WO2024053020A1 - System and method for estimating factor of difference between simulation result and actual result - Google Patents

Abstract

This system receives, from a user, an input of domain knowledge relating to a reproduction object by a simulator, and stores, in a storage device, influencing factor candidate data which is based on the domain knowledge and which represents the parent‐child relationship between factors. On the basis of data representing simulation results and the influencing factor candidate data, the system generates a first causal sequence model for entities relating to simulation differences (differences between the simulation results and actual results), regarding the entities and candidate factors influencing the entities. Further, on the basis of actual result data representing the actual results and the influencing factor candidate data, the system generates a second causal sequence model for the entities, regarding the entities and candidate factors influencing the entities. The system extracts one or more differences between the first causal sequence model and the second causal sequence model.

Description

System and method for estimating factors behind differences between simulation results and actual results

The present invention generally relates to a technique for estimating the cause of the difference between simulation results and actual results.

In order to deal with the large number of shipments and manufacturing instructions that occur every day in distribution warehouses and factories, work plans are created to control worker placement, work order, etc. before the start of work.

In recent years, especially in the manufacturing industry, there has been an increase in small-lot, high-mix products, and in distribution warehouses, with the expansion of e-commerce distribution, there has been an increase in small-lot shipments. Therefore, it is necessary to respond to the conditions of each client, making it difficult to plan using a simple and uniform method, and requiring the creation of a detailed work plan.

It is difficult to plan for such complex shipping and manufacturing instructions using the manual discretion of managers, so operations have begun to use computer calculations such as mathematical optimization to find optimal plans.

In the process of finding the optimal solution, a recursive process is performed in which the solution candidates generated in the search process are evaluated, and based on the results, the solution is adjusted to find a better solution and evaluated again. As an evaluation in this process, a simulator that simulates the behavior of actual factories and warehouses is constructed, and information about solution candidates is given to this simulator to perform simulations. Solution candidates are evaluated based on the results of this simulation. For example, when searching for a plan that provides the fastest work, the work completion time is evaluated through simulation.

From the above, a planning system consisting of an optimization function and a simulator that works with it is constructed as appropriate.

Patent Document 1 discloses a method for optimizing simulation conditions in manufacturing.

JP 2019-101683 Publication

Regarding work bases such as warehouses and factories, there are entities such as processing machines and workers for each process. Regarding these entities, there are parameters such as processing time. A simulator is constructed using such entities, parameters, etc. as simulation factors.

However, the simulator deviates from the reproduction target as time passes. Specifically, as time passes, factors that were not considered when constructing the simulator may become necessary, or factors that were considered when constructing the simulator may change or become unnecessary.

More specifically, for example, when building the simulator, assume that the type of item was not taken into account as a factor because there was no difference in work speed depending on the type of item handled in the warehouse. Suppose that over time, the number of types of items increases, and there are noticeable differences in work speed depending on the type of item. In this case, the type of item is one of the factors that causes the difference between the simulation result and the actual result. However, in this case, simply by readjusting the existing parameters, it is not possible to estimate the cause of the difference between the simulation result and the actual result, and the simulator cannot be made to follow the reproduction target. Additionally, there is a possibility that unrelated parameters may be erroneously adjusted.

Neither this problem nor a method for solving such a problem is disclosed or suggested in Patent Document 1.

There are applications based on simulation results (for example, optimization of work plans), but if the simulator deviates from what is being reproduced (the accuracy of the simulation decreases), the application will be adversely affected (for example, optimization results (reducing the quality or performance of the results).

The difference factor estimation system receives input of domain knowledge regarding the reproduction target of the simulator from the user, and stores influencing factor candidate data, which is data based on the domain knowledge and represents parent-child relationships of factors, in the storage device. Based on the simulation result data representing the results of simulation by the simulator and the influence factor candidate data, the difference factor estimation system estimates the entity and the relevant A first causal relationship model, which is a model of causal relationships with candidate factors that influence the entity, is generated. Further, the difference factor estimation system generates a second model, which is a model of the causal relationship between the entity and the candidate factor that influences the entity, for the entity based on the performance data representing the performance and the influencing factor candidate data. Generate a causal model. A difference factor estimation system extracts one or more differences between the first causal relationship model and the second causal relationship model.

According to the present invention, it is possible to contribute to maintaining the accuracy of simulation even over time.

1 is a diagram illustrating a logical configuration example of a difference factor estimation system according to a first embodiment; FIG. It is a figure which shows the example of a structure of a simulation result table. It is a figure which shows the example of a structure of an operation performance table. FIG. 3 is a diagram illustrating a configuration example of an entity table. It is a figure showing an example of composition of an influencing factor candidate table. It is a figure which shows the example of a structure of a 1st modeling result table. It is a figure which shows the example of a structure of a 2nd modeling result table. It is a figure showing the flow of the whole processing performed by a difference factor estimation system. 8 is a diagram showing the flow of difference extraction processing (S701 in FIG. 7). FIG. 8 is a diagram showing the flow of influence factor modeling processing (S702 in FIG. 7). FIG. FIG. 8 is a diagram showing the flow of influencing factor selection processing (S703 in FIG. 7). 8 is a diagram showing the flow of correction method selection processing (S704 in FIG. 7). FIG. 8 is a diagram showing an example of a first visualization screen displayed in the visualization process (S705 in FIG. 7). FIG. 8 is a diagram showing an example of a second visualization screen displayed in the visualization process (S705 in FIG. 7). FIG. FIG. 8 is a diagram showing an example of a third visualization screen displayed in the visualization process (S705 in FIG. 7). It is a figure showing an example of a logical composition of a difference factor estimation system concerning a 2nd embodiment. It is a figure showing an example of composition of a cooperation system parameter table. It is a figure which shows an example of the visualization screen which recommends changing the number of searches. FIG. 6 is a diagram illustrating an example of a visualization screen that recommends changing settings of an approximate model. FIG. 2 is a diagram showing an example of a physical configuration of a difference factor estimation system.

