Information processing method, information processing apparatus, and information processing program

Abstract

To increase average values of processing capabilities of workers and prevent sharp decline of the capabilities, while reducing absolute work time of the workers from a long-term perspective.SOLUTION: An information processing method to be executed by a processor of an information processing apparatus, for determining interventions to facilitate behavioral change of a user includes: acquiring attribute data of the user, a command value of intervention effect, an actual measurement value of intervention effect to be obtained from log information obtained by executing an intervention on the user; calculating a difference between the actual measurement value of intervention effect and the command value; adjusting a parameter of a model that determines interventions based on the difference; calculating a predicted value of intervention effect based on the user attribute data and information on the intervention; and repeatedly adjusting the parameter of the model until the difference between the predicted value of intervention effect and the command value of intervention effect falls below a predetermined threshold.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] FIELD Embodiments of the present invention relate to an information processing method, an information processing device, and an information processing program. [Background technology]

[0002] In recent years, in order to improve the efficiency of item picking work in logistics warehouses and other locations, systems have been introduced in which automated guided vehicles (AGVs) are used to transport shelves containing items to multiple picking stations (PSs).At each picking station, picking means such as workers or robots pick items from the shelves transported by the AGVs.

[0003] For example, a delay in the work of a picking worker can affect the next shelf transport, causing a decrease in the throughput of the entire process. Therefore, by displaying the target work time to the worker via a terminal such as a display, processing capacity can be improved and the work time can be shortened. In this way, when it is difficult for people to act as expected by the system, appropriate intervention measures can be implemented to encourage desirable behavioral changes, leading to overall optimization and the expected effect of improving the operational efficiency of social infrastructure.

[0004] On the other hand, since there are individual differences in the responses of picking workers to intervention measures, ingenuity is required to realize the intervention effects envisioned by the infrastructure manager. For example, Patent Document 1 proposes a method known as feedback control, in which an intervention measure is selected according to the difference between the person's behavior information after the intervention and a target value, and the selection parameters are dynamically updated. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Patent No. 6833032 Summary of the Invention [Problem to be solved by the invention]

[0006] When the characteristics of the person being intervened change from moment to moment, the technology described in Patent Document 1 has the problem that it is difficult to update parameters to keep up with those changes, making it difficult to achieve the expected intervention effect.

[0007] This invention has been made in light of the above circumstances, and its purpose is to provide a method for updating the parameters of a model that determines intervention measures so that the target intervention effect can be achieved even when human characteristics change from moment to moment. [Means for solving the problem]

[0008] In one embodiment, an information processing method executed by a processor of an information processing device for determining an intervention measure to encourage behavioral change in a user includes acquiring attribute data of the user, a command value for an intervention effect, and an actual measured value of the intervention effect obtained from log information of the intervention measure being implemented on the user, calculating the difference between the actual measured value of the intervention effect and the command value, adjusting parameters of a model that determines the intervention measure based on the difference, calculating a predicted value of the intervention effect based on the attribute data of the user and information about the intervention measure, and repeating the parameter adjustment of the model until the difference between the predicted value of the intervention effect and the command value for the intervention effect becomes smaller than a predetermined threshold. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram conceptually illustrating an example of the configuration of a picking system according to an embodiment. [Figure 2] FIG. 2 is a block diagram showing an example of the configuration of a picking system. [Figure 3] FIG. 3 is a block diagram illustrating an example of the configuration of a host management device in the picking system according to the embodiment. [Figure 4] FIG. 4 is a block diagram showing an example of the configuration of an AGV in the picking system according to the embodiment. [Figure 5] FIG. 5 is a diagram illustrating an example of functions of the upper management device in the picking system according to the embodiment. [Figure 6] FIG. 6 is a diagram illustrating an example of a process in which the work data creation unit creates picking work data. [Figure 7] FIG. 7 is a diagram illustrating an example of a process in which the data reforming unit reforms picking operation data. [Figure 8] FIG. 8 is a diagram illustrating an example of processing performed by the processing capacity prediction unit and the work planning unit. [Figure 9] FIG. 9 is a block diagram for explaining the process of determining an intervention measure for a user. [Figure 10] FIG. 10 is a flowchart for explaining an example of the operation of the upper management device in the retrieval work of an item in the picking system according to the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, an information processing method, an information processing device, and an information processing program will be described in detail with reference to the drawings. In the following embodiments, parts with the same numbers perform the same operations, and redundant description will be omitted. For example, when there are multiple identical or similar elements, a common symbol may be used to describe each element without distinguishing between them, or a subnumber may be used in addition to the common symbol to describe each element with distinction between them.

