Planning and organizing plan making device, planning and organizing plan making method using the planning and organizing plan making device, and computer program

Abstract

To determine a plan and arrangement proposal on the basis of any KPI priority information.SOLUTION: An operation arrangement proposal generation device 10 comprises: a learning data generation part 132 for giving corresponding KPI priority information to each history of an operation arrangement proposal which is a modification method proposal of a plan and generating learning data; and an operation arrangement proposal generation part 1200 for determining the operation arrangement proposal maximizing a predetermined object function on the basis of the plan needed to be modified and any KPI priority information by using an operation arrangement restriction model 1202 having the learning data by machine learning as teacher data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a project consolidation planning device, a project consolidation planning method using the project consolidation planning device, and a computer program. [Background technology]

[0002] The present invention preferably relates to a train schedule creation device and a train schedule creation method.

[0003] One of the tasks in railway operation management is to correct disruptions to the schedule due to delays and other factors, and to revert to the planned schedule. In this task, it is necessary to change the schedules of many trains by combining several rescheduling operations to return them to the planned schedule.

[0004] There are many ways to return to the planned timetable, but it is necessary to select the optimal rescheduling method (rescheduling plan) by taking into account multiple KPIs (Key Performance Indexes). For example, minimizing the total delay time, which is the sum of the delay times for each train at each station, is important from the perspective of passenger satisfaction. In addition, reducing the number of rescheduling operations must also be considered from the perspective of operational costs. These KPIs cannot necessarily be maximized simultaneously, and which KPI should be prioritized will differ depending on the situation.

[0005] Generally, train rescheduling work involves experienced commanders determining priority KPIs based on their experience, and then creating train rescheduling plans to meet those KPIs. In planning problems like this, where there are multiple KPIs, and the number of KPIs is large and complex, or where the planner's decisions are based on their experience and intuition, it is difficult to create a plan using a computer, and the resulting plan is unlikely to meet the user's requirements.

[0006] Therefore, it is conceivable to create a system in which KPI priorities are externally set by a commander and train rescheduling plans are automatically proposed based on that. To achieve this, a model is needed that somehow restricts the search method for rescheduling plans for the specified KPIs. When creating such a model, it is necessary to use past data. To use past data in model construction, information on KPIs that were prioritized in the past is required, but such data is not generally stored.

[0007] To address this issue, there is a method for estimating so-called tacit constraint conditions from past data. For example, the technology described in Patent Document 1 estimates the priority of constraints to be relaxed from past planning data, and describes that "a production planning device and a production planning method are provided that are capable of formulating an appropriate production plan that is in line with the actual situation on-site," and that "the production planning processing unit relaxes the constraints in order of the relaxation priority based on the prediction results, thereby formulating a production plan that satisfies all of the multiple constraints." [Prior art documents] [Patent documents]

[0008] [Patent Document 1] Japanese Patent Application Publication No. 2018-120342 Summary of the Invention [Problem to be solved by the invention]

[0009] The above-mentioned Patent Document 1 learns the priority of relaxing constraints on KPIs that are common to all past data, and creates a plan based on that priority. However, this document does not address situations where the priority of KPIs differs for individual events, and is therefore unable to estimate the priority of KPIs for individual past events.

[0010] The present invention has been made in consideration of the above-mentioned problems, and aims to provide a planning arrangement drafting device that can determine a planning arrangement based on arbitrary KPI priority information, a planning arrangement drafting method using the planning arrangement drafting device, and a computer program. [Means for solving the problem]

[0011] In order to solve the above-mentioned problems, a plan reorganization device according to one aspect of the present invention includes a learning data generation unit that generates learning data by assigning corresponding KPI priority information to each of the histories of plan reorganization proposals, which are plan revision method proposals, and a plan reorganization decision unit that determines a plan reorganization proposal that maximizes a predetermined objective function based on the plan that needs to be revised and arbitrary KPI priority information, using a plan reorganization proposal restriction model that has been machine-learned using the learning data as training data. Equipped with. [Effects of the Invention]

[0012] According to the present invention, it is possible to realize a plan reorganization device capable of determining a plan reorganization plan based on arbitrary KPI priority information, a plan reorganization method using the plan reorganization device, and a computer program. [Brief explanation of the drawings]