In the following description, an "interface device" may be one or more interface devices. The one or more interface devices may be at least one of the following:
- One or more I/O (Input/Output) interface devices. The I/O (Input/Output) interface device is an interface device for at least one of an I/O device and a remote display computer. The I/O interface device for the display computer may be a communication interface device. The at least one I/O device may be a user interface device, eg, an input device such as a keyboard and pointing device, or an output device such as a display device.
- One or more communication interface devices. The one or more communication interface devices may be one or more of the same type of communication interface device (for example, one or more NICs (Network Interface Cards)) or two or more different types of communication interface devices (for example, one or more NICs (Network Interface Cards)). It may also be an HBA (Host Bus Adapter).

Also, in the following description, "memory" refers to one or more memory devices, typically a main storage device. At least one memory device in the memory may be a volatile memory device or a non-volatile memory device.

Also, in the following description, "persistent storage" refers to one or more persistent storage devices. The persistent storage device is typically a non-volatile storage device (for example, an auxiliary storage device), and specifically, for example, an HDD (Hard Disk Drive) or an SSD (Solid State Drive).

Furthermore, in the following description, a "storage device" may be at least a memory and a persistent storage device.

Also, in the following description, a "processor" refers to one or more processor devices. The at least one processor device is typically a microprocessor device such as a CPU (Central Processing Unit), but may be another type of processor device such as a GPU (Graphics Processing Unit). At least one processor device may be single-core or multi-core. The at least one processor device may be a processor core. At least one processor device may be a broadly defined processor device such as a hardware circuit (for example, a field-programmable gate array (FPGA) or an application-specific integrated circuit (ASIC)) that performs part or all of the processing.

In addition, in the following explanation, functions may be explained using the expression "yyy part", but functions may be realized by one or more computer programs being executed by a processor, or one or more computer programs may be executed by a processor. It may be realized by the above hardware circuit (for example, FPGA or ASIC), or a combination thereof. When a function is realized by a program being executed by a processor, the specified processing is performed using a storage device and/or an interface device as appropriate, so the function may be implemented as at least a part of the processor. good. A process described using a function as a subject may be a process performed by a processor or a device having the processor. Programs may be installed from program source. The program source may be, for example, a program distribution computer or a computer-readable recording medium (for example, a non-temporary recording medium). The description of each function is an example, and a plurality of functions may be combined into one function, or one function may be divided into a plurality of functions.

In addition, in the following explanation, data that provides an output in response to input may be explained using expressions such as "xxx table," but the data may have any structure (for example, It may be structured data or unstructured data), or it may be a learning model such as a neural network, genetic algorithm, or random forest that generates an output based on input. Therefore, the "xxx table" can be called "xxx data." In addition, in the following explanation, the configuration of each table is an example, and one table may be divided into two or more tables, or all or part of two or more tables may be one table. It's okay.

Hereinafter, some embodiments of the present invention will be described with reference to the drawings.
[First embodiment]

It is assumed that work plans are optimized (including creation of work plans) on a daily basis at work bases such as warehouses and factories. For such purposes, the results of simulations performed by work simulators are used. If there is a difference between the simulation results and the actual operating results, the cause of the difference is estimated.

The work simulation may be, for example, any of the following simulations (A) and (B).
(A) Simulation of the time-series transition of the state of one or more entities in which a series of processes such as warehouse arrival, storage, picking, distribution processing, and truck loading interact with each other.
(B) Simulation of the time-series transition of the state of one or more entities that behave in cooperation with multiple processes such as the manufacturing process of a factory.

In both (A) and (B) above, the one or more entities may be, for example, at least one of a person, a machine, and an item. Furthermore, both of the above simulations (A) and (B) may be processes executed by a computer.

Hereinafter, as a work simulation, a work simulation related to work in a warehouse (an example of a work base) will be taken as an example.

FIG. 19 is a diagram showing an example of the physical configuration of the difference factor estimation system 100 in the first embodiment.

The difference factor estimation system 100 is a physical computer system (one or more physical computers), and includes an interface device (I/F) 51, a storage device 52, and a processor 53 connected thereto. The difference factor estimation system 100 may be a logical computer system based on such a physical computer system (for example, a cloud computing system based on a cloud infrastructure).

An input device 72 and a display device 71 are connected to the interface device 51. Data is input by the user from the input device 72 via the interface device 51 . A user interface screen such as a GUI (Graphical User Interface) is displayed on the display device 71 via the interface device 51 . A touch panel may be employed as the display device 71 and the input device 72. The difference factor estimation system 100 may be a computer having the display device 71 and the input device 72, or the difference factor estimation system 100 may be a server, and the display device 71 and the input device 72 may be the same as the difference factor estimation system 100. It may be provided in the communicating client.

A data source 60 is connected to the interface device 51. Data source 60 may include one or more sensors located in the warehouse. Measurement data (data including values measured by sensors) may be input from the data source 60 via the interface device 51 .

The storage device 52 stores input/output data. Further, a computer program executed by the processor 53 is stored. The processor 53 reads the program from the storage device 52 and executes it.

FIG. 1 is a diagram showing an example of a logical configuration of a difference factor estimation system 100.

The storage device 52 of the difference factor estimation system 100 stores a simulation result table 108, an operation record table 110, an entity table 112, an influencing factor candidate table 114, a first modeling result table 116, and a second modeling result table 118. be done. Further, by the processor 53 of the difference factor estimation system 100 executing the computer program, the difference extraction unit 101, the first modeling unit 102, the second modeling unit 103, the influencing factor selection unit 104, the updating unit 105, and the visualization unit Functions such as 106 are realized. Although not shown, a work simulator may be realized by the processor 53 executing a computer program. The work simulator may exist in a computer system separate from the difference factor estimation system 100.

FIG. 2 is a diagram showing an example of the configuration of the simulation result table 108.