[0011] In the following description, the term "A" or "B" means at least one of A or B, and the term "A," "B," or "C" means at least one of A, B, or C. Furthermore, the term "A" and "B" also means at least one of A and B, and the term "A," "B," and "C" means at least one of A, B, and C.

[0012] First, a picking system according to an embodiment is a system included in a logistics system installed in a logistics center, a warehouse, or the like, and is a system for picking items from shelves. The picking system transports shelves to a picking station using an automated guided vehicle (AGV). The picking system picks items from shelves at the picking station and controls a picking operation to sort the picked items into designated trays. The picking system causes picking means such as a worker or a robot to pick items from shelves.

[0013] FIG. 1 is a diagram conceptually illustrating an example of the configuration of a picking system 100 according to an embodiment.

[0014] FIG. 2 is a block diagram showing an example of the configuration of the picking system 100.

[0015] 1 and 2, the picking system 100 includes a plurality of picking stations P (three in the example of FIG. 1), an upper management device 1, a network 2, an external system 3, an AGV 7, and an AGV shelf 8. The picking station P includes a tray 10, a user interface 112, and a user 113. The network 2 is connected to the upper management device 1, the AGV 7, and the user interface 112. The network 2 is also connected to the external system 3.

[0016] The external system 3 transmits an order list indicating the items to be picked to the upper management device 1. The external system 3 can be realized by one or more computers. For example, the external system 3 may be equipped with a warehouse management system (WMS) that is responsible for managing the inbound and outbound or inventory management of items.

[0017] The host management device 1 (information processing device) manages information related to items stored in a warehouse. For example, the host management device 1 stores item management information indicating the items stored on each AGV shelf 8. The host management device 1 also acquires, from an external system 3, an order list indicating the items to be picked and a control target (command value) of the intervention effect to be achieved by the intervention measure. The host management device 1 transports the AGV shelf 8 to each picking station P based on the order list. Furthermore, the host management device 1 instructs a user 113 waiting at the picking station P to perform the picking work via a user interface 112. Furthermore, the host management device 1 acquires information related to the processing of items at each picking station P.

[0018] Furthermore, the upper management device 1 executes intervention measures for the user 113 via the user interface 112. Details of the control objectives and intervention measures will be described later.

[0019] A warehouse control system (WCS) for controlling the warehouse facilities may be provided between the host management device 1 and the user interface 112 and AGV 7. Some of the functions of the host management device 1, which will be described later, may be performed by the WMS or WCS.

[0020] The network 2 relays communications between the upper management device 1, the AGV 7, and the user interface 112. For example, the network 2 is a local area network (LAN). The AGV 7 may be connected to the network 2 wirelessly.

[0021] The AGV 7 transports the AGV shelf 8 to the picking station P. The AGV 7 operates based on control signals from the upper management device 1 or the user interface 112. For example, the AGV 7 travels toward a designated loading position and lifts up the AGV shelf 8 at the designated loading position. The AGV 7 travels toward a designated unloading position and unloads the AGV shelf 8 at the designated unloading position (for example, a specified picking station P). After the picking operation at the picking station P is completed, the AGV 7 returns the AGV shelf 8 to the item storage area.

[0022] The AGV shelf 8 is arranged in the item storage area. The AGV shelf 8 is a shelf that stores items. For example, the AGV shelf 8 stands upright on four support pillars. The height below the shelf of the AGV shelf 8 (height from the floor to the bottom of the shelf) is higher than the height of the AGV 7. This allows the AGV 7 to slip under the shelf of the AGV shelf 8. After slipping under the shelf, the AGV 7 uses a pusher to lift the AGV shelf 8 so that the tips of the support pillars are a few centimeters above the floor, and then travels with the AGV shelf 8 lifted. In this way, the AGV 7 transports the AGV shelf 8.

[0023] Furthermore, shelf identification information that can be read by a fixed camera or a mobile camera may be attached to the AGV shelves 8. Item identification information that can be read by a fixed camera or a mobile camera may also be attached to the items. For example, the shelf identification information and item identification information are barcodes or two-dimensional codes. Note that the picking system 100 may be equipped with multiple readers that read this shelf identification information and item identification information, in addition to the fixed or mobile cameras.

