configuration of an information processing apparatus. [Diagram 2] FIG. 2 is a block diagram showing an example of a main functional configuration of an information processing apparatus. [Diagram 3] 1 is a flowchart showing an example of a procedure of an optimization method. [Figure 4] 11 is a flowchart showing an example of specifying the order of operations. [Diagram 5] FIG. 13 is a diagram showing an example of a table showing the number of work days for each task for each company. [Figure 6] 1 is a flowchart showing an example of a procedure of an optimization method. [Figure 7] FIG. 13 is a diagram showing an example of a rule-based work plan. [Figure 8] FIG. 13 is a diagram showing an example of a work plan obtained by a mixed integer programming solver under the condition that a plan must be created in the minimum total number of days. [Figure 9] FIG. 13 is a diagram showing an example of a work plan obtained by an Ising machine under the condition that a plan is created in the minimum total number of days. [Figure 10] This is an example of a work plan calculated using a mixed integer programming solver under the condition that the maximum allowable total number of days is 23 days. [Figure 11] This is an example of a work plan calculated using a mixed integer programming solver under the condition that the maximum allowable total number of days is 24 days. [Figure 12] FIG. 13 shows an example of a work plan obtained by an Ising machine under the condition that the maximum allowable total number of days is 24 days. [Figure 13] FIG. 13 shows an example of a work plan obtained by an Ising machine under the condition that the maximum allowable total number of days is 26 days. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0018] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. In the following description, the same or similar components will be given common reference numerals and duplicated descriptions will be omitted. In addition, when there are multiple elements having the same or similar functions, different subscripts will be added to the same reference numerals. In addition, when there is no need to distinguish between multiple elements, the subscripts will be omitted.
[0019] The optimization problem solving process according to the present invention can be applied to many situations, but in the following example, we consider a resource-constrained scheduling problem in which the resources such as personnel, companies, and machines in charge of each task are fixed, and the resource allocation and start time for each task are determined. As an example, we consider a case where there are constraints on the order of each task.
[0020] In such scheduling problems, the most important objective may be to minimize the time when the entire work is completed. However, in the actual work planning process, optimization that takes into account various indicators in addition to minimizing the work completion time is required, such as reducing the number of people involved in the work, performing work by the same person as much as possible in succession, and leveling out the workload among people.
[0021] Therefore, as an embodiment of the present invention, an example of creating a schedule that places importance on the overall work completion time while also taking other items into consideration will be shown. The embodiment described in this disclosure is not limited to cases where minimizing the work completion time is important, but can also be applied to cases where minimizing total costs, maximizing profits, etc. are important while also optimizing other items of relatively low importance. EXAMPLES
[0022] Fig. 1 shows an example of the hardware configuration of an information processing device that executes work plan optimization. The information processing device 100 shown in Fig. 1 includes a processor 101, a main memory device 102, an auxiliary memory device 103, an input device 104, an output device 105, a communication device 106, one or more combinatorial optimization devices 107, and a system bus 108 that communicatively connects these devices.
[0023] The information processing device 100 may be realized, for example, by using a virtual information processing resource such as a cloud server provided in part or in whole by a cloud system. The information processing device 100 may also be realized, for example, by a plurality of information processing devices connected to each other so as to be able to communicate with each other and to operate in cooperation with each other.
[0024] The processor 101 is configured using, for example, a central processing unit (CPU) or a micro processing unit (MPU). The main storage device 102 is a device that stores programs and data, and is, for example, a read only memory (ROM) (static random access memory (SRAM), non-volatile RAM (NVRAM), mask read only memory (mask ROM), programmable ROM (PROM), etc.), a random access memory (RAM) (dynamic random access memory (DRAM)), etc.). The auxiliary storage device 103 is a hard disk drive, a flash memory, a solid state drive (SSD), an optical storage device (compact disc (CD), digital versatile disc (DVD), etc.), etc. The programs and data stored in the auxiliary storage device 103 are read into the main storage device 102 as needed.
[0025] The input device 104 is a user interface that accepts information input from a user, such as a keyboard, a mouse, a card reader, a touch panel, etc. The output device 105 is a user interface that provides information to a user, such as a display device (LCD (Liquid Crystal Display), a graphic card, etc.) that visualizes various information, an audio output device (speaker), a printer, etc. The communication device 106 is a communication interface that communicates with other devices, such as a NIC (Network Interface Card), a wireless communication module, a USB (Universal Serial Interface) module, a serial communication module, etc.
[0026] The combinatorial optimization device 107 is a device that solves an input combinatorial optimization problem. The combinatorial optimization device 107 may be dedicated hardware that is designed specifically to execute a metaheuristic algorithm such as simulated annealing, or may be dedicated hardware that executes a solution search for an optimization problem expressed as an Ising model. However, without providing a dedicated combinatorial optimization device 107, the general-purpose calculation processor 101 may function as a combinatorial optimization device to solve the combinatorial optimization problem.
[0027] The combinatorial optimization device 107 may take the form of an expansion card mounted on the information processing device 10, such as a graphics processing unit (GPU). The combinatorial optimization device 107 may be configured with hardware such as a complementary metal oxide semiconductor (CMOS) circuit, a field programmable gate array (FPGA), or an application specific integrated circuit (ASIC). The combinatorial optimization device 107 includes a control device, a storage device, an interface for connecting to the system bus 5, and the like, and transmits and receives commands and information to and from the processor 101 via the system bus 108. The combinatorial optimization device 107 may be communicably connected to another combinatorial optimization device 107 via a communication line, for example, and may operate in cooperation with the other combinatorial optimization device 107.
[0028] Fig. 2 shows the main functional configuration (software configuration) of the information processing device 100. The information processing device 100 shown in Fig. 2 includes a problem data acquisition unit 111, a formulation generation unit 112, a variable definition unit 113, an objective function setting unit 114, a combinatorial optimization unit 115, an optimization result determination unit 116, a solution output unit 117, and a storage unit DB.