The simulation result table 108 represents the results of simulation by the work simulator. The simulation result table 108 indicates which process started and when it ended for each work unit. In the case of warehouse work, boxes are prepared and packed for each shipping destination in accordance with orders from the shipping destination, so boxes are used as an example of the unit of work. In this embodiment, the simulation result table has an entry for each workbox, and each entry represents the ID of the workbox and the start time and end time of each process. In addition, each entry may include data representing data regarding entities related to the workbox (for example, entity ID, entity type, or entity details (for example, number of items, number of workers)). Examples of entities are workers, material handling equipment, or items.

FIG. 3 is a diagram showing an example of the configuration of the operation record table 110.

The configuration of the operation performance table 110 may be the same as the configuration of the simulation result table 108. The operation performance table 110 may be constructed based on data input from the data source 60 or an input device, for example. The operation performance table 110 may be prepared for each period.

FIG. 4 shows a configuration example of the entity table 112.

The entity table 112 has an entry for each simulation difference (difference between simulation results and operation results), and each entry includes the time period in which the simulation difference occurred, the process in which the simulation difference occurred, the entity related to the simulation difference, and This is data representing the amount of difference. For example, in a warehouse, workers and material handling equipment perform work, but if there is a simulation difference in the number of work boxes at a certain picking station, the ID of the picking process is recorded as the process ID, and the picking process is recorded as the entity ID. The ID of the station is recorded, the difference in the number of work boxes is recorded as the difference amount, and the start date and time of the time zone in which the difference occurred is recorded as the time zone. The length of the time period in which the difference occurs may or may not be uniform (for example, one hour).

For each entity, data representing details of the entity may be specified from an entity master table (not shown) using the ID of the entity as a key.

FIG. 5 shows a configuration example of the influencing factor candidate table 114.

The influencing factor candidate table 114 represents influencing factor candidates. An "influencing factor candidate" is a candidate for an influencing factor. "Influencing factors" are elements defined based on the knowledge possessed by people such as warehouse managers (domain knowledge regarding the objects to be reproduced in work simulators), and are typically defined based on the state of the entity (entity (Example of output value for)

The influencing factor candidate table 114 has an entry for each predetermined management unit such as a process, and each entry is data representing a parent-child relationship between elements. According to parent-child relationships between elements, parent elements influence child elements. For example, when there are many items, it is necessary to access the shelves where each item is stored, which increases the working time. Therefore, the number of items to be picked can be considered as a factor that affects the working time in the picking process, and in this case, the working time is a child element and the number of items is a parent element. In an entry, the influencing factor parameter may be the ID (eg, name) of the parent element, and the influenced factor parameter may be the ID of the child element.

Since parent-child relationships between elements may be affected by the warehouse configuration, the influencing factor candidate table 114 may be input by the warehouse manager using the input device 72, but between warehouses with different configurations However, since there may be a common parent-child relationship, a common influencing factor candidate table 114 may be prepared for a plurality of warehouses.

Also, at least one parent-child relationship may be a causal relationship. For example, if it is clearly predicted in advance that the work time will be long in proportion to the number of items, the parent-child relationship may be a relational expression that takes the number of items and the work time as elements.

Also, one or more parent elements exist for one child element.

FIG. 6A is a diagram showing a configuration example of the first modeling result table 116. FIG. 6B is a diagram showing a configuration example of the second modeling result table 118.

The first and second modeling result tables 116 and 118 represent models of causal relationships, respectively. For example, the model may be a conditional probability calculation formula in which output values of various parameters are determined probabilistically under specific conditions. In this case, the influencing factor (parent element) and the influenced factor (child element) are parameter items, and the value of each element may be a parameter value. The model may be a model obtained using machine learning or a statistical processing library instead of such a calculation formula, and in that case, both tables 116 and 118 may be stored as binary files.

Both the first and second modeling result tables 116 and 118 have an entry for each combination. A combination represents a plurality of sets and probabilities for the plurality of sets. One set is a set of a factor and a value of the factor. For example, the values of influence factor 1 and influence factor 1 are one set, and the values of influence factor 2 and influence factor 2 are another set.

Hereinafter, an example of the processing performed in this embodiment will be described.

FIG. 7 is a diagram showing the overall flow of processing performed by the difference factor estimation system 100.

In S701, the difference extraction unit 101 performs difference extraction processing. In S702, the first modeling unit 102 creates a first modeling result table 116, and the second modeling unit 103 creates a second modeling result table 118. In S703, the influencing factor selection unit 104 performs influencing factor selection processing. In S704, the update unit 105 performs correction method selection processing. In S705, the visualization unit 106 performs visualization processing.

FIG. 8 is a diagram showing the flow of the difference extraction process (S701 in FIG. 7).

In S801, the difference extraction unit 101 refers to the simulation result table 108 and the operation performance table 110. In S801, a simulation may be performed by a work simulator, and a simulation result table 108 representing the results of the simulation may be generated or updated. Further, in S801, the operation record table 110 may be generated or updated based on data input from the data source 60 or the input device 72.

In S802, the difference extraction unit 101 selects one unevaluated process (a process that has not yet been targeted for S802) based on the simulation result table 108 and the operation record table 110, and selects the selected process (FIG. 8 In the explanation of "Target process"), a difference evaluation is performed for each indicator item related to work progress. An example of the "indicator item" here may be the number of work boxes. Specifically, for example, the following (S802-1) to (S802-3) may be performed.
(S802-1) The difference extraction unit 101 counts the number of work boxes per unit time for the target process based on the simulation result table 108, and generates a simulation process time series that is a time series of the cumulative value of the number of work boxes. calculate.
(S802-2) The difference extraction unit 101 counts the number of work boxes per unit time for the target process based on the operation performance table 110, and calculates the actual process time series that is the time series of the cumulative value of the number of work boxes. calculate.
(S802-3) The difference extraction unit 101 compares the simulation process time series and the actual process time series for the target process. Specifically, for example, the difference between the cumulative value in the simulation process time series and the cumulative value in the actual process time series is calculated for each time period. If there is a time period in which the difference between the simulation process time series and the actual process time series is equal to or greater than a certain value, the difference extraction unit 101 identifies the time period in which the difference exists as the difference occurrence time period.