[0013] [Figure 1] FIG. 1 is a diagram illustrating an example of the configuration of a timetable replanning proposal generating device, which is a plan replanning proposal generating device according to an embodiment. [Figure 2] FIG. 10 is a diagram showing an example of a KPI achievement level output by a schedule evaluation unit in the embodiment. [Figure 3] FIG. 2 is a diagram showing an example of a planned timetable in an embodiment. [Figure 4] FIG. 10 is a diagram showing an example of a timetable rescheduling pattern database in the embodiment. [Figure 5] FIG. 10 is a diagram showing an example of a train schedule rescheduling plan in an embodiment. [Figure 6] FIG. 10 is a diagram showing the flow of the overall processing by the inference unit in the embodiment. [Figure 7]FIG. 2 is a diagram showing an example of the configuration of a past timetable rescheduling database in the embodiment. [Figure 8] FIG. 10 is a diagram showing the flow of the overall processing in the model generation unit in the embodiment. [Figure 9] FIG. 10 is a diagram showing a flow of a learning data generation process performed by a model generation unit in the embodiment. [Figure 10] FIG. 10 is a diagram showing a flow of a perturbation addition process performed by a model generation unit in the embodiment. [Figure 11] FIG. 10 is a diagram showing the flow of a KPI priority estimation process performed by a model generation unit in the embodiment. [Figure 12] FIG. 10 is a graph showing an example of a planned timetable stored in a past database in the embodiment. [Figure 13] FIG. 10 is a graph showing an example of a delayed timetable stored in a historical database in an embodiment. [Figure 14] FIG. 10 is a graph showing an example of a past performance timetable stored in a past database in the embodiment. [Figure 15] FIG. 10 is a graph showing an example of a revised timetable after adding perturbations stored in a past database in an embodiment. [Figure 16] FIG. 10 is a diagram showing an example of a value of KPI achievement level obtained by evaluating a past performance timetable stored in a past database and a revised timetable after adding perturbations in an embodiment. [Figure 17] FIG. 10 is a diagram showing the region of the KPI priority parameter that is restricted by an inequality obtained by the perturbation process by the learning data generation unit in the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0014] Hereinafter, embodiments of the present invention will be described with reference to the drawings. Note that the embodiments described below do not limit the scope of the invention as claimed, and not all of the elements and combinations thereof described in the embodiments are necessarily essential to the solution of the invention.

[0015] In the drawings explaining the embodiments, parts having the same functions are given the same reference numerals, and repeated explanations thereof will be omitted.

[0016] In the following description, expressions such as "xxx data" may be used as an example of information, but the data structure of the information may be any. In other words, to indicate that the information does not depend on the data structure, "xxx data" may be referred to as "xxx table." Furthermore, "xxx data" may be simply referred to as "xxx." In the following description, the structure of each piece of information is an example, and the information may be stored divided or combined.

[0017] In the following explanation, processing may be described using a "program" as the subject, but since a program is executed by a processor (e.g., a CPU (Central Processing Unit)) to perform a predetermined process using storage resources (e.g., memory) and / or communication interface devices (e.g., ports) as appropriate, the subject of the processing may also be the program. Processing described using a program as the subject may also be processing performed by a processor or a computer having that processor.

[0018] In the following explanation, when the subject of an operation is described as "the ○○ unit," it means that the processor reads the processing content of the ○○ unit, which is a program, from memory, loads it into memory, and then realizes the function of the ○○ unit (details described below).

[0019] The plan reorganization proposal formulation device of this embodiment has the following configuration, for example.

[0020] The timetable rescheduling proposal generation device, which is a planning and rescheduling proposal generation device of this embodiment, is a timetable creation device that proposes timetable rescheduling proposals based on specified KPIs, and is characterized by having a part that estimates and saves the KPI priority at that time from individual timetable rescheduling history in past data, a model that assists in generating rescheduling proposals based on KPI priority information learned using the saved data, and a part that generates rescheduling proposals.

[0021] According to this embodiment, it is possible to propose a timetable rescheduling plan that takes into consideration the designated KPI priority. [Example]

[0022] This embodiment will be described below with reference to FIGS.

[0023] The timetable rescheduling proposal device of this embodiment, which is a plan rescheduling proposal device, proposes a new plan to resolve a delay situation when a delay occurs in train operation.When multiple KPIs need to be considered when resolving a delay situation and the KPI priorities are provided from outside the system, the timetable rescheduling proposal device of this embodiment proposes a timetable rescheduling proposal that maximizes an objective function based on the KPI priorities using a model that limits the search space for rescheduling proposals according to the KPI priorities.The timetable rescheduling proposal device of this embodiment includes a model generation unit that creates the model, and the model generation unit includes a learning data generation unit that estimates and supplements data related to KPI priorities required for model generation from past timetable rescheduling data.

[0024] FIG. 1 shows an example of the configuration of a timetable replanning plan generating device 10.

[0025] The timetable rescheduling plan generation device 10 includes, as hardware, a storage device 101 such as RAM (Random Access Memory), ROM (Read Only Memory), HDD (Hard Disk Drive), SSD (Solid State Drive), etc., an input device 102 consisting of a keyboard, mouse, touch panel, etc., an output device 103 consisting of a monitor, printer, etc., a processor 104 such as a CPU (Central Processing Unit) that executes programs stored in the storage device, and a memory 105 consisting of RAM, etc.

[0026] The storage device 101 stores a predictive timetable simulator 110, a train rescheduling operation database 111, an objective function 112, an inference unit 120, and a model generation unit .

[0027] The processor 104 executes a program stored in the storage device 101 to implement a function of generating a train schedule rescheduling according to the KPI priority 1208 described later.

[0028] The predicted timetable simulator 110 predicts and outputs the actual timetable based on the timetable information and information on plan changes corresponding to the timetable information.

[0029] Timetable information is information that shows train operation plans used to control train fleets. Timetable information includes at least the arrival and departure times, or passing times, of all trains at all stations. Specific timetable information is explained in detail in Figure 3.

[0030] The information on the plan change is the delay status data 1203 , which will be described later, or the timetable rescheduling proposal 1207 , or the timetable rescheduling proposal 1414 stored in the past timetable rescheduling database 1400 .