[0029] These functions are realized by the processor 101 reading and executing a program stored in the main memory device 102, or by hardware included in the combinatorial optimization device 107. In addition to the above functions, the information processing device 100 may also include other functions, such as an operating system, a file system, a device driver, and a DBMS (DataBase Management System).
[0030] The storage unit DB includes a problem data storage unit DB1, a formulation data storage unit DB2, a variable data storage unit DB3, a tentative solution storage unit DB4, and a calculation program storage unit DB5.
[0031] The problem data acquisition unit 111 stores information input from outside to set a resource-constrained scheduling problem as problem data D1 in the problem data storage unit DB1 of the storage unit DB. The information input from outside (problem data D1) includes the order relationship of each task, information on each resource for each task, information on the time required for each task, items to be set as objective functions for optimization, information on the importance of each objective function term, etc. Information input from outside is performed, for example, by a user via a user interface (input device, output device, communication device, etc.).
[0032] The formulation generation unit 112 mathematically expresses (formulates) the resource-constrained scheduling problem as a combinatorial optimization problem from the problem data information D1 stored in the problem data storage unit DB1, and stores the formulated data D2 in the formulated data storage unit DB2 of the storage unit DB. The variable definition unit 113 refers to the problem data information D1 stored in the problem data storage unit DB1 and the formulated data D2 stored in the formulated data storage unit DB2, defines variables for solving the resource-constrained scheduling problem, and stores the variables as variable data D3 in the variable data storage unit DB3.
[0033] The objective function setting unit 114 determines the objective function to be optimized from problem data information D1 regarding the importance of each objective function stored in the problem data storage unit DB1, and transmits the objective function selected as the optimization target to the combinatorial optimization unit 115.
[0034] The combinatorial optimization unit 115 uses the variable data D3 stored in the variable data storage unit DB3 to solve the formulation data D2 (resource-constrained scheduling problem) described in the formulation data storage unit DB2. The combinatorial optimization unit 115 executes the solution of the resource-constrained scheduling problem using the combinatorial optimization device 107 and the processor 101. The combinatorial optimization unit 115 receives the objective function acquired from the objective function unit 114, the mathematical expression of the resource-constrained scheduling problem D2 stored in the formulation data storage unit DB2, and the variables created in the variable definition unit 113 as inputs, and outputs the solution of the resource-constrained scheduling problem D2 to the optimization result determination unit 116 and the solution output unit 117. The optimization result determination unit 116 receives the solution obtained by the combinatorial optimization unit 115 as inputs and determines whether to continue the solution search.
[0035] When the combinatorial optimization unit 115 finishes finding a solution, the solution output unit 117 reads out the solution obtained by the combinatorial optimization unit 115 and constant values and the like used by the combinatorial optimization unit 115, and stores them as a tentative solution D4 in a tentative solution storage unit DB4 of the storage unit DB. In addition, the solution output unit 117 may output the result read out from the combinatorial optimization unit 115 to the output device 105 or the communication device 106.
[0036] The computation program memory unit DB5 is a program executed by the variable definition unit 113 when defining variables while referring to the problem data memory unit DB1, and a program executed by the combinatorial optimization unit 115 when solving a resource-constrained scheduling problem using the combinatorial optimization device 107 or the processor 101.
[0037] Fig. 3 is a flow chart showing an example of the procedure of the optimization process in the present invention. The optimization method shown in Fig. 3 is preferably executed by each functional unit shown in Fig. 2 using the information processing device 100 configured with the hardware shown in Fig. 1. Hereinafter, the process executed by the information processing device 100 shown in Fig. 3 will be referred to as optimization process S200. In the following, the letter "S" before the reference numerals means a processing step. The optimization process S200 is started, for example, by receiving an instruction from a user via the input device 104.
[0038] The optimization process S200 will be described below with reference to FIG.
[0039] First, in processing step S201, the objective function setting unit 114 refers to the information on the importance of each optimization objective (problem data information D1) stored in the problem data storage unit DB1, and sets only items with particularly high importance as the objective function, and the combinatorial optimization unit 115 executes the optimization process.
[0040] Taking a resource-constrained scheduling problem as an example, possible optimization objectives include minimizing task completion times, minimizing costs, minimizing the number of workers, leveling out the workload of each worker, and having the same worker perform tasks as consecutively as possible. Each optimization objective may be input by the user via a user interface. The importance of each optimization objective may also be input by the user.
[0041] The method of specifying the importance of each item to be optimized includes, for example, specifying it by a numerical value as a weighting for each objective function term, specifying it by an order of priority, etc. Also, the user may directly specify the objective function terms to be optimized in each of the process step S201 and the subsequent process step S203. In the optimization in process step S201, the combinatorial optimization unit 115 may be used multiple times to perform repeated optimization calculations. After the optimization in process step S201 is completed, the obtained solution is stored in the provisional solution storage unit DB4 of the storage unit DB.
[0042] In process step S202, after the process of process step S201, the solution obtained in process step S201 is set as a tentative solution D4, and the variable definition unit 113 redefines the variable D3 using information on the tentative solution D4. In the variable definition in process step S202, the definition of unnecessary variables can be avoided by using information obtained from the tentative solution D4. In other words, in process step S202, pruning of the variable D3 used in process step S201 is performed. This pruning of the variable D3 means that calculations related to variables whose values can be analytically determined in advance are removed from the subsequent process in order to calculate a feasible solution that satisfies all the constraints and to maintain the objective function value already optimized in process step S201. This makes it possible to reduce the number of variables to be calculated in the subsequent optimization process.
[0043] In process step S203, the variables pruned in process step S202 are used to perform optimization including items that were not considered as optimization objectives in process step S201. The optimization in process step S203 does not necessarily have to include items that were optimization objectives in process step S201. In the optimization in process step S203, similar to process step S201, the objective function setting unit 114 determines the objective function to be optimized by referring to the problem data information D1 related to the optimization objective stored in the problem data storage unit DB1, and the combinatorial optimization unit 115 performs optimization. The user may input in advance whether the item that was optimization objective in process step S201 is set as the optimization objective in process step S203, and store it in the problem data storage unit DB1. After the optimization in process step S203 is completed, the solution obtained in process step S203 is stored as a provisional solution D4 in the provisional solution storage unit DB4 of the storage unit DB.