In addition, when there are multiple sequential processes, the progress difference between the processes other than the first process (i.e., the middle process and the last process) is due to the delay or front-loading of the previous process of the process. It may be affected. Therefore, in such a case, the difference extraction unit 101 executes the simulation again under the same conditions as the operation of the previous process of the target process, and based on the operation results and the results of the second simulation, the difference extraction unit 101 performs the above (S802-1). (S802-3) may be performed. As a result, it is possible to expect a difference evaluation for the target process excluding the influence of the previous process (for example, assuming that the start time of the target process is the same in both the simulation and the actual operation).

For example, the data on which the simulation results table 108 is based and the data on which the operation results table 110 is based include, for each entity involved in the target process, the ID of the work box processed by the entity and the data processed by the entity. The data may include data representing the work time period (for example, work start time and work end time) for each work box. The difference extraction unit 101 may count the number of work boxes for each entity in the target process from the data on which the simulation result table 108 is based, and calculate the simulation entity time series of the cumulative value of the number of work boxes. Further, the difference extraction unit 101 may count the number of work boxes for each entity in the target process from the data on which the operation performance table 110 is based, and calculate the actual entity time series of the cumulative value of the number of work boxes. . The difference extraction unit 101 may compare the simulation entity time series and the actual entity time series for each entity in the target process. Specifically, for example, the difference between the cumulative value in the simulation entity time series and the cumulative value in the actual entity time series may be calculated for each time period. The difference may be registered in the entity table 112 as a difference amount. Furthermore, data representing the ID of the target process, the ID of the entity, and the time period may be registered in the entity table 112 as data corresponding to the difference amount.

In S803, the difference extraction unit 101 determines whether a difference occurrence time period has been identified for the target process based on the processing result of S802 (S803).

If the determination result in S803 is true (S803: Yes), in S804, the difference extraction unit 101 identifies the target entity from the target process and one or more difference occurrence time periods identified between the target process. If there is only one entity per process, the target entity is identified as is. If there is a process that includes multiple entities, for example, the difference extraction unit 101 may identify, from the entity table 112, entities whose difference amount is equal to or greater than a certain value for the target process and the reoccurrence time period.

In S805, the difference extraction unit 101 determines whether all processes have been evaluated. If all processes have not been evaluated (S805: No), in other words, if there is an unevaluated process, the process returns to S802. If all steps have been evaluated, the process ends.

FIG. 9 is a diagram showing the flow of the influence factor modeling process (S702 in FIG. 7).

In S901, each of the first modeling unit 102 or the second modeling unit 103 selects one unselected entity (an entity not selected in S901) from among the entities extracted in the difference extraction process (S701 in FIG. 7). ). The entity selected here will be referred to as a "target entity" in the explanation of FIG.

In S902, each of the first modeling unit 102 or the second modeling unit 103 identifies influence factor candidates related to the target entity from the influence factor candidate table 114. For example, for each process in which a difference occurrence time zone is specified for the target entity, an influencing factor candidate (parent element) of which the target entity is an affected factor (child element) is identified.

In S903, the second modeling unit 103 performs modeling using the influencing factor candidates identified in S902 and the operation performance table 110. In S903, the following may be performed.
(a) The second modeling unit 103 determines, based on the relationship between the target entity (child element) and the influencing factor candidate (parent element) and the operation record table 110, that the element is a node and the relationship between the elements is an edge. Create a graph that is . The element corresponding to the node in this graph is an element specified from the operation record table 110 (for example, an entity specified using the workbox ID in the operation record table 110 as a key), and the target entity (child element) It may be an influencing factor candidate (parent element) related to .
(b) The second modeling unit 103 generates a causal relationship model (for example, a Bayesian network model) based on the graph and the operation record table 110. For this model, the second modeling unit 103 calculates a score such as BIC (Bayesian Information Criterion). This score may correspond to the evaluation of the influencing factor candidate.

The second modeling unit 103 may perform (a) and (b) for each influencing factor candidate. Note that other methods may be adopted as the evaluation method for influencing factor candidates. For example, a single graph may be generated and evaluated for each influencing factor candidate, or a graph may be generated using all influencing factor candidates and the amount of change in score when each influencing factor candidate is removed one by one. The evaluation may be based on.

Additionally, a causal relationship model may be generated for each period of operation performance. For example, even if the causality model is generated based on the operation performance table 110 of the last time when the work simulator to be evaluated is updated, and the causality model is generated based on the operation performance table 110 of the most recent period. good.

In S904, the first modeling unit 102 performs modeling using the influencing factor candidates identified in S902 and the simulation result table 108. In the details of the explanation of S904, "second modeling unit 103" in the details of S903 is replaced with "first modeling unit 102", and "operation performance table 110" in the details of the explanation is replaced with "simulation result table 108". ”. If the influencing factors considered by the target entity can be obtained from data representing the specifications of the simulation, the causal relationship may be determined based on that data.

In S905, the first modeling unit 102 or the second modeling unit 103 determines whether all sets have been evaluated (S902 to S904 are performed for all entities extracted in the difference extraction process (S701 in FIG. 7)). (Whether or not) is determined. If all sets have not been evaluated (S905: No), the process returns to S901. If all sets have been evaluated (S905: Yes), the process ends.

The second modeling result table 118 may be updated by the second modeling unit 103 every time S903 is performed. The first modeling result table 116 may be updated by the first modeling unit 102 every time S904 is performed.

FIG. 10 is a diagram showing the flow of the influencing factor selection process (S703 in FIG. 7).

In S1001, the influencing factor selection unit 104 selects one unselected entity (entity not selected in S1001) from the entities extracted in the difference extraction process (S701 in FIG. 7). The entity selected here will be referred to as a "target entity" in the explanation of FIG.

In S1002, the influencing factor selection unit 104 obtains a causal relationship model (simulation causal relationship model) for the target entity from the first modeling result table 116, and obtains a causal relationship model for the target entity from the second modeling result table 118. Obtain the model (causal relationship model of operation results). The causal relationship models acquired here may be models represented by entries having parameter items of affected factors that match the target entity.