[0024] In the example shown in Fig. 1, a picking station P is placed in a retrieval work area. The picking station P receives an AGV shelf 8 transported by an AGV 7. The picking station P picks an item from the received AGV shelf 8.

[0025] Trays 10 corresponding to each piece of allocated order information are prepared in the picking station P. At the picking station P, picked items are placed on the trays 10. At the picking station P, trays 10 corresponding to each piece of allocated order information are prepared, and when all the necessary items are stored in a tray 10, it is ejected from the picking station P and a new empty tray 10 is replenished. Once the tray 10 has been ejected and is filled with the necessary items, it is sent to the subsequent process in the warehousing work area, where it is packed, sorted, and loaded.

[0026] The user interface 112 manages the picking work at the picking station P. The user interface 112 includes a display unit, an operation unit, and the like. For example, the user interface 112 is a digital assortment system (DAS). The user interface 112 displays information indicating the items to be picked on the display unit in accordance with control from the upper management device 1. The user interface 112 also accepts input of an operation indicating the items that have been picked from a worker as a user 113. The user interface 112 transmits information indicating the input operation to the upper management device 1.

[0027] The DAS as the user interface 112 scans the barcode of an item picked by the user 113 from the AGV shelf 8, and a lamp corresponding to the tray 10 into which the scanned item is to be sorted lights up. When the user 113 has finished sorting the items onto the tray 10, the worker presses a switch located at the position of that tray 10, completing the work. At this time, the user interface 112 stores work completion information in response to the switch being pressed. Then, based on the received work completion information, the user interface 112 creates a work log that indicates work results (work performance), such as work time. After creating the work log, the user interface 112 supplies the created work log to the upper management device 1.

[0028] The user 113 picks up an item from the AGV shelf 8 in accordance with instructions from the upper management device 1. In this embodiment, the user 113 is a worker. As shown in FIG. 1 and other figures, only the user 113 is shown as a picking means, but it goes without saying that the user 113 may include both a worker and a picking robot.

[0029] The user 113 visually checks the items displayed on the user interface 112 and picks the items. The user 113 places the picked items on the tray 10. After placing the items on the tray 10, the user 113 inputs an operation to the user interface 112 indicating the items that have been picked. The user 113 also receives an intervention measure via the user interface 112.

[0030] The picking station P may be equipped with a camera or the like that monitors the actions of the user 113. The start and end of the work of the user 113 may be determined based on images captured by the camera. In addition, the presence or absence of interruption processing or abnormalities during work may be determined, and the information may be added to the work log.

[0031] Next, the configuration of the upper management device 1 in the picking system 100 according to the embodiment will be described.

[0032] FIG. 3 is a block diagram showing an example of the configuration of the upper management device 1 in the picking system 100 according to the embodiment.

[0033] The upper management device 1 is configured by one or more computers. For example, the upper management device 1 includes a processor 11, a ROM 12, a RAM 13, an auxiliary storage device 14, a communication interface 15, and the like.

[0034] The processor 11, the ROM 12, the RAM 13, the auxiliary storage device 14, and the communication interface 15 are connected to one another via a data bus or an interface.

[0035] The processor 11 has the function of controlling the overall operation of the host management device 1. The processor 11 may also include an internal cache and various interfaces. The processor 11 performs various processes by executing programs stored in advance in the internal memory, the ROM 12, or the auxiliary storage device 14.

[0036] For example, the processor 11 is a central processing unit (CPU). The processor 11 may be realized by hardware such as a large scale integration (LSI), an application specific integrated circuit (ASIC), or a field-programmable gate array (FPGA).

[0037] The ROM 12 is a non-transitory computer-readable storage medium that stores the above-mentioned programs, and also stores data and various setting values ​​used by the processor 11 when performing various processes.

[0038] The RAM 13 is a memory used for reading and writing data, and is used as a so-called work area for storing data that is temporarily used when the processor 11 performs various processes.

[0039] The auxiliary storage device 14 is a non-transitory computer-readable storage medium and may store the above-mentioned programs. The auxiliary storage device 14 also stores data used by the processor 11 when performing various processes, data generated by the processes of the processor 11, various setting values, etc.

[0040] The auxiliary storage device (storage device) 14 pre-stores item management information indicating items stored on each AGV shelf 8. For example, the item management information stores a shelf code indicating the AGV shelf 8 and an item code indicating the item stored on the AGV shelf 8 in association with each other.