[0031] The predictive timetable simulator 110 holds information about constraints that the output timetable must satisfy. Constraints include, for example, minimum stop times at stations and minimum time intervals between trains. The predictive timetable simulator 110 outputs a predictive timetable that satisfies the constraints.

[0032] The timetable evaluation unit 113 calculates the index value of a predetermined evaluation index for the revised timetable 1206 based on the planned timetable 1204 and the post-delay predicted timetable 1205. Hereinafter, the index value output by the timetable evaluation unit 113 will be referred to as the KPI achievement degree.

[0033] The KPI achievement level consists of at least one or more numerical data, and when there are multiple KPIs (evaluation indexes), it becomes a numerical vector. For example, when there are three evaluation indexes, the KPI achievement level outputs a vector consisting of three-dimensional real numbers. In this embodiment, the case where there are three types of evaluation indexes is handled as an example, but the number of evaluation indexes in the present invention is not limited. A specific example of the KPI achievement level will be described later using FIG. 2.

[0034] The timetable rescheduling operation database 111 is a list of all available timetable rescheduling operations. A timetable rescheduling operation (hereinafter simply referred to as timetable rescheduling) is an operation that changes the schedule of at least one train. For example, an operation that delays the departure time of a train at a certain station is one example of a timetable rescheduling operation. Details of the timetable rescheduling operation database 111 will be described later with reference to FIG. 4.

[0035] The objective function 112 is a function that returns a final evaluation value based on the KPI achievement level, which is the output of the diagram evaluation unit 113. The objective function 112 is a function that uses the value of the KPI priority 1208, which will be described later, as a parameter. For example, if the KPI achievement level is given by three evaluation indexes KPI1, KPI2, and KPI3, and the KPI priority 1208 is given by three variables w1, w2, and w3, the objective function 112 is defined as a function expressed as w1*KPI1+w2*KPI2+w3*KPI3.

[0036] The inference unit 120 outputs a timetable rescheduling proposal 1207 based on delay status data 1203 and KPI priority 1208. The inference unit 120 includes a timetable rescheduling proposal generation unit 1200, a timetable rescheduling restriction model 1202, delay status data 1203, a planned timetable 1204, a post-delay predicted timetable 1205, a revised timetable 1206, a timetable rescheduling proposal 1207, a KPI priority 1208, and a timetable rescheduling operation list 1209. Specific processing details of the inference unit 120 will be explained later using FIG. 6.

[0037] The configuration of the inference unit 120 will be described below.

[0038] The delay status data 1203 is data that indicates the actual delay status at the site. For example, when an accident causes a situation in which a certain train must stop for a certain period of time at a certain station, information specifying the train in question on the planned timetable, information specifying the station in question on the planned timetable, and information on the expected delay time are input. The delay status data 1203 contains information necessary for the timetable simulator 110 to read it and output a post-delay predicted timetable 1205.

[0039] Planned timetable 1204 is timetable information at the time of planning before the occurrence of the delay. A specific example of the timetable will be explained later with reference to FIG.

[0040] The post-delay predicted timetable 1205 is timetable information after a delay that is predicted to be realized if a delay corresponding to the delay status data 1203 occurs in relation to the planned timetable 1204 and no train rescheduling is implemented. The post-delay predicted timetable 1205 is created using the predicted timetable simulator 110 based on the planned timetable 1204 and the delay status data 1203.

[0041] The revised timetable 1206 is a timetable that is predicted to be realized when a timetable rescheduling plan 1207 generated by the timetable rescheduling plan generating unit 1200 is applied to the post-delay predicted timetable 1205. The revised timetable 1206 is generated using the predicted timetable simulator 110 based on the post-delay predicted timetable 1205 and the timetable rescheduling plan 1207.

[0042] The timetable replanning plan 1207 is a list of timetable replanning operations. An example of the implementation of the timetable replanning plan 1207 will be explained later using Figure 5. The timetable replanning plan 1207 is created by the timetable replanning plan generation unit 1200 in step s3020 of Figure 6.

[0043] The KPI priority 1208 is a parameter that defines the objective function 112. It is assumed that the KPI priority 1208 is determined by a commander and input using an input device, but it may also be determined by another external algorithm and input using an input device.

[0044] The timetable replanning operation list 1209 is a list that defines timetable replanning operations that the timetable replanning plan generation unit 1200 uses when creating the timetable replanning plan 1207. For example, it is a list that specifies some of the timetable replanning operations defined in the timetable replanning operation database 111. It is generated by the timetable replanning restriction model 1202, which will be described later.

[0045] The timetable rescheduling plan generation unit 1200 uses a timetable rescheduling operation list 1209 to output a timetable rescheduling plan 1207 that maximizes the objective function 112 based on the planned timetable 1204 and the post-delay predicted timetable 1205. Specifically, for example, the timetable rescheduling plan generation unit 1200 calculates the KPI achievement rate for all timetable rescheduling plans 1207 that can be created by combining the timetable rescheduling operation list 1209, and outputs the one that maximizes it.