In S1003, the influencing factor selection unit 104 compares the models obtained in S1002 (the simulation causal relationship model and the operational performance causal relationship model), and selects an influencing factor based on the comparison result. For example, if there is an influencing factor that appears in the causal relationship model of the operation performance but does not appear in the causal relationship model of the simulation, the influencing factor selection unit 104 selects the influencing factor. In addition, an influencing factor that appears in both models but has a different structure of a relational expression with the output of an entity that is an influenced factor may be selected.

In S903 and S904 of FIG. 9, when causality models are generated for each period, differences between these causality models may be included in the evaluation. For example, if there is an influencing factor that does not appear in the causal relationship model based on the operational performance table 110 at the time when the work simulator was last updated, but appears in the causal relationship model based on the most recent operational performance table 110, the influencing factor is It is presumed that this is a factor that has come to influence the behavior of the warehouse, and is therefore presumed to be a factor that has increased the progress differences in recent work simulators. Therefore, the influencing factor selection unit 104 may narrow down the influencing factors (α) below to the factors corresponding to (β) below.
(α) There is a difference between the causal relationship model of the work simulator (the model identified from the first modeling result table 116) and the causal relationship model of the operation performance (the model identified from the second modeling result table 118). Certain influencing factors.
(β) A factor that appears as a difference between the causal relationship model of the operation results at the time of the last update of the work simulator and the causal relationship model of the most recent operation results.

In S1004, the influencing factor selection unit 104 determines whether all sets have been evaluated (whether S1002 and S1003 have been performed for all entities extracted in the difference extraction process (S701 in FIG. 7)). If all sets have not been evaluated (S1004: No), the process returns to S1001. If all sets have been evaluated (S1004: Yes), the process ends.

FIG. 11 is a diagram showing the flow of the correction method selection process (S704 in FIG. 7).

In S1101, the updating unit 105 specifies all the influencing factors selected in the influencing factor selection process (S703 in FIG. 7).

In S1102, the updating unit 105 generates a correction method to be verified through simulation for each of the identified influencing factors. A plurality of types of correction methods may be generated, such as a correction method that takes into account influencing factors regarding all entities extracted in the difference extraction process (S701 in FIG. 7), and a correction method that partially incorporates influencing factors.

The influencing factor corresponds to the difference between the causal relationship model of the simulation result and the operational performance, and therefore, the influencing factor can be a factor of the difference between the simulation result and the operational performance. Therefore, if the difference as an influencing factor is eliminated, it is expected that the difference between the simulation result and the operating performance will be reduced (preferably eliminated). Therefore, the correction method generated for each influencing factor is a model for reducing the difference between the simulation result and the operating performance regarding the influencing factor. For example, the update unit 105 traces the simulation (for example, traces the simulation log (not shown)), and for each influencing factor, based on the tracing result, updates the operation performance of the simulation result with respect to the influencing factor. Generate a correction scheme to reduce the difference in . The updating unit 105 generates a correction method by, for example, adjusting (for example, changing and/or adding) parameters of the simulation conditions of the selected entity regarding the influencing factor. For example, when the influencing factor candidate in the work progress simulation is the "number of items" (work time varies depending on the number of items), the updating unit 105 changes the definition formula of the work speed for the number of items in the simulation conditional formula to: A causal relationship model (e.g., especially a simulation causal relationship model (conditional probability model, etc.)) and/or an approximate model (calculated from the performance of factors extracted in the causal relationship model (i.e., work time and number of items)) regression model, etc.) or update it. Note that the "approximate model" will be described with respect to the second embodiment.

In S1103, the updating unit 105 selects one unselected correction method (the correction method not yet selected in S1103) from among the correction methods created in S1102.

In S1104, the updating unit 105 causes the simulator to execute a simulation that reflects the correction method (for example, applies the correction method to a work simulator and causes the work simulator to execute the simulation). The updating unit 105 compares the results of the simulation with the actual operating results, and evaluates the improvement effect by determining how much the difference (for example, the difference in working hours) has been reduced. Moreover, instead of or in addition to the improvement effect, the updating unit 105 evaluates the degree of similarity between the work simulator to which the correction method is applied and the work simulator before the correction method is applied. The degree of similarity may be based on, for example, the degree of overlap between the input data to the work simulator to which the correction method is applied and the input data to the work simulator before the correction method is applied. If the degree of similarity is high, it can be expected that the number of man-hours required to repair the work simulator will be small. By evaluating both the improvement effect and the degree of similarity, it is possible to determine a correction method that can be expected to reduce the number of man-hours required for modifying the work simulator and reduce differences. In this embodiment, the evaluation items include improvement effects and repair man-hours.

In S1105, the update unit 105 determines whether evaluation of all sets (all correction methods) is completed. If all sets have not been evaluated (S1105: No), the process returns to S1103. If all sets have been evaluated (S1105: Yes), the process ends.

The visualization unit 106 performs visualization processing (S705 in FIG. 7) based on the results of the previous processing (S701 to S704 in FIG. 7). In the visualization process, a screen such as a GUI (Graphical User Interface) (hereinafter referred to as a "visualization screen" for convenience) is displayed on the display device 71. Some examples of screens will be described with reference to FIGS. 12 to 14.

FIG. 12 shows the first visualization screen.

The first visualization screen 1200 is a screen based on data obtained by the difference extraction process (S701 in FIG. 7 and FIG. 8). The first visualization screen 1200 displays a graph representing a time series of differences in the number of work boxes for each process. The graph is based on the result of S802 in FIG. 8, for example. That is, for each process, a time series of the difference (difference in the number of work boxes) between the simulation process time series and the actual process time series is displayed as a graph.

According to the first visualization screen 1200, for process 3, there is a time period in which a difference occurs, that is, a time period in which the difference between the simulation process time series and the actual process time series is equal to or greater than a certain value. The time period in which the difference occurred is highlighted. Regarding the difference occurrence time period, text representing details (for example, the maximum value of the differences (differences in the number of work boxes) belonging to the current time period) is displayed.

FIG. 13 shows the second visualization screen.