[0041] The host management device 1 may be transferred with the above program stored in the ROM 12 or the auxiliary storage device 14, or may be transferred without the above program stored therein. In the latter case, the host management device 1 reads the above program stored in a removable storage medium such as an optical disk or semiconductor memory, and writes the read program to the auxiliary storage device 14. Alternatively, the host management device 1 downloads the above program via the network 2 or the like, and writes the downloaded program to the auxiliary storage device 14.

[0042] The communication interface 15 is an interface for transmitting and receiving data to and from various devices. The communication interface 15 is connected to the AGV 7, the user interface 112, and the like. The communication interface 15 also acquires order lists and the like from the external system 3, and the like. For example, the communication interface 15 supports LAN (local area network) connections and the like. The communication interface 15 may also be configured to include an interface for transmitting and receiving data to and from the external system 3, an interface for transmitting and receiving data to and from the AGV 7, and an interface for transmitting and receiving data to and from the user interface 112.

[0043] Next, the configuration of the AGV 7 in the picking system 100 according to the embodiment will be described.

[0044] FIG. 4 is a block diagram showing an example of the configuration of the AGV 7 in the picking system 100 according to the embodiment.

[0045] The AGV 7 includes a processor 71, a ROM 72, a RAM 73, an auxiliary storage device 74, a communication interface 75, a drive unit 76, a sensor 77, a battery 78, a charging mechanism 79, and tires 70.

[0046] The processor 71 has the function of controlling the overall operation of the AGV 7. The processor 71 may also include an internal cache and various interfaces. The processor 71 performs various processes by executing programs stored in advance in the internal memory, the ROM 72, or the auxiliary storage device 74.

[0047] For example, the processor 71 is a CPU. The processor 71 may be realized by hardware such as an LSI, an ASIC, or an FPGA.

[0048] The processor 71 executes a program stored in the ROM 72 or the like to generate control signals based on transport instructions received via the communication interface 75, and performs processes such as calculations and controls required for operations such as acceleration, deceleration, stopping, direction changes, and raising and lowering of the AGV shelf 8. The processor 71 also transmits work performance data of the AGV 7 to the upper management device 1 or the user interface 112 via the communication interface 75.

[0049] For example, the host management device 1 or the user interface 112 transmits a control signal to move the AGV 7 from its current position to a first position (the position of the target AGV shelf 8) and then from the first position to a second position (the position of the target picking station P). The host management device 1 or the user interface 112 also transmits a control signal to move the AGV 7 from the second position to the first position. The processor 71 of the AGV 7 outputs a drive signal in response to the control signal transmitted from the host management device 1 or the user interface 112. This causes the AGV 7 to move from its current position to the first position, from the first position to the second position, and from the second position to the first position. The processor 71 also outputs a drive signal in response to an instruction to load or unload the AGV shelf 8, which is included in the control signal transmitted from the host management device 1 or the user interface 112. This causes the AGV 7 to lift the AGV shelf 8 using its pusher and lower the lifted AGV shelf 8.

[0050] The ROM 72 is a non-transitory computer-readable storage medium that stores the above-mentioned programs. The ROM 72 also stores data or various setting values ​​used by the processor 71 when performing various processes. The RAM 73 is a memory used for reading and writing data. The RAM 73 is used as a so-called work area for storing data that is temporarily used by the processor 71 when performing various processes.

[0051] The auxiliary storage device 74 is a non-transitory computer-readable storage medium and may store the above-mentioned programs. The auxiliary storage device 74 also stores data used by the processor 71 when performing various processes, data generated by the processes of the processor 71, various setting values, etc.

[0052] The communication interface 75 is an interface for transmitting and receiving data to and from the higher level management device 1 or the user interface 112 via a wireless LAN access point, etc. For example, the communication interface 75 supports wireless LAN connection.

[0053] The drive unit 76 is a motor or the like, and rotates or stops the motor based on a drive signal output from the processor 71. The power of the motor is transmitted to the tires 70 and then to the steering mechanism. The power from such a motor moves the AGV 7 to a target position.

[0054] Furthermore, with the AGV 7 submerged under the AGV shelf 8, the drive unit 76 rotates the motor (forward rotation) based on the drive signal output from the processor 71. The power from this motor raises the pusher, lifting the AGV shelf 8. Furthermore, after the AGV 7 reaches the destination position, the drive unit 76 rotates the motor (reverse rotation) based on the drive signal output from the processor 71. The power from this motor lowers the pusher, lowering the AGV shelf 8 to the floor.