[0044] In process step S204, the optimization result determination unit 116 uses the solution obtained in process step S203 as an input and determines whether to end the optimization process S200 or to repeat a part of the optimization process S200 again. The conditions for determining whether to end the solution search in process step S204 may be, for example, whether the objective function value has been improved from the previous solution, whether a preset time limit has been reached, etc.
[0045] Processing step S205 is executed when it is not determined in processing step S204 that the solution has been completed. In processing step S205, the conditions for the variable pruning process executed in processing step S202 are updated. However, the conditions do not necessarily have to be updated, and the conditions used in the previous processing step S202 may remain unchanged.
[0046] Here, the pruning conditions refer to conditions on how to use the information of the tentative solution D4, such as how to reflect the task completion time found in the tentative solution D4 and whether to reflect the allocation of resources to each task in the variable definition. After the process of process step S205, the variables are redefined in process step S202 by referring to the new tentative solution and the updated pruning conditions. Then, process step S203 is executed again. This series of processes consisting of process steps S205, S202, and S203 is repeatedly executed until it is determined in process step S204 that the solution has been found.
[0047] In short, the optimization process in Figure 3 can be very simply described as a two-stage optimization process in which a first optimization process is performed using only some of the objective functions to be optimized, selected according to their importance, and then a second optimization process is performed using all objective functions, including those with less importance, using the interim solution obtained in the first optimization process.
[0048] An embodiment of the optimization process S200 will be described below using an example of a resource-constrained scheduling problem. In the resource-constrained scheduling problem handled in this example, a problem is assumed in which a company in charge of each task is assigned to each task as a resource for a series of tasks, and the start date of each task is determined. However, the order of each task is specified, and the same company cannot perform multiple tasks simultaneously.
[0049] The problem we will deal with as an example here is to optimize the relationship between multiple tasks and multiple resources (companies in charge of the tasks). We assume that a total of 11 tasks, Task 01 to Task 11 in Figure 4, must be performed in the order shown in the figure.
[0050] Figure 4 is an example of task sequence specification in a resource-constrained scheduling problem, and the execution order between multiple tasks is carried out in accordance with the sequence of tasks in Figure 4. Here, the start and end points of each arrow indicate the sequence of tasks, and the task at the start point must be completed before the task at the end point can be started. For example, tasks 02 and 03 must be completed before task 06 can be started, and tasks 08 and 09 cannot be started unless task 06 is completed.
[0051] In this problem, the multiple resources (companies in charge of tasks) are five companies in total, namely, Company A to Company E. Figure 5 is a table showing an example of a list of the number of work days for each company for each of the 11 tasks shown in Figure 4.
[0052] Each company can only take on tasks for which numbers are entered. For example, company A cannot take on tasks 01 and 05-08, but can take on tasks 02, 03, 04, 09, 10, and 11, with required days of 10, 8, 2, 2, 9, and 6, respectively. Below, we will explain the procedure for scheduling optimization, which assigns each task to each company and determines the start date of each task while adhering to the task sequence shown in Figure 4.
[0053] Fig. 6 is a flowchart showing an example of a scheduling optimization procedure. Hereinafter, the process shown in Fig. 6 and executed by the information processing device 100 will be referred to as an optimization process S300. A series of steps (processing steps S301, S302, S303, S304, S305, S306, and S307) constituting the optimization process S300 corresponds to processing step S201 of the optimization process S200 shown in Fig. 3. Similarly, processing step S309 corresponds to processing step S202, processing steps S308 and S310 correspond to processing step S203, processing step S311 corresponds to processing step S204, and processing step S312 corresponds to processing step S205.
[0054] In processing step S301 in Fig. 6, a tentative work plan (provisional solution) is first created by rule-based processing. In rule-based processing, for example, the company that requires the shortest number of days to complete each task may be assigned to the task. In this case, in the example in Fig. 5, the companies that can complete tasks in the shortest number of days are assigned to tasks 01, 02, and 03, such as Company D (5 days), Company B (2 days), and Company D (4 days), respectively.
[0055] After all the company assignments have been decided, the start date for each task is decided according to the task sequence shown in Figure 4. At this time, the start date is decided so that each task is completed as early as possible while preventing the same company from performing different tasks at the same time.
[0056] Figure 7 shows a work plan created by the rule-based processing described above. The horizontal axis is the number of days elapsed, and the vertical axis is the work number. In this example, we focus on work 01, which is positioned first in the work sequence in Figure 4, and set its start date as the first day. Furthermore, referring to Figure 5, work 01 is assigned to Company D, which can complete the work in the shortest number of days, so the schedule for work 01 is set to days 1 through 5.
[0057] Next, we focus on task 02, which has the end of task 01 as its start condition, and set the start date of task 02 as the sixth day immediately after task 01 ends, and determine the schedule for the two days (seventh day) as determined in Figure 5. Next, we determine the start date of task 03. In terms of the construction order, task 03 can start on the first day, but since both tasks 01 and 03 are handled by Company D and the same company cannot carry out different tasks simultaneously, task 03's start date is set to the sixth day immediately after task 01 ends. In the same way, we can determine the start date of each task from task 04 onwards, while respecting the construction order and ensuring that the same company does not handle multiple tasks at the same time. Note that tasks 06 and 07 can be started only when multiple upstream tasks have been completed, so the start date of the task is determined according to the completion of the task with the latest completion date among the multiple upstream tasks.
[0058] Figure 7 shows an example of a work plan created by the rule-based processing described above. It can be seen that the plan satisfies all of the construction sequence specifications shown in Figure 4, and does not allow each company to work on different tasks at the same time. In addition, the plan created by this rule-based processing has a completion date for the entire work on the 24th.