[0046] The traffic rescheduling restriction model 1202 is a model that creates a traffic rescheduling operation list 1209 based on a post-delay predicted timetable 1205, a planned timetable 1204, and a KPI priority 1208. The traffic rescheduling restriction model 1202 is executed to estimate a traffic rescheduling operation list 1209 that is necessary and sufficient to maximize the objective function 112.

[0047] The model generation unit 130 generates a timetable rescheduling restriction model 1202 based on past data. The model generation unit 130 includes a model learning unit 131, a learning data generation unit 132, a learning database 133, and a past timetable rescheduling database 134.

[0048] The past timetable replanning database 134 is a database that stores one or more pieces of past timetable replanning data 1401, which will be described later. The past timetable replanning data 1401 includes information about the history of timetable replanning that has been carried out in the past. An example of the past timetable replanning data 1401 will be described later using FIG. 6.

[0049] The learning database 133 is a database that stores one or more pieces of learning data, which will be described later. The learning data is a set of data that includes information on the past timetable adjustment data 1401 and KPI priority information at that time.

[0050] The learning data generation unit 132 creates learning data based on information in the past timetable adjustment database 134. The learning data generation unit 132 is executed to estimate and add KPI priority information when it is not included in the past timetable adjustment data 1401. Details of the processing by the learning data generation unit 132 will be explained in detail later using FIG. 9.

[0051] The model learning unit 131 generates a timetable rescheduling restriction model 1202 using the learning data stored in the learning database 133. Specifically, for example, the model learning unit 131 creates a model that associates KPI priorities with the types of timetable rescheduling used, based on the planned timetable 1204 in the learning data.

[0052] 2 shows an example of KPIs output by the timetable evaluation unit 113. In this embodiment, the timetable evaluation unit 113 outputs three KPIs for delay time, train intervals, and the number of train order changes.

[0053] As shown in Figure 2, the KPI (201) for delay time is an indicator that takes a larger value the shorter the delay time, and the range of possible values ​​is from 0 to 1. For example, it is the improvement rate of the total delay time of the revised timetable relative to the total delay time of the original delayed timetable.

[0054] The KPI (202) relating to train intervals is an index whose possible values ​​range from 0 to 1 and whose value increases as the train intervals at a station become more uniform, for example, the improvement rate of the variance of train intervals.

[0055] The KPI (203) relating to the number of order changes is an index that takes a larger value the fewer the number of order changes, and the range of possible values ​​is from 0 to 1. For example, if no order changes are made, the index is 1, and decreases by 0.2 each time the order is changed, becoming 0 if it is made five times or more.

[0056] In this embodiment, three KPIs shown in FIG. 2 are used to provide a simple example, but this does not limit the types or number of KPIs of the present invention.

[0057] An example of the planned timetable 1204 is shown in Fig. 3. In the example of the planned timetable 1204 shown in Fig. 3, four trains with train IDs (211) 1 to 4 are given as a table corresponding to arrival times (213), departure times (214), and platform numbers (215) used at seven stations with station IDs (212) A to G.

[0058] The timetable information is not limited to these, and may also include, for example, information about train connections, information about crew members, and information about passing stations.

[0059] Fig. 4 shows an example of the timetable replanning operation database 111. In the example of the timetable replanning operation database 111 shown in Fig. 4, five types of timetable replanning are available, and for each timetable replanning operation, a timetable replanning ID (221), a character string (222) of the operation content, and flag information (223) indicating information required to execute the timetable replanning operation are provided as a corresponding table.

[0060] The flag information required to execute a traffic rescheduling operation is, for example, one row assigned to each candidate piece of information required for each traffic rescheduling operation, with "1" entered in the row corresponding to the item and "0" entered in the row corresponding to an item that is not required.

[0061] In the example of Figure 4, the candidates for necessary information are the train's timetable ID (2231), the station ID (2232), the target train ID (2233), and the time of day (2234) for the operation. For example, when executing a timetable rescheduling operation (224) indicated by a timetable rescheduling ID of 1, the train ID, station ID, and target train ID columns are filled with 1, indicating that these three pieces of information are required. In this embodiment, five types of timetable rescheduling are considered as shown in Figure 4 to provide a simple example, but this does not limit the types of timetable rescheduling in the present invention.

[0062] 5 shows an example of a timetable rescheduling plan 1207. In this example, the timetable rescheduling plan 1207 is a table that lists a column (230) in which the ID of the timetable rescheduling operation is entered, and columns (231, 232, 233, 234) corresponding to the information required for the timetable rescheduling operation.

[0063] In each row of the table in Figure 5, a traffic rescheduling operation ID and the information required for that operation are entered, and in columns where no information is required, a different symbol (in this example, "-") is entered to indicate that no information is required. For example, the traffic rescheduling operation in the first row indicates that a traffic rescheduling operation with a traffic rescheduling ID of 1 will be performed, and the train ID, station ID, and target train ID information are stored as information required for the operation, but the time is not required, so a "-" symbol is entered.

[0064] 6 shows an example of the past train rescheduling database 134. The past train rescheduling database 134 is data that includes one or more pieces of past train rescheduling data 1401. The past train rescheduling data 1401 includes information on train rescheduling that has been carried out in the past, and includes, for example, information on a planned train timetable 1411, a predicted train timetable after delay 1412, a revised train timetable 1413, and an implemented train rescheduling plan 1414 in a train timetable revision that has been carried out in the past.