[0055] The sensor 77 is a plurality of reflective sensors. Each reflective sensor is attached around the AGV 7. Each reflective sensor emits a laser beam, detects the time it takes for the laser beam to be reflected by an object and return, detects the distance to the object based on the detected time, and notifies the processor 71 of a detection signal.

[0056] The processor 71 outputs a control signal for controlling the travel of the AGV 7 based on a detection signal from the sensor 77. For example, the processor 71 outputs a control signal for stopping at a specified position, decelerating to avoid a collision, or stopping, based on the detection signal. Note that a camera may be provided in addition to the sensor 77, and the camera may capture images of the surrounding area and output the captured images to the processor 71. In this case, the processor 71 analyzes the captured images and outputs a control signal for decelerating or stopping to avoid a collision with an object.

[0057] The battery 78 supplies the necessary power to the drive unit 76 etc. The charging mechanism 79 is a mechanism that connects the charging station and the battery 78, and the battery 78 is charged with power supplied from the charging station etc. via the charging mechanism 79.

[0058] Next, functions of the upper management device 1 in the picking system 100 according to the embodiment will be described.

[0059] FIG. 5 is a diagram showing an example of functions of the upper management device 1 in the picking system 100 according to the embodiment.

[0060] The host management device 1 has a database 121 provided in a storage device such as the internal memory of the processor 11, the ROM 12, or the auxiliary storage device 14. The database 121 as a storage device stores various types of data. For example, the processor 11 performs necessary processing by executing a program stored in the database 121. The data stored in the database 121 is output to the outside of the host management device 1 via the communication interface 15. The data input to the host management device 1 via the communication interface 15 is saved in the database 121.

[0061] The various data stored in the database 121 include order lists obtained from the external system 3, control targets based on intervention measures, picking work data, work logs, and formatted data based on work results, data on items stored in the warehouse, data on users 113, data on AGVs 7, work plan data, program data, model data, etc.

[0062] The work log is data related to the work results for the instructed picking work. The work log is data in which, for example, the work results (data of the user 113, work time, whether or not there was an interrupt process or an abnormality, etc.) acquired from the user interface 112 are added to each performed picking work unit (item ID, number of items picked, item storage position on the shelf, tray position to which the item is inserted, whether or not a step ladder is used, etc.).

[0063] The work log for AGV 7 is, for example, a log in which work results (equipment ID, position at start of work, movement route, whether collision avoidance action was taken, movement distance, power consumption, transport time, etc.) obtained via communication interface 15 are attached to each transport work unit (shelf ID of AGV shelf 8 transported, initial position of AGV 8 shelf, destination picking station ID).

[0064] Formatted data is preprocessed into the input data format required to run various programs, and is generated by formatting picking work data related to picking work. Formatted data is generated and saved as appropriate when an order list or work log is acquired.

[0065] The item data is data about the items to be picked. Examples of the item data include the item ID, quantity, weight, volume, shape, material, storage shelf ID, storage position on the shelf, and storage status (whether the cardboard box is opened or unopened).

[0066] The data of user 113 is user attribute data, which is data related to user 113. The attribute data of user 113 as data of user 113 includes, for example, user ID, age, sex, height, exercise experience, years of employment, employment type, and duration of employment. Note that the data of user 113 may not be stored in database 121, but may be stored in external system 3. The upper management device may acquire necessary data of user 113 from external system 3 when necessary, temporarily store the data in database 121, etc., and delete it from database 121 when it is no longer necessary.

[0067] The data of the AGV 7 is data related to the AGV 7. Examples of the data of the AGV 7 include the device ID, years of use, charge rate, and status (standby, in use, returning).

[0068] The work plan data is data that represents plans for allocating picking work to picking stations P and controlling AGVs 7 to transport AGV shelves 8 in order to process the acquired order list. These plans are revised and updated as appropriate according to the progress of work in the warehouse.

[0069] The programs stored in the database 121 are programs executed by the processor 11, such as programs related to work data creation, data shaping, processing capacity prediction, work planning, policy effect prediction, policy decision-making, model evaluation, and model update.