The second visualization screen 1300 is a screen based on the data obtained by the influencing factor modeling process (S702 in FIG. 7 and FIG. 9) and the data obtained by the influencing factor selection process (S703 in FIG. 7 and FIG. 10). be.

The second visualization screen 1300 displays a graph representing a causal relationship model of simulation and a graph representing a causal relationship model of operational performance. Each graph is, for example, a DAG (Directed Acyclic Graph) with factors (elements) as nodes and parent-child relationships between factors as edges. Note that the illustrated "condition parameter" is a parameter (an example of an element) that influences the influenced factor.

In addition, on the second visualization screen 1300, influencing factors as differences between the causal relationship model of the simulation and the causal relationship model of the operational performance, that is, the influencing factors selected in the influencing factor selection process are highlighted. According to the example shown in FIG. 13, influencing factors A and B are the selected influencing factors. Although the influencing factor A exists in both models, the relationship (for example, degree of dependence or parameter) between the influencing factor A and the influenced factor differs between these models. Influence factor B is an influence factor that exists in the causal relationship model of operation performance but does not exist in the causal relationship model of simulation.

Regarding the influencing factor A (difference P), the configuration between the simulation DAG and the actual operation DAG is the same. Therefore, the difference with respect to the influencing factor A is displayed from a different perspective. Specifically, a histogram (hereinafter referred to as a simulation histogram) representing the relationship between the value of the influencing factor A and the value of the influenced factor is displayed based on the simulation result table 108 and the simulation causal relationship model. Furthermore, a histogram (hereinafter referred to as performance histogram) representing the relationship between the value of the influencing factor A and the value of the influenced factor is displayed based on the operation performance table 110 and the causal relationship model of the performance performance. Also, the difference between those histograms is displayed. In place of or in place of the histogram, a mathematical formula or other types of information representing the relationship between the influencing factor A and the influenced factor may be displayed.

FIG. 14 shows the third visualization screen.

The third visualization screen 1400 is a screen based on data obtained through the correction method selection process (S704 in FIG. 7 and FIG. 11) in addition to the data on which the second visualization screen 1300 is based.

On the third visualization screen 1400, recommendations are displayed for each correction method for which a simulation evaluation higher than a predetermined evaluation has been obtained. According to the example shown in FIG. 14, recommendation 1 is a recommendation of a correction method corresponding to the influence factor A (difference P) shown in FIG. Recommendation 2 corresponds to the correction method corresponding to influence factor B (difference Q) shown in FIG. Each of Recommendations 1 and 2 is an example of a recommendation corresponding to a correction method that obtained a simulation evaluation higher than a predetermined evaluation. "Simulation evaluation of a predetermined evaluation or higher" means a simulation evaluation in which the evaluation value for each of one or more evaluation items is a predetermined value or higher. The evaluation items are, for example, each of the above-mentioned improvement effect and repair man-hour, and evaluation values are calculated for each of the improvement effect and repair man-hour in S1104 of FIG. 11. The evaluation value may be a numerical value (for example, the number of work boxes) or a level value (for example, three levels of large, medium, and small).

According to Recommendation 1, the improvement effect of updating the relationship between the influencing factor A and the influenced factor in the simulation causality model is high (the number of work boxes as a difference becomes smaller), and the influencing factor A is Since this factor exists in both the simulation causality model and the operational performance causality model, there is no need to add or delete the influencing factor itself, and therefore the number of man-hours required for modification is small. Specifically, for example, based on the average difference between the actual work completion time and the actual work completion time when simulations are performed for Y days (for example, several days) before and after applying Recommendation 1 (correction method), improvement effects are identified and displayed.

According to Recommendation 2, the improvement effect of adding influencing factor B to the simulation causal relationship model is higher than Recommendation 1, but influencing factor B needs to be added to the simulation causal relationship model, and therefore, the number of repair man-hours is reduced. is large.

The visualization unit 106 displays (A) the simulation results of the work simulator before modification, (B) the simulation results when recommendation 1 is applied to the work simulator, and (C) the simulation results of recommendation 2 on the third visualization screen 1400. The simulation results when applied to a work simulator and (D) the results of actual operation are shown. Each of (A) to (D) may be, for example, a time series of the remaining work amount (remaining number of work boxes to be worked on). (A) to (D) are displayed in the same display area (for example, in the same orthogonal coordinate system). From the displays (A) to (D), the user can see the difference in the improvement effect between Recommendation 1 and Recommendation 2. Note that in S1104 of FIG. 11, the correction method is evaluated for each correction method, and each of Recommendation 1 and Recommendation 2 in FIG. For example, it corresponds to one of the top two correction methods). The graphs of Recommendation 1 and Recommendation 2 each represent simulation results when evaluating the correction method.

According to the third visualization screen 1400, the user can easily decide which recommendation to adopt from the viewpoint of a plurality of evaluation items (for example, improvement effect and repair man-hours). For example, the user can easily find a recommendation (correction method) that can be expected to have an improvement effect (reduction of differences) with less man-hours for repairing the work simulator.
[Second embodiment]

A second embodiment will be described. At that time, differences with the first embodiment will be mainly explained, and explanations of common points with the first embodiment will be omitted or simplified.

In formulating daily work plans for warehouses and factories, it is desirable to be able to automatically create work plans that combine optimization search and warehouse and factory work simulations. In this case, the work planning stage is likely to take place during the limited time before the start of work. For this reason, an approximation model (machine learning model) that simulates the behavior of the work simulator is prepared in advance, and when planning work, optimization search and predictive processing (processing using an approximation model that responds faster than the work simulator) are combined. It is desirable to automatically create a work plan.

It is thought that optimization search and prediction processing using the results of the work simulator are each performed based on set conditions. Therefore, it may be necessary to modify the setting conditions for optimization search and prediction processing as the work simulator is updated.

FIG. 15 is a diagram illustrating a logical configuration example of a difference factor estimation system according to the second embodiment.