[0065] 7 is a flowchart illustrating the timetable replanning plan generation process performed by the inference unit 120. The timetable replanning plan generation process is a process for proposing and outputting a timetable replanning plan 1207 based on data stored in the storage device 101.

[0066] The inference unit 120 first acquires the delay status data 1203 and the KPI priority information 1208 (s3005). The delay status data 1203 and the KPI priority information 1208 are input, for example, by a commander via the input device 102. Next, the inference unit 120 generates a post-delay predicted timetable 1205 using the timetable simulator 110, the planned timetable 1204, and the delay status data 1203 (s3010). Next, the inference unit 120 creates a timetable rescheduling operation list 1209 using the timetable rescheduling restriction model 1202, the post-delay predicted timetable 1205, the planned timetable 1204, and the KPI priority 1208 (s3015). Next, the inference unit 120 generates a timetable rescheduling plan 1207 using the timetable rescheduling plan generation unit 1200 based on the planned timetable 1204, the post-delay predicted timetable 1205, and the objective function 112 (s3020). Next, the inference unit 120 uses the timetable simulator 110 to generate a revised timetable 1206 based on the planned timetable 1204, the predicted timetable after delay 1205, and the timetable rescheduling proposal 1207 (s3025). Next, the inference unit 120 outputs the timetable rescheduling proposal 1207 and the revised timetable 1206 to the output device 103 (s3030), and the process ends.

[0067] FIG. 8 is a flowchart illustrating the model generation process performed by the model generation unit 130.

[0068] The model generation unit 130 first performs a learning data generation process (s3105).

[0069] The learning data generation process s3105 is a process of creating learning data to be used for learning based on the data in the past database 1334 and storing the data in the learning database 133. An example of the specific processing content of the learning data generation process s3105 will be described in detail later with reference to Figs. 9 to 11.

[0070] Next, the model generation unit 130 generates a model using the model learning unit 131 and the learning data in the learning database 133 (s3110). Next, the model generation unit 130 saves the obtained model as the timetable rescheduling restriction model 1202 (s3115), and ends the processing.

[0071] An example of the learning data generation process (s3105) will be described using Figures 9 to 11. In this example, a change (perturbation) is made to the timetable rescheduling plan 1414 included in the past data, the resulting change in the achieved KPI is calculated, and based on that information, the KPI priority when the timetable rescheduling plan 1414 was created in the past is estimated and saved as learning data.

[0072] FIG. 9 is a flowchart showing an example of the learning data generation process performed by the learning data generation unit 132.

[0073] First, the learning data generation unit 132 acquires one piece of past timetable adjustment data 1401 from the past timetable adjustment database 134 and stores it in the memory 105 or the like (s3205).

[0074] Next, the learning data generation unit 132 executes a perturbation addition process (s3210) on the acquired past timetable replanning data 1401. The perturbation addition process (s3210) is a process of making at least one change to a previously executed timetable replanning plan 1414, generating a revised timetable using the timetable replanning plan after the change, and calculating the achievement KPI. A specific example of this process will be explained later using FIG. 10.

[0075] Next, the learning data generation unit 132 executes a KPI priority estimation process (s3215). The KPI priority estimation process (s3215) is a process for estimating KPI priority using the past timetable adjustment data 1401 acquired in the past data acquisition process (s3210) and the achieved KPI for the revised timetable obtained in the perturbation addition process. A specific example of this process will be described later with reference to FIG. 11.

[0076] Next, the learning data generation unit 132 stores the KPI priority estimated in the KPI priority estimation process (s3215) and the past timetable adjustment data 1401 acquired in the past data acquisition process (s3210) in the learning database 133 (s3220).

[0077] Next, the learning data generation unit 132 determines whether the above process has been completed for all data in the past timetable adjustment database 134 (s3225), and if not, returns to step s3205. If the above process has been completed for all data, the learning data generation process ends.

[0078] FIG. 10 is a flowchart showing an example of the perturbation addition process (s3210).

[0079] First, the learning data generation unit 132 initializes a counter n and sets the required number of data N (s3300). Next, the learning data generation unit 132 generates a post-perturbation timetable replanning proposal by making some kind of change (perturbation) to the timetable replanning proposal 1414 of the past timetable replanning data 1401, and records the perturbed timetable replanning proposal in a memory or the like (s3305).

[0080] A perturbation is the addition of one or more changes to a past timetable rescheduling plan. For example, adding one new timetable rescheduling operation to a past timetable rescheduling plan is an example of a perturbation.

[0081] Next, the learning data generation unit 132 uses the predictive timetable simulator 110 to obtain a new revised timetable based on the planned timetable 1411 in the past timetable rescheduling data 1401, the predicted timetable after delay 1412 in the past timetable rescheduling data 1402, and the timetable rescheduling proposal after perturbation (s3310).

[0082] Next, the learning data generation unit 132 uses the timetable evaluation unit 113 to obtain the KPI achievement degree using the planned timetable 1411 in the past timetable rescheduling data 1401, the predicted timetable after delay 1412 in the past timetable rescheduling data 1401, and the newly revised timetable obtained by adding perturbations (s3315).