[0059] In the example of the processing step S301 of the rule-based processing described above, the start dates are determined simply in the order of the numbers of each task, that is, in the order of task 01, task 02, task 03, etc., but this order is not necessarily required. However, when determining the start date of each task, the start dates must already be determined for all tasks that must be completed before starting the task in the task order shown in FIG. 4. For example, when determining the start date of task 06, the start dates of tasks 02 and 03 must already be determined. This means that, for example, when trying to determine the start date of a certain task, if there is even one task that has not been determined as a start date among the tasks that must be completed in advance, the determination of the start date of that task can be postponed and the determination of the start date of the other tasks can be prioritized. For example, if the start date of task 02 has not been determined when trying to determine the start date of task 06, the process of determining the start date of task 06 can be postponed until after the process of determining the start date of task 02.
[0060] In the example of the processing step S301 of the rule-based processing described above, the company in charge of each task is assigned to the company with the smallest number of days required, but other methods of assignment may be used. For example, the assignment of companies to each task and the determination of the start date of each task may be performed in order so that the completion date of each task is the earliest. In this case, for example, task 01 is started by company D on the 1st day. Next, task 02 is started by company B on the 6th day. Task 03 is started by company A on the 1st day. If task 03 is assigned to the company with the smallest number of days required, task 03 is started by company D on the 6th day and completed on the 9th day. However, if the companies are assigned so that the completion date of each task is the earliest, the completion date of this task 03 will be the 8th day. In this way, if the assignment of companies is performed so that the completion date of each task is the earliest, it is possible to advance the completion date of each task while considering the influence of the constraint that the same company cannot be in charge of different tasks at the same time.
[0061] 6 again, in process step S302, a formulation is generated for solving the input resource-constrained scheduling problem with a mixed integer programming solver. An example of the formulation is shown below.
[0062] To formulate the resource constrained scheduling problem, we first define the following variables: s i、j、d : A binary variable that is 1 if company i starts operation j at elapsed day d, and 0 otherwise δ i : A binary variable that is 1 if company i is responsible for at least one task, and 0 otherwise D: A continuous or integer variable that corresponds to the upper bound of the total work completion date Next, we define the following constants: d i、j : Number of days required by company i for operation j d max : Maximum number of total working days allowed M i : The maximum number of tasks that company i can handle N i :Total number of companies w k : Weighting coefficient for each objective function (k=1, 2, ...) The above definition can be more specifically described as follows. First, i、j is the number of days required for each task by each company, as shown in Figure 5. max is set in process steps S303 and S308 of FIG. 6 by referring to the tentative work plan obtained by the rule-based process. i is the table of required days shown in Figure 5 and d max For example, d max In the case of M = 13, Company A can only be responsible for a maximum of three tasks: Task 04, Task 09, and Task 11. A社 = 3. N j is the total number of companies, which is 5 in this resource-constrained scheduling problem. k is the objective function Q k is a weighting factor for the , which may be specified by the user.
[0063] Next, we define the following sets: P: A set of pairs of work with ordering constraints I j: A set of companies that can handle task j J i : A set of tasks that company i can handle The above definition can be more specifically described as follows. First, P includes, for example, (task 01, task 02), (task 02, task 05), etc. j and J. i can be determined from the table of required days shown in Figure 5. For example, I 作業02 = {Company A, Company B, Company C, Company E}, and J A社 ={Task 02, Task 03, Task 04, Task 09, Task 10, Task 11}.
[0064] The objective function in this resource-constrained scheduling problem is expressed as the objective function term Q, for example, as shown in the following equations (1) to (4) using the variables, constants, and sets defined above. k It can be expressed as:
[0065]
number
[0066]
number
[0067]
number
[0068]
number
[0069] The objective function term Q1 corresponds to the upper bound of the total work completion date, and the work completion date can be minimized by minimizing the objective function term Q1 while satisfying the constraint terms described below. The objective function term Q2 is the number of companies that can be assigned at least one person to the work. δ iBy minimizing objective function term Q2 while satisfying the constraints on , the number of companies to which work can be assigned can be minimized. Objective function term Q3 is the total number of consecutive tasks by the same company multiplied by -1. Here, consecutive tasks refer to starting one task immediately after completing another. Multiplying by -1 is done in order to maximize the number of consecutive tasks by minimizing objective function term Q3. Objective function term Q4 is the variance in the number of working days among all companies in the entire work schedule. By minimizing objective function term Q4, the number of working days among companies can be equalized.
[0070] In the conceptual explanation of the optimization process in FIG. 3, in the process step S201, it is stated that "only items with high importance are optimized in the first optimization process", which means that a part of the objective function terms Q is selected from the objective function terms Q1, Q2, Q3, and Q4.
[0071] The overall objective function H in this resource-constrained scheduling problem is the above objective function term Q k Weighting factor w k This can be expressed as the weighted sum of the following equation (5).
[0072]
number
[0073] 6 in this embodiment, in order to create a formulation for a mixed integer programming solver, the weighting coefficient w3 of the objective function term Q3 and the weighting coefficient w4 of the objective function term Q4, which are nonlinear terms, are always set to 0. In this example, the optimization process for only the items with high importance in the processing step S201 in FIG. 3 is performed using a known mixed integer programming solver, but other methods such as simulated annealing, genetic algorithm, tabu search, and stochastic evolution method may also be used.
[0074] Next, we formulate the constraints. In this resource-constrained scheduling problem, we need to consider the constraints shown in the following equations (6) to (10).
[0075]
number
[0076]
number
[0077]
number
[0078]
number
[0079]
number
[0080] Equation (6) indicates that each task must be executed exactly once. Equation (7) reflects the task ordering shown in Figure 4, and all task pairs (j p , j n ) for work j p is working n Equation (8) prohibits the same company from working on multiple tasks at the same time. Equation (9) indicates that δ i The constraint on δ is that if company i is assigned at least one task, i becomes 1. By minimizing the objective function term Q2 while satisfying this equation (9), δ iis 0 only if company i has not been assigned any tasks, and the number of companies performing tasks can be expressed by the objective function term Q2. Equation (10) sets an upper bound for the task completion date, indicating that any task will be completed by the D-1th day. By minimizing the objective function term Q1 while satisfying this equation (10), the overall task completion date can be minimized.