A cooperative system that is a system that cooperates with the difference factor estimation system 100 includes an optimization search unit 1508 and an approximate model generation unit 1509. The optimization search unit 1508 and the approximate model generation unit 1509 operate using the parameters represented by the cooperative system parameter table 1510. The difference factor estimation system 100 is equipped with a parameter adjustment unit 1506, and the parameter adjustment unit 1506 adjusts parameters (updates the cooperative system parameter table 1510). The optimization search unit 1508, the approximate model generation unit 1509, and the cooperative system parameter table 1510 may be provided in a computer system separate from the difference factor estimation system 100, or may be provided within the difference factor estimation system 100. That is, at least the parameter adjustment section 1506 among the parameter adjustment section 1506, the optimization search section 1508, and the approximate model generation section 1509 is realized by the processor 53 executing a computer program.

FIG. 16 is a diagram showing a configuration example of the cooperative system parameter table 1510.

The cooperative system parameter table 1510 has an entry for each cooperative system. For each cooperative system, the entry is data representing the ID of the cooperative system and one or more parameters of the cooperative system. One parameter is a pair of a parameter name and a parameter value.

The parameter adjustment unit 1506 performs parameter adjustment processing including updating parameters for each cooperative system. Taking one cooperative system and one correction method (correction method selected in the correction method selection process) as an example, the following is an example. The parameter adjustment unit 1506 acquires one or more parameters corresponding to the cooperative system from the cooperative system parameter table 1510. The parameter adjustment unit 1506 identifies a parameter to be corrected among the one or more parameters based on the correction method, and determines a recommendation for adjusting the parameter (or generates a corrected parameter for the parameter). ).

For example, assume that the cooperative system is an optimization search and the correction method is a model that adds new influencing factors. In this case, since variations due to these influencing factors are included, the conditions that determine the behavior of the simulator increase. Therefore, the search space may be expanding. Therefore, the parameter adjustment unit 1506 determines the recommendation for adjusting the parameter value for the parameter name “search count” among the parameters corresponding to the optimization search, and the visualization unit 106 makes the recommendation as shown in FIG. A visualization screen 1700 is displayed.

Further, for example, assume that the cooperative system generates an approximate model (generates an approximate model that simulates a simulator). In this case, by inputting input data that is the same as or similar to the input data of the simulator to the approximate model (machine learning model), data corresponding to the output of the simulator is output from the approximate model. Approximate models are faster from data input to data output (prediction) than simulators. From this, if the number of influencing factors to be considered in the simulator is increasing, it is possible that the approximation accuracy will be higher if feature quantities corresponding to these factors are provided. Therefore, the parameter adjustment unit 1506 determines a recommendation for adjusting the feature amount (for example, adding a feature amount) according to the correction method, and the visualization unit 106 displays the visualization screen 1800 on which the recommendation is displayed, as shown in FIG. Display. The "feature quantity" referred to here may be an example of a parameter, that is, it may correspond to at least a parameter value. According to the example shown in FIG. 18, the parameter names "Param α" and "Param β" of the approximate model correspond to the parameter names "Parameter 1" and "Parameter 2" of the simulator, but the parameter names added for the simulator Since there is no parameter name corresponding to "parameter

Although several embodiments of the present invention have been described above, these are illustrative examples for explaining the present invention, and are not intended to limit the scope of the present invention to these embodiments. The present invention can also be implemented in various other forms.

For example, the object to be reproduced by a simulator is not limited to a work base (also called a field) such as a warehouse or a factory. For example, the flow of the entire supply chain of suppliers, factories, warehouses, stores, etc. may also fall under the "reproduction target" of the simulation. If the flow of the entire supply chain is to be "reproduced," at least one of the following may be adopted, for example.
- Entities may be suppliers, assembly plants, warehouses, trucks that travel between them, etc.
- Influencing factors may be the productivity of each factory or warehouse, or the transportation time of trucks.
- The simulation difference may be the amount of goods delivered to the store at a certain point in time (for example, the amount of goods delivered up to a certain time is decreasing due to a decline in the productivity of a factory or warehouse).

Note that the above description can be summarized, for example, as follows. The following summary may include supplementary explanations and explanations of modifications to the above explanation.

The difference factor estimation system 100 includes an interface device 51, a storage device 52, and a processor 53 connected to the interface device 51 and the storage device 52.

The processor 53 receives input of domain knowledge regarding the object to be reproduced by the simulator from the user via the interface device 51, and inputs an influence factor candidate table 114 (influence factor An example of candidate data) is stored in the storage device 52.

Based on the simulation result table 108 (an example of simulation result data) representing the results of simulation by the simulator and the influencing factor candidate table 114, the processor 53 calculates the simulation difference, which is the difference between the simulation result and the actual performance of the target to be reproduced. A first causal relationship model (simulation causal relationship model), which is a model of the causal relationship between the entity and a candidate for a factor that influences the entity, is generated for the entity related to the entity. Further, the processor 53 determines, with respect to the entity, the causal relationship between the entity and the candidate factors that influence the entity, based on the operation performance table 110 (an example of performance data) representing the operation performance and the influencing factor candidate table 114. A second causal relationship model (actual causal relationship model) is generated. Processor 53 extracts one or more model differences that are one or more differences between the first causal relationship model and the second causal relationship model. For continuous operation using simulation (for example, for continuous operation of a digital twin), in order to follow changes in the reproduction target, find the difference between the simulation results and the cause of the difference. need to be identified and improved. Since the first and second causal relationship models express the relationship between the influencing factor candidate and the influenced factor candidate, the identified model difference may correspond to a factor of the simulation difference. Therefore, it is possible to appropriately improve the simulator to follow changes in the reproduction target, that is, it can contribute to maintaining the accuracy of the simulation over time.

For each of one or more model differences, the processor 53 performs a correction method to reduce or eliminate the model difference (for example, adjustment of parameters of simulation conditions between an influencing factor and an affected factor, or adjusting a parameter of an influencing factor for simulation). addition or deletion) and evaluate the correction method. As a result, it is possible to avoid improving the simulator by taking unnecessary detailed factors into consideration, and it is therefore possible to expect efficient improvement of the simulator.