[0083] Next, the learning data generation unit 132 compares the KPI achievement degree in the past traffic rescheduling data 1401 with the KPI achievement degree calculated for the new revised timetable, and determines whether the new revised timetable obtained by adding the perturbation is useful for KPI priority (s3320). The criterion for determining usefulness is, for example, that if all KPIs decrease due to the perturbation, it is determined to be unuseful.

[0084] If the learning data generation unit 132 determines that the perturbation is not useful, the process returns to step s3305 and generates a new post-perturbation reorganization plan. If the learning data generation unit 132 determines that the perturbation is useful, the counter n is incremented by 1 and the post-perturbation KPI achievement information is stored in memory (s3325).

[0085] Next, if the number of the counter is equal to or greater than N set in s3300, the process ends. If the number of the counter is not equal to or greater than N, the process returns to step s3305 for generating perturbation conditions.

[0086] FIG. 11 is a flowchart showing an example of the KPI priority estimation process (s3215).

[0087] The learning data generation unit 132 creates a conditional expression related to KPI priority using the N post-perturbation KPI achievement degrees and the KPI achievement degree of the past actual performance schedule (s3405). For example, assuming that the past actual performance schedule maximizes the objective function 112, the learning data generation unit 132 obtains an inequality that the objective function evaluation value calculated using the KPI achievement degree for the past actual performance schedule is greater than the objective function evaluation value calculated using the post-perturbation KPI achievement degree. N conditional expressions are obtained for the N post-perturbation KPI achievement degrees.

[0088] Next, the training data generation unit 132 searches for a range of solutions that satisfy these inequalities (s3410). If the training data generation unit 132 does not find a solution, it selects one of the conditional expressions, removes it, and returns to the search step s3410 (s3420). If a solution is found, it determines a representative point of the solution and stores it in the training database 133 as an estimated KPI priority (s3425).

[0089] Specific examples of how to obtain estimated values ​​of KPI priorities in the learning data generation process will be described below with reference to FIGS.

[0090] In the following specific example, the diagram evaluation unit 113 calculates the KPIs (hereinafter referred to as KPI1, KPI2, and KPI3), which are the three evaluation indexes described in Fig. 2, as the KPI achievement level. Also, in the following specific example, the KPI priority, which is a parameter of the objective function 112, is three weight parameters w1, w2, and w3 corresponding to the importance of each KPI achievement level, and the objective function 112 is a function given by w1*KPI1+w2*KPI2+w3*KPI3. The KPI priorities w1, w2, and w3 satisfy w1+w2+w3=1.

[0091] The purpose of the learning data generation process shown in Figure 9 is to estimate the KPI priority parameters w1, w2, and w3 when a timetable change is made from the past planned timetable 1411, the predicted timetable after delay 1412, the past corrected timetable 1413, and the past timetable change proposal 1414 in the past timetable change data 1401.

[0092] FIG. 12 shows an example of a planned timetable 1204 depicted on a graph. In FIG. 12, the horizontal axis represents time, the vertical axis represents stations, and the solid lines indicate the position of each train at each time. The same applies to FIGS. 13 to 15 below. FIG. 12 depicts the planned timetable 1204 of FIG. 3 on a graph, and the solid lines 4001, 4002, 4003, and 4004 represent the planned timetables of trains represented by train IDs 1, 2, 3, and 4 in the timetable of FIG. 3, respectively. Below, these will be referred to as train 1, train 2, train 3, and train 4, respectively. The lines representing such train schedules will be called train routes.

[0093] Figure 13 shows an example of a post-delay timetable. Figure 13 is a post-delay timetable predicted if train 3 is delayed by 45 minutes (4100) at station D. As shown in the train path after the delay for train 3 (4013), train 3 is delayed at all stations from station D onwards. As a result of the delay for train 3, train 4 is also delayed (4014). In this case, the delays at the terminal stations are 0 minutes for train 1, 0 minutes for train 2, 45 minutes for train 3, and 40 minutes for train 4, for a total of 85 minutes. Comparing the train intervals at the terminal stations, the interval between trains 1 and 2 is 15 minutes, between trains 2 and 3 is 60 minutes, and between trains 3 and 4 is 10 minutes.

[0094] FIG. 14 shows an example of a past revised timetable 1413 in the past train rescheduling data 1401. In FIG. 13, train 4 changes its departure order with train 3 at station D (4110), eliminating the delay of train 4 (4024) and improving the total delay time. Specifically, train 1 is 0 minutes, train 2 is 0 minutes, train 3 is 0 minutes, and train 4 is 45 minutes, for a total of 45 minutes. The train intervals are 15 minutes between train 1 and train 2, 30 minutes between train 2 and train 3, and 30 minutes between train 3 and train 4.

[0095] Figure 15 shows the post-perturbation timetable when delays 4201 and 4202 are added to train 2 in addition to the original rescheduling plan in step s3305. Train route 4032 for train 2 is delayed by 10 minutes at the end point. The total delay from trains 1 to 4 is 55 minutes, which is an increase from the revised timetable for the past data shown in Figure 14. On the other hand, the train intervals at the end points are 20 minutes between trains 1 and 2, 25 minutes between trains 2 and 3, and 30 minutes between trains 3 and 4, which is closer to being even than the revised timetable shown in Figure 14.