[0081] In process step S303 of FIG. 6, the work completion date in the provisional plan obtained in process step S301 is multiplied by a constant d max However, it is not necessary to use the actual completion date in the tentative plan, but you can use the value obtained by adding a few days to the completion date, or by subtracting a few days from the completion date. max Assume that the 24th is substituted for the date of completion of the work in the plan created by the rule-based processing shown in FIG.
[0082] In processing step S304 of FIG. 6, the variable s i、j、d , δ i , D. Define the variable δ i For all companies, δ A社 , δ B社 However, in the process step S305 described later, if the objective function term Q2 is not included in the objective function H, the variable δ i The variable D is defined as a continuous variable or an integer variable. i、j、d is the set of tasks J that can be handled by each company i. i Also, for company i and operation j, we define the variable s i、j、d The range of dates d that defines max In addition, the variable s i、j、d is defined only for the date d on which company i can start task j, with reference to the construction sequence specification shown in Figure 4.
[0083] Specifically, the earliest finish date for all paths of pre-tasks for task j is calculated, and the day after the maximum of these is set as the earliest start date for task j. Taking task 06 in Figure 4 as an example, the pre-task paths are (task 01, task 02) and (task 03), and the earliest finish date can be calculated by adding up the minimum number of days required for each task for these two paths. As a result, the earliest finish date for (task 01, task 02) is the 7th day, and the earliest finish date for (task 03) is the 4th day, so the earliest start date for task 06 is the 8th day. This argument means that since it takes a minimum of 7 days to complete the pre-tasks for task 06, task 06 can only start on the 8th day at the earliest.
[0084] In addition, calculate the minimum number of construction days for all passes of the subsequent tasks of task j, and the maximum of these is d max The value subtracted from (=24 days) is the latest date that is permitted as the end date of task j. Taking task 06 in Figure 4 as an example, the paths of subsequent tasks are (task 08, task 11) and (task 09, task 11), and the minimum number of construction days for these two paths can be calculated by adding up the minimum number of construction days for each task. As a result, the minimum number of construction days for (task 08, task 11) and (task 09, task 11) are found to be 10 days and 4 days, respectively, so the latest allowable end date for task 06 is d. max -10=day 14. This argument means that in order for the successor to activity 06 to be completed by day 24, activity 06 must be completed no later than day 14.
[0085] The earliest start date of task j obtained by the above method is d j start , the latest finish date is d j finish Then, the variable s ijd The range of days d that defines j start , d j finish -d ij For example, the variable s for Company D for task 06 shown in Figure 4 can be limited to D社、作業06、dThe range of days d that defines the operation is [8, 10]. That is, the variable s ijd s D社、作業06、8日目 , s D社、作業06、9日目 , s D社、作業06、10日目 On the other hand, for the range of days other than these, the variable s D社、作業06、d is always 0, so there is no need to define it.
[0086] 6, an optimal work plan is created by finding a combination of variable values that minimizes the objective function value using, for example, a known mixed integer programming solver for the formulation generated in process step S302 and the variables defined in process step S304. In the optimization process in process step S305, the tentative work plan obtained in process step S301 may be used as an initial solution.
[0087] Hereinafter, examples will be described separately for a case where only objective function term Q1 (task completion date) is specified as the item of high importance in processing step S201 in Fig. 3, and a case where two objective function terms Q1 (task completion date) and Q2 (number of companies performing the work) are specified. Note that which objective function term is selected as the item of high importance is determined by the objective function setting unit 114 in Fig. 2 with reference to the information on the importance of each objective function term stored in the problem data storage unit DB1.
[0088] First, when only objective function term Q1 (task completion date) is set as a high-importance item and minimizing the task completion date is a prerequisite for creating a plan, in process step S305 in Fig. 6, the objective function H is optimized with only objective function term Q1, i.e., H = w1Q1. This can be achieved by substituting 0 for all of the weighting coefficients w2, w3, and w4 for the objective function H generated in process step S302. Also, when objective function term Q2 is not included in objective function H, equation (9), which is a constraint on the number of companies, does not need to be taken into account as a constraint condition.
[0089] Figure 8 is a Gantt chart of a work plan created from the solution of minimizing the objective function H (=w1Q1) described in equation (5) under the constraints described in equations (6) to (10). This work plan is created by minimizing the variable s ijd and the number of days required for each task d ij The solution can be created by referencing the value of the variable s ijd For example, if we take a variable that includes Company A as a subscript, then A社、作業04、4 =1, s A社、作業09、15 = 1, and all others are 0. Also, the value of variable D in the obtained solution is 23 days, and the overall work completion date is 23 days. In the rule-based work plan shown in Fig. 7, the work completion date is 24 days, and the work completion date has been shortened by one day through optimization in S305.
[0090] The mixed integer programming solver can guarantee that the solution obtained is an exact solution. Therefore, the obtained work completion date of 23 days is the exact minimum work completion date. In this way, when the mixed integer programming solver has completed the solution, a work plan with the minimum number of days can be created.
[0091] Referring again to FIG. 6, in process step S306, referring to the work plan obtained in process step S305, d max As a method of updating, for example, the value of d max In this case, at the time of processing step S303 in FIG. max = 24, but in processing step S306, max = 23, where d max The method of updating is not limited to the above method, and a value obtained by adding or subtracting a few days from the task completion date in the task plan obtained in processing step S305 may be substituted. maxThe method of updating may be specified by the user. The optimization process in process step S305 may be terminated midway, for example, by setting a limit on the calculation time. If the optimization process in the mixed integer programming solver is terminated midway, there is no guarantee that the solution obtained in process step S305 is an exact solution. Therefore, for example, in process step S306, max By making smaller than the previous value, the number of variables defined in process step S304 can be reduced, and as a result, the calculation time required to obtain an exact solution in the next process step S305 can be shortened.