For each of the one or more model differences, the evaluation of the generated correction scheme may include at least one of the following: This can contribute to selecting a correction method based on the trade-off between improvement effect and repair load.
- The amount of reduction in simulation differences when the correction method is applied to the simulator, and/or the difference between the simulation result and the actual result when the correction method is applied to the simulator.
- Simulator modification load obtained based on the similarity between the simulator and the simulator to which the correction method is applied.

The processor 53 may visualize the results of the evaluation of the correction scheme generated for each of the one or more model differences. This makes it easy for the user to select the correction method.

The processor 53 may visualize one or more model differences. This makes it easy for the user to identify the cause of simulation differences. That is, the processor 53 can contribute to maintaining the accuracy of the simulation over time even if it performs visualization as shown in FIG. 13 and does not perform visualization as shown in FIG. 14.

Of the one or more model differences, the processor 53 generates a second causality model based on the actual performance data of the latest creation time or update time of the simulator, and a second causal relationship model based on the actual performance data of the most recent performance, for the entity. One or more model differences corresponding to one or more differences with the causal relationship model may be extracted. As a result, factors that have recently appeared can be identified as model differences by comparing the causal relationship model determined from the simulator's latest creation or update results with the causal relationship model determined from recent results. Accurate estimation of the factors is expected.

The processor 53 may adjust the parameters used in the optimization process of formulating a plan based on the simulation result table 108 based on the correction method applied to the simulator among one or more correction methods. This is expected to lead to better optimization processing.

The processor 53 may adjust parameters used in a proxy simulator in which the input and output of the simulator are approximated by a machine learning model based on the correction method applied to the simulator among one or more correction methods. This is expected to lead to a better proxy simulator.

100...Difference factor estimation system, 101...Difference extraction unit, 102...First modeling unit, 103...Second modeling unit, 104...Influence factor selection unit, 105...Update unit, 106... Visualization unit, 108... Simulation result table, 110... Operation record table, 112... Entity table, 114... Influence factor candidate table, 116... First modeling result table, 118... Second modeling result table

Claims (10)

1. comprising an interface device, a storage device, and a processor connected to the interface device and the storage device,
   The processor includes:
   Accepting input of domain knowledge regarding the reproduction target of the simulator from the user via the interface device, and storing influencing factor candidate data, which is data based on the domain knowledge and representing a parent-child relationship of the factors, in the storage device;
   Based on the simulation result data representing the simulation result by the simulator and the influencing factor candidate data, the entity and the entity regarding the simulation difference, which is the difference between the simulation result and the actual performance in the reproduction target, are determined. Generate a first causal relationship model that is a model of causal relationship with candidate factors that influence
   A second causal relationship model, which is a model of a causal relationship between the entity and a candidate factor that influences the entity, is generated for the entity based on the performance data representing the performance and the influencing factor candidate data. ,
   extracting one or more model differences that are one or more differences between the first causal relationship model and the second causal relationship model;
   Difference factor estimation system.
2. For each of the one or more model differences, the processor:
   generating a correction method to reduce or eliminate the model difference;
   evaluate the correction method;
   The difference factor estimation system according to claim 1.
3. For each of the one or more model differences, the evaluation of the generated correction scheme includes at least one of the following:
   - the amount of reduction in the simulation difference when the correction method is applied to the simulator, and/or the difference between the simulation result and the actual performance when the correction method is applied to the simulator,
   - Simulator modification load obtained based on the similarity between the simulator and the simulator to which the correction method is applied;
   The difference factor estimation system according to claim 2.
4. the processor visualizes the results of the evaluation of the correction scheme generated for each of the one or more model differences;
   The difference factor estimation system according to claim 2.
5. the processor visualizes the one or more model differences;
   The difference factor estimation system according to claim 1.
6. The processor is configured, for the entity, to create a second cause-and-effect relationship model based on performance data of the most recent creation time or update time of the simulator, and a second causal relationship model based on performance data of the most recent performance among the one or more model differences. extracting one or more model differences corresponding to one or more differences with a second causal relationship model;
   The difference factor estimation system according to claim 1.
7. The processor selects a correction method to be applied to the simulator from among the one or more correction methods to determine parameters used in an optimization process for creating a plan based on simulation result data representing simulation results of the simulator. Adjust based on
   The difference factor estimation system according to claim 2.
8. The processor adjusts parameters used in a proxy simulator that approximates input and output of the simulator with a machine learning model based on a correction method applied to the simulator among the one or more correction methods.
   The difference factor estimation system according to claim 2.
9. A computer receives input of domain knowledge regarding a reproduction target of the simulator from a user, and stores influencing factor candidate data, which is data based on the domain knowledge and representing a parent-child relationship of the factors, in a storage device;
   A computer determines, based on simulation result data representing a simulation result by the simulator and the influencing factor candidate data, an entity related to a simulation difference that is a difference between the simulation result and the performance in the reproduction target. Generate a first causal relationship model that is a model of the causal relationship between and a candidate for a factor that affects the entity,
   A computer generates, for the entity, a second causal relationship model that is a model of a causal relationship between the entity and a candidate for a factor that influences the entity, based on the performance data representing the performance and the influencing factor candidate data. generate,
   the computer extracts one or more model differences that are one or more differences between the first causal relationship model and the second causal relationship model;
   Difference factor estimation method.
10. Accepting input of domain knowledge regarding the reproduction target of the simulator from the user, storing influencing factor candidate data, which is data based on the domain knowledge and representing parent-child relationships of the factors, in a storage device;
    Based on the simulation result data representing the simulation result by the simulator and the influencing factor candidate data, the entity and the entity regarding the simulation difference, which is the difference between the simulation result and the actual performance in the reproduction target, are determined. Generate a first causal relationship model that is a model of causal relationship with candidate factors that influence
    A second causal relationship model, which is a model of a causal relationship between the entity and a candidate factor that influences the entity, is generated for the entity based on the performance data representing the performance and the influencing factor candidate data. ,
    extracting one or more model differences that are one or more differences between the first causal relationship model and the second causal relationship model;
    A computer program that causes a computer to do something.

Images