[0096] Figure 16 shows an example of KPIs obtained by evaluating Figures 14 and 15. Figure 16 shows an example in which the result of evaluating the past performance timetable shown in Figure 13 was (KPI1, KPI2, KPI3) = (0.6, 0.7, 0.5), and the result of evaluating the perturbed timetable shown in Figure 15 was (KPI1, KPI2, KPI3) = (0.5, 0.8, 0.5).

[0097] Figure 17 is a diagram illustrating on a plane the range of possible values ​​of w1 and w2 that are restricted by the inequality obtained in step s3405 using the KPI obtained in the example of Figure 16. In step s3405, the learning data generation unit 132 substitutes the KPIs of the revised timetable and the perturbed revised timetable in the example of Figure 15 into the objective function 112 and compares them, thereby obtaining the following inequality. w1*0.5+w2*0.7+w3*0.8>w1*0.4+w2*0.8+w3*0.5

[0098] The learning data generation unit 132 rearranges the above inequality to obtain w1 > w2. In step s4310, the learning data generation unit 132 estimates from this inequality that the value falls within the region indicated by 5110 in FIG. 17, and in step 3245 selects a representative point within this region, for example, randomly, and terminates the KPI priority estimation process. The learning data generation unit 132 can obtain multiple inequalities by creating multiple perturbation diagrams and performing similar processing, thereby further restricting the values ​​that the KPI priority can take and enabling more accurate estimation.

[0099] As described above in detail, the timetable replanning proposal generation device 10 of this embodiment makes it possible to determine a timetable replanning proposal based on arbitrary KPI priority information.

[0100] The above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to those including all of the described components. Furthermore, some of the components of each embodiment can be added to, deleted from, or replaced with other components.

[0101] Furthermore, the above-described configurations, functions, processing units, processing means, etc. may be partially or entirely implemented in hardware, for example, by designing them as integrated circuits. The present invention can also be realized by software program code that implements the functions of the embodiments. In this case, a storage medium on which the program code is recorded is provided to a computer, and a processor included in the computer reads the program code stored in the storage medium. In this case, the program code itself read from the storage medium implements the functions of the above-described embodiments, and the program code itself and the storage medium on which it is stored constitute the present invention. Examples of storage media for providing such program code include flexible disks, CD-ROMs, DVD-ROMs, hard disks, solid-state drives (SSDs), optical disks, magneto-optical disks, CD-Rs, magnetic tapes, non-volatile memory cards, and ROMs.

[0102] Furthermore, the program code that realizes the functions described in this embodiment can be implemented in a wide range of program or script languages, such as assembler, C / C++, perl, Shell, PHP, Java (registered trademark), and Python.

[0103] Furthermore, all or part of the program code of the software that realizes the functions of each embodiment may be stored in the memory device 101 in advance, or, if necessary, may be stored in the memory device 101 from a non-temporary storage device of another device connected to the network, or from a non-temporary storage medium via an external I / F (not shown) provided in the timetable replanning plan generation device 10.

[0104] Furthermore, the program code of the software that realizes the functions of the embodiments may be distributed via a network and stored in a storage means such as a computer's hard disk or memory, or in a storage medium such as a CD-RW or CD-R, and the processor of the computer may read and execute the program code stored in the storage means or storage medium.

[0105] In the above-described embodiment, the control lines and information lines are shown as those considered necessary for the explanation, and not all control lines and information lines are necessarily shown in the product. All components may be interconnected.

[0106] Another example of an embodiment of the present invention is a system for automatically changing a production plan in a factory that manufactures multiple products. This system automatically creates a new plan to change the original production plan when some abnormality occurs in the original production plan, such as equipment failure, a problem with material procurement, or a sudden increase in demand.

[0107] In an automatic production plan modification system, the history of a proposed plan is information that includes, for example, the original plan, a post-abnormal plan that arose as a result of an abnormality that occurred in the original plan, a revised plan that has been modified, and a revision proposal that indicates what modifications have been made. A production plan includes, for example, a manufacturing schedule that indicates which products multiple pieces of equipment in a factory will produce at what time, or whether they will not produce anything and will perform other work such as maintenance, procurement plan information that indicates when materials required for product production will be ordered and how much will be available at what time, and demand plan information that indicates how much of each product is needed at what time.

[0108] A revision is a method of changing a production plan, such as changing product A that equipment was originally scheduled to produce at a certain time or range of times to a different product B, or changing the time or quantity of materials to be ordered in procurement plan information.