[0092] In step S307 of FIG. 6, it is determined whether to end the solution by the mixed integer programming solver or to repeat the procedure of steps S304, S305, and S306. The determination conditions include, for example, whether an exact solution has been obtained in step S305, whether the objective function value has been improved since the previous solution was obtained, and the like. The determination conditions may be specified by the user, for example, d max The solution may be repeatedly found while decreasing or increasing the value of . In the following description, it is assumed that a solution with the strictly minimum task completion date is obtained in processing step S305.
[0093] In processing step S308 of FIG. 6, after the mixed integer programming solver has completed solving, a formulation for an Ising machine is generated. In the flowchart shown in FIG. 6, it is assumed that the optimization process S203 including items of low importance in FIG. 3 is executed by an Ising machine. In the formulation for an Ising machine, a method of incorporating the constraint equations shown in equations (6) to (10) into an objective function using a known Lagrangian function method or a method of using a known alternating direction multiplier method may be adopted. In addition, all constraint equations may be written as penalty terms and incorporated into the objective function. When the constraint conditions are written as penalty terms, for example, the constraint equation (6) may be expressed as equation (11), multiplied by a weighting coefficient w6, and added to the objective function.
[0094]
number
[0095] In processing step S309 of FIG. 6, the total work completion date is d max Define the variables so that they are within d max = 23, and for each company i and each operation j, the variable s ijd By obtaining the strictly minimum task completion date in the processing step S305 of FIG. 6, the variable s ijd It is possible to reduce the range of days for which the above is defined, thereby reducing the number of variables to be handled.
[0096] Also, the variable s ijd and the variable δ i may be defined only for companies that have been assigned at least one task in the tentative solution obtained in processing step S305. However, in the case of the tentative solution shown in Fig. 8, tasks are assigned to all five candidate companies, so in this embodiment, variables are defined for all companies. The case where variables are defined for only some of the companies in processing step S309 will be described later.
[0097] In processing step S310 of FIG. 6, optimization is performed using an Ising machine, taking into account the objective function terms that were not considered in processing step S305. In the optimization process in processing step S310, the tentative solution obtained in processing step S305 may be used as the initial solution. In this case, the newly added objective function terms are Q2, Q3, and Q4, and the objective function values multiplied by the weighting coefficients w2, w3, and w4 designated by the user are minimized. However, when constraint conditions are incorporated into the objective function as the Lagrangian function method, the alternating direction multiplier method, or penalty terms, the objective function values including the terms derived from the constraints are optimized. In this embodiment, in particular, the objective function terms Q3 and Q4 are quadratic objective functions, which are relatively difficult to handle by a mixed integer programming solver, and are objective function terms that the Ising machine is good at.
[0098] The weighting coefficient w of the objective function term used in the processing step S310 of FIG. kis the weighting coefficient w used in processing step S305 k For example, in process step S305, the objective function term Q1 relating to the task completion date is specified as the overall objective function H, but in process step S310, w1=0 may be set, and the task completion date may be excluded from the optimization target. This type of processing is effective in creating a task plan where minimizing the task completion date is a prerequisite. That is, the task completion date is strictly minimized at the time of process step S305, and this minimum task completion date is set as d in process step S306. max In step S309, the work plan is always set to d max If the variables are defined so that the work is completed by the minimum work completion date, the optimization process in process step S310 will always obtain a work plan in which all work is completed by the minimum work completion date. In this case, the objective function H does not need to include the objective function term Q1 related to the work completion date. Then, in process step S310, by performing optimization including the objective function term not considered in process step S305, it is possible to create a work plan that takes into account additional optimization indices such as the leveling out of the number of consecutive work tasks and the number of work days while maintaining the minimum work completion date.
[0099] Figure 9 shows the max This is an example of a work plan obtained by performing the optimization process of processing step S310 in Fig. 6 with an Ising machine, with 23 days set as the minimum work completion date for Q. In this example, the objective function H to be optimized consists of objective function term Q3 and objective function term Q4, and does not include objective function terms Q1 and Q2. In other words, the number of consecutive tasks (objective function term Q3) and leveling out the number of work days (objective function term Q4) are specified as items of low importance in processing step S203 in Fig. 3.
[0100] In the work plan example shown in Fig. 9, the work completion date is the 23rd, which is the same value as the minimum work completion date calculated in processing step S305 in Fig. 6. In addition, the number of consecutive executions is 1 (task 03 and task 04 by company A) in the work plan in Fig. 8, but after optimization processing using the Ising machine, it increases by 2 to 3 (task 03 and task 04 by company A, task 05 and task 09 by company C, and task 08 and task 10 by company E) in the work plan in Fig. 9.
[0101] This increases the number of consecutive tasks performed by the same company as much as possible within the scope of the work sequence, minimizing the number of unnecessary idle days for each company. To equalize the number of working days, it is sufficient to compare the variance of the number of working days for each company. In the work plan in Figure 8, the number of working days for each company is as follows: Company A: 12 days, Company B: 5 days, Company C: 2 days, Company D: 12 days, Company E: 11 days, and the variance of the number of working days between companies is 17.04.
[0102] On the other hand, as a result of optimization processing using an Ising machine, in the work plan in Figure 9, the number of working days for each company is Company A: 10 days, Company B: 5 days, Company C: 10 days, Company D: 12 days, and Company E: 11 days, and the variance in the number of working days between the companies has decreased from the plan in Figure 8 to 5.84. This has resulted in the burden between the companies being leveled as much as possible.
[0103] In this way, a plan can be created that first strictly minimizes the most important optimization index, the task completion date, and then adheres to the task completion date while also taking into account the additional optimization indexes of maximizing the number of consecutive tasks and minimizing the variance in the number of working days. In this embodiment, we assume that minimizing the task completion date is a necessary condition. This assumption is a case that frequently occurs in resource-constrained scheduling problems.