[0109] Examples of KPIs include whether the amount of manufactured products is large, whether the production volume of products meets demand, whether the product inventory is greater than the required amount, whether the product inventory is less than the upper limit, whether the number of times equipment is changed over in the factory is low, whether the percentage of equipment operating time is high, whether the material inventory is greater than the required amount, or whether the material inventory is less than the upper limit. [Explanation of symbols]

[0110] 10...Train rescheduling plan generating device 101...Storage device 104...Processor 105...Memory 110...Predictive timetable simulator 111...Train rescheduling operation database 112...Objective function 113...Timetable evaluation unit 120...Inference unit 130...Model generation unit 131...Model learning unit 132...Learning data generation unit 133...Learning database 134...Past train rescheduling database 1200...Train rescheduling plan generating unit 1202...Train rescheduling restriction model 1203...Delay situation data 1204...Planned timetable 1205...Predicted timetable after delay 1206...Revised timetable 1207...Train rescheduling plan 1208...KPI priority 1209...Train rescheduling operation list 1334...Past database 1400...Past train rescheduling database 1401...Past train rescheduling data 1402...Past train rescheduling data 1411...Past planned timetable 1412...Predicted timetable after delay 1413...Past revised timetable 1414...Past train rescheduling plan

Claims

1. a learning data generation unit that generates learning data by assigning corresponding KPI priority information to each of the histories of plan reorganization proposals, which are plan modification method proposals; a plan reorganization proposal determination unit that determines the plan reorganization proposal that maximizes a predetermined objective function based on the plan that needs to be modified and any of the KPI priority information, using a plan reorganization proposal restriction model that has been machine-learned using the learning data as training data; A planning and organizing device comprising:

2. 2. The planning system according to claim 1, a plan generation unit that generates a revised plan based on the plan reorganization proposal; a plan evaluation unit that calculates a KPI achievement level related to the plan; Equipped with the learning data generation unit generates a post-perturbation plan reorganization by applying perturbation at one or more locations to the history of the plan reorganization; the plan generation unit creates the revised plan from each of the plan reorganization proposal and the perturbed plan reorganization proposal; The plan evaluation unit calculates the KPI achievement degree for each of the revised plan based on the history of the plan reorganization proposal and the revised plan based on the perturbed plan reorganization proposal; The learning data generation unit estimates the KPI priority corresponding to the history of the plan reorganization proposal by comparing each of the KPI achievement degrees calculated by the plan evaluation unit, and assigns the KPI priority to the history of the plan reorganization proposal. A planning and organizing device characterized by:

3. 3. The project organizing and planning system according to claim 2, The learning data generation unit applies the KPI achievement degrees of the revised plan based on the history of the plan reorganization proposal and the revised plan based on the perturbed plan reorganization proposal to the objective function to generate an inequality, and estimates the KPI priority corresponding to the history of the plan reorganization proposal based on the inequality. A planning and organizing device characterized by:

4. 4. The planning system according to claim 3, The KPI achievement level has a plurality of evaluation indexes, and the objective function is a sum of values ​​obtained by multiplying each of the evaluation indexes by a variable corresponding to each of the evaluation indexes. A planning and organizing device characterized by:

5. 5. The planning and organizing device according to claim 4, The learning data generation unit applies the perturbation to the history of the plan reorganization by partially or temporally changing at least one of the items constituting the plan reorganization. A planning and organizing device characterized by:

6. 6. The project organizing and planning system according to claim 5, The learning data generation unit obtains a range that can be taken by the plurality of variables that constitute the KPI priority from the inequality, and estimates the KPI priority using a representative point within this range. A planning and organizing device characterized by:

7. 4. The planning system according to claim 3, The timetable rescheduling plan drafting device is a timetable rescheduling plan drafting device that drafts a timetable rescheduling plan, which is a proposed method for amending a railway train schedule, the learning data generation unit generates a post-perturbation timetable replanning proposal by applying perturbation at one or more locations to the history of the timetable replanning proposal; the plan generation unit generates a revised train schedule from the timetable rescheduling proposal and the perturbed timetable rescheduling proposal, The plan evaluation unit calculates the KPI achievement degree for each of a revised train schedule based on the history of the train schedule replanning proposal and a revised train schedule based on the perturbed train schedule replanning proposal; the learning data generation unit applies the KPI achievement degree to a predetermined objective function to generate the inequality, estimates a KPI priority corresponding to the history of the timetable replanning proposal based on the inequality, and assigns the KPI priority to the history of the timetable replanning proposal; A planning and organizing device characterized by:

8. 8. The project organizing and planning system according to claim 7, The timetable rescheduling plan includes a combination of a timetable rescheduling operation and information required for the timetable rescheduling operation, The learning data generation unit generates the perturbed timetable replanning plan by perturbing information on one or more timetable replanning operations with respect to the timetable replanning plan history. A planning and organizing device characterized by:

9. A planning method using a planning device, The planning and organizing device includes: a learning data generation step of generating learning data by assigning corresponding KPI priority information to each of the histories of plan reorganization proposals, which are plan modification method proposals; a plan reorganization decision step of deciding the plan reorganization that maximizes a predetermined objective function based on the plan that needs to be revised and any of the KPI priority information, using a plan reorganization restriction model that has been machine-learned using the learning data as training data; A planning method using a planning device that executes the above.

10. A computer program executed by a computer, When this computer program is executed, the computer a learning data generation function that generates learning data by assigning corresponding KPI priority information to each of the histories of plan reorganization proposals, which are plan revision method proposals; a plan reorganization decision function that decides the plan reorganization that maximizes a predetermined objective function based on the plan that needs to be revised and any of the KPI priority information, using a plan reorganization restriction model that has been machine-learned using the learning data as training data; A computer program that executes the following:

Images