[0104] In the conventional method using an Ising machine, in order to minimize the work completion date, it was necessary to repeat the optimization process multiple times and heuristically search for an optimal solution. In particular, the number of times the optimization process is repeated to find the shortest work completion date is unknown, and furthermore, it is not possible to guarantee that the obtained work completion date is strictly the shortest. As illustrated in this embodiment, the work completion date is strictly minimized using a mixed integer programming solver, and then an optimization process using an Ising machine is performed including a quadratic objective function, so that the overall procedure is a single loop, and it is possible to create a plan that reliably minimizes the work completion date and also takes into account other optimization indices.
[0105] 6, it is determined whether to end the solution search by the Ising machine or to execute it again according to a previously specified criterion. The criterion may be, for example, whether all constraint conditions are satisfied, whether the objective function value is improved from the previous solution by the Ising machine, whether the optimization process by the Ising machine has been executed a number of times specified by the user, etc.
[0106] In process step S312 of FIG. 6, the operation plan obtained in process step S310 is referred to, and d max This is effective when minimizing the work completion date is not a necessary condition for creating a schedule, for example, when you want to prioritize minimizing other objective function terms such as the number of consecutive work operations or the variance in the number of working days between companies, even if it means delaying the overall work completion date.
[0107] Next, a detailed description will be given of a case where two objective function terms Q1 (task completion date) and Q2 (number of companies in charge of the work) are specified as optimization items with high importance in processing step S201 in Fig. 3. Explanations of procedures similar to those in the case where only objective function term Q1 (task completion date) is specified will be omitted as appropriate.
[0108] 6 again, in process steps S301 and S302, a rule-based tentative plan is created and a formulation for a mixed integer program solver is generated by the method already described. In process step S303 in FIG. 6, for example, 23 days, which is one less than the work completion date in the tentative plan (see FIG. 7) obtained in process step S301, is calculated as d max Suppose we substitute
[0109] Next, processing steps S304 and S305 in FIG. 6 are executed in the same manner as in the case where only the objective function term Q1 is considered. As a result, for example, a work plan as shown in FIG. 10 is obtained. In this work plan, the work completion date is the 23rd, and the number of companies in charge of the work is 5 (all of companies A to E). Here, the variable s ijdTherefore, it is not possible to create a schedule in which the work completion date is on the 24th day or later in the optimization process in process step S305.
[0110] In processing step S306 of FIG. max The value of is increased from the previous setting of 23 to, for example, 24. Then, in processing step S307 of FIG. 6, it is determined that the mixed integer programming solver is to be used again to find a solution. max The update method and the criteria for continuing the solution are, for example, "d max The value of d can be increased by 1 and the solution can be repeated. max You can also specify "increase the value of 1 and repeat twice". max does not need to be increased, and d max The optimization process may be repeated while decreasing .
[0111] FIG. 11 shows that in the second loop, d max This shows an example of a work plan calculated by a mixed integer programming solver under = 24 days. In this work plan, the work completion date is 24 days, and the number of companies in charge of the work is 4 (only companies A to D are in charge of the work, and company E is not in charge). When the weight coefficient w2 of the objective function term Q2 is sufficiently large compared to the weight coefficient w1 of the objective function term Q1, and reducing the number of companies is important, d max By changing and re-executing the optimization process, a plan can be obtained that reduces the number of companies even if it means extending the work completion date. However, if it is not necessary to strictly minimize the work completion date, d max Alternatively, the value of θ may be set to a sufficiently large value to reduce the number of times processing steps S304 to S307 are repeated.
[0112] After obtaining the operation plan shown in FIG. 11, in the process step S306 of FIG. maxFor example, it is assumed that the task completion date of 24th obtained in processing step S305 is directly substituted into . Furthermore, in processing step S307 in Fig. 6, it is assumed that it is determined to end the optimization process with the mixed integer programming solver, and in processing step S308, a formulation for the Ising machine is generated by the method already described.
[0113] In processing step S309 of FIG. 6, the total work completion date is d max Define the variable to be within (=24 days). Variable s ijd The range of the date d for defining the variable s can be determined by the same procedure as that of the processing step S304 in FIG. 6. In addition, if the reduction in the number of companies in charge, which is described in the objective function term Q2, is important, the variable s ijd and the variable δ i In the tentative solution ( FIG. 11 ) obtained in processing step S305, the variable is defined only for the companies that appear in the plan, i.e., only for Company A, Company B, Company C, and Company D, and it is not necessary to define the variable for Company E. In this case, since the variables themselves are defined only for four companies, Company A, Company B, Company C, and Company D, the work plan obtained by the optimization processing using the Ising machine will always have four or fewer companies in charge.
[0114] In processing step S310 of FIG. 6, optimization processing is performed using an Ising machine. At this time, for example, an objective function term Q3 related to continuous work is added as an additional objective function, and the objective function terms to be optimized are three terms w1Q1, w2Q2, and w3Q3. However, in processing step S309, the variable s ijd and the variable δ i is defined, the objective function term Q2 regarding the number of companies does not need to be included in the objective function H. Note that the objective function term added here does not need to be limited to Q3, and may be Q4 or other optimization indexes.
[0115] Figure 12 shows the maxThis is a work plan obtained by performing optimization processing using an Ising machine under the condition that Q1, Q2, and Q3 are added as objective function terms. Compared to the provisional solution (FIG. 11) obtained in processing step S305 of FIG. 6, the work completion date remains unchanged at 24 days, and the number of work companies remains unchanged at four companies in total. In processing step S309 of FIG. 6, variables are defined with reference to the provisional solution, and these optimization indices can be set so that they do not deteriorate compared to the provisional solution. However, due to the effect of the newly added objective function term Q3 related to consecutive work, the number of consecutive work operations increases from 3 in FIG. 11 to 4 in FIG. 12 while maintaining the values of the high-priority optimization indices of the work completion date and the number of companies.
[0116] After the work plan shown in FIG. 12 is obtained, it is determined in processing step S311 of FIG. 6 that the solution is to be re-executed using the Ising machine. Furthermore, in processing step S312, d max Assume that 26 days is set for d. These are the judgment conditions in processing step S311 and d max The method of updating d can be specified by the user in advance. For example, "Update d until the objective function value no longer improves." max In addition, when repeatedly solving the problem with an Ising machine, the solution obtained in the previous solving process may be used as the initial solution.
[0117] In step S309 of FIG. max Under the condition that the number of days is 26, variables are defined only for four companies, A to D. Furthermore, in processing step S310, optimization is performed using an Ising machine.
[0118] Figure 13 shows an example of a work plan calculated by the Ising machine under the above conditions. max The value of has been increased, so the work completion date has been extended to 26 days, but the number of companies remains unchanged at four, A to D. The number of consecutive tasks has also increased from four to seven. If you are willing to delay the work completion date a little in order to improve other objective function terms, you can use dmax Then, the optimization process can be performed again using the Ising machine by changing the value of
[0119] Although one embodiment has been described above in detail, the present invention is not limited to the above embodiment, and various modifications can be made without departing from the spirit of the present invention. For example, the above embodiment has been described in detail to explain the present invention in an easy-to-understand manner, and the present invention is not necessarily limited to those having all of the configurations described. In addition, it is possible to add, delete, or replace part of the configuration of the above embodiment with other configurations.
[0120] In addition, the above-mentioned configurations, functional units, processing units, processing means, etc. may be realized in part or in whole by hardware, for example, by designing them as integrated circuits. In addition, the above-mentioned configurations, functions, etc. may be realized in software by a processor interpreting and executing a program that realizes each function. Information such as the program, table, file, etc. that realizes each function can be stored in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), an IC card, an SD card, a DVD, or other recording medium.
[0121] In addition, in each of the above figures, the control lines and information lines are shown as those considered necessary for the explanation, and do not necessarily show all the control lines and information lines in the actual implementation. For example, it may be considered that almost all the components are actually connected to each other.
[0122] The above-described layout of the various functional units, various processing units, and various databases of the information processing device 100 is merely an example. The layout of the various functional units, various processing units, and various databases can be changed to an optimal layout in terms of the performance, processing efficiency, communication efficiency, and the like of the hardware and software included in the information processing device 100.
[0123] Furthermore, the configuration (schema, etc.) of the database that stores the various types of data described above can be flexibly changed from the perspective of efficient use of resources, improved processing efficiency, improved access efficiency, improved search efficiency, and the like. [Explanation of symbols]
[0124] 100: Information processing device 107: Combinatorial optimization device 111: Problem data acquisition section 112: Formulation generator 113: Variable definition section 114: Objective function setting section 115: Combinatorial Optimization Department 116: Optimization result judgment unit 117: Solution output unit DB: Storage section DB1: Problem data storage DB2: Formalized Data Storage DB3: Variable Data Storage DB4: Temporary solution storage DB5: Calculation program storage section
Claims
1. An information processing device that processes a combinatorial optimization problem consisting of a plurality of objective function terms using a computer, In a first stage, a combinatorial optimization unit performs an optimization process for objective function terms with high importance among a plurality of objective function terms, and a variable definition unit performs processing to reduce the number of variables to be defined by referring to a tentative solution obtained by the combinatorial optimization unit; In a second stage, the combinatorial optimization unit performs an optimization process for objective function terms that also take into account objective function terms with a relatively low degree of importance, and the optimization result determination unit determines whether to terminate the optimization process based on a specified determination criterion. When the optimization process is to be re-executed, the conditions for defining the variables are updated and the second stage optimization process is repeatedly executed.
2. 2. The information processing device according to claim 1, An information processing device characterized in that, in the first and second stages of optimization processing, objective function terms are divided into those with high importance and those with low importance according to the importance of each objective function term specified in advance by a user, and the combinatorial optimization unit selects the objective function terms to be optimized.
3. 2. The information processing device according to claim 1, The mathematical optimization methods used in the optimization process may be different from each other, and the information processing device is characterized by comprising a program for executing each mathematical optimization method and a combinatorial optimization device.
4. 4. The information processing device according to claim 3, An information processing device comprising: a formulation generation unit that formulates a combinatorial optimization problem in accordance with a mathematical optimization method used in each of the first and second optimization processing stages, and a storage unit that stores the formulation.
5. 2. The information processing device according to claim 1, An information processing apparatus characterized in that, before executing a first stage optimization process, a feasible provisional solution is obtained by a rule-based method different from the mathematical optimization method used in the first and second stages.
6. 2. The information processing device according to claim 1, 1. An information processing apparatus comprising: a first optimization process step and a second optimization process step, the first optimization process step and the second optimization process step being performed in a stepwise manner;
7. 2. The information processing device according to claim 1, An information processing device characterized by having a variable definition unit that, when reducing the number of variables to be defined, prunes variables whose values can be analytically determined based on information obtained from feasible tentative solutions obtained in advance and constraints imposed on the combinatorial optimization problem, without defining the variables.
8. 2. The information processing device according to claim 1, An information processing device comprising a variable definition unit that, when reducing the number of variables to be defined, prunes some variables without defining them by referring to information obtained from feasible provisional solutions obtained in advance and the importance of each objective function term.
9. 6. The information processing device according to claim 5, An information processing device characterized in that, prior to the first and second stages of optimization processing, pruning of variables is performed by referring to the feasible tentative solution, and the optimization processing is repeatedly performed by changing the conditions used for pruning of variables depending on the optimization processing results, the constraint conditions, and the importance of the objective function terms.
10. 1. An information processing method for processing a combinatorial optimization problem consisting of a plurality of objective function terms using a computer, comprising: In a first step, an optimization process is performed for objective function terms having high importance among a plurality of objective function terms, and a process of reducing the number of variables to be defined is performed by referring to a provisional solution obtained by the optimization process; In a second step, an optimization process is performed for the objective function terms taking into account objective function terms with a relatively low degree of importance, and a process is performed to determine whether or not to terminate the optimization process based on a specified determination criterion. When the optimization process is to be re-executed, the conditions for defining the variables are updated and the second stage optimization process is repeatedly executed.
Citation Information
Patent Citations
Optimization device, optimization method, and optimization program
JP2021190019A