[0070] 5, the upper management device 1 includes a task data creation unit 101, a data shaping unit 102, a processing capacity prediction unit 103, a task planning unit 104, a policy decision unit 105, a model evaluation unit 106, a policy effect prediction unit 107, a model update unit 108, and a simulator 109. The task data creation unit 101, the data shaping unit 102, the processing capacity prediction unit 103, the task planning unit 104, the policy decision unit 105, the model evaluation unit 106, the policy effect prediction unit 107, the model update unit 108, and the simulator 109 are functions realized by the processor 11 executing programs stored in a database 121 of the processor 11 or a storage device such as the internal memory of the processor 11, the ROM 12, or the auxiliary storage device 14.

[0071] The processor 11 executes a program that creates picking operation data from the order list, thereby functioning as an operation data creation unit 101. The operation data creation unit 101 creates a list of picking operation units (item ID, number of items to be picked, weight, volume, item storage position within the AGV shelf 8, position of tray 10 into which the items are to be placed, etc.) to be performed at each picking station P based on the order list acquired from the external system 3 and item data stored in the database 121.

[0072] The processor 11 functions as a data formatting unit 102 by executing a program that formats picking operation data. The data formatting unit 102 formats picking operation data (including data with operation results attached) into a format acceptable to various programs for processing capacity prediction, operation planning, action effectiveness prediction, and action decision-making. The formatting process includes linking the picking operation data with data from the user 113 and data on the intervention measures to be implemented, as well as preprocessing such as variable transformation and scaling for each data column in the table data. Note that before the operation results are obtained as a log, if the information to be linked has not been determined, preprocessing is performed on the finalized picking operation data or on data to which information is provisionally linked based on the predicted situation.

[0073] The processor 11 functions as a processing capacity prediction unit 103 by executing a program that predicts the processing capacity of the user 113. The processing capacity prediction unit 103 generates data on the processing capacity of each user 113 for each picking task to be performed based on the formatted picking task data, and outputs the generated data to the task planning unit 104, which will be described later.

[0074] The processing capacity of the user 113 is, for example, the work time required for the picking work. The processor 11 may use a digital assortment system to calculate the work time (the processing capacity of the user 113) from the difference between the arrival time of the AGV 7 and the time when a switch located at the position of the tray 10 to which the items are to be placed is pressed. Alternatively, the processor 11 may determine the start and end times of the picking work from images captured by a camera installed at the picking station P, and calculate the work time.

[0075] The processor 11 executes a program for carrying out work planning, thereby functioning as a work planning unit 104. The work planning unit 104 formulates work plans for work allocation to each picking station P and for the AGVs 7 used for shelf transport, based on the processing capacity data of the user 113 output by the processing capacity prediction unit 103 and data such as the operational status within the warehouse.

[0076] The processor 11 executes a program for determining an intervention measure, thereby functioning as the measure determination unit 105. The measure determination unit 105 outputs an intervention measure required to cause the user 113 to make a target behavioral change.

[0077] The processor 11 functions as a model evaluation unit 106 by executing a program for evaluating the performance of a model related to the determination of an intervention measure. As a first function, the model evaluation unit 106 compares an actual measurement value of the intervention effect obtained by executing an intervention measure with a command value input from the external system 3, and calculates an evaluation result regarding the control performance of the measure decision unit 105. As a second function, it compares the actual measurement value of the intervention effect with a predicted value output by the measure effect prediction unit 107, which will be described later, and calculates an evaluation result regarding the prediction performance of the measure effect prediction unit 107. If the performance does not satisfy a predetermined standard, the model evaluation unit 106 outputs a model update flag to the model update unit 108.

[0078] The processor 11 functions as a measure effect prediction unit 107 by executing a program that predicts the effect of an intervention measure. The measure effect prediction unit 107 calculates the intervention effect on the user 113 from information about the user 113 who is the intervention target and information about the intervention measure to be executed. Here, the intervention effect is, for example, the amount of increase in the processing capacity of the user 113 (the amount of reduction in picking work time) when the intervention measure is executed, with respect to the case where the intervention measure is not executed. Alternatively, the intervention effect may be a binary value indicating whether the processing capacity of the user 113 increases (whether the picking work time decreases) when the intervention measure is executed, with respect to the case where the intervention measure is not executed.

[0079] The processor 11 executes a program for updating a model, thereby functioning as a model updating unit 108. When the model updating unit 108 receives a model update flag from the model evaluation unit 106, it updates the model used by the policy effect prediction unit 107.

[0080] The processor 11 functions as a simulator 109. The simulator 109 will be described in detail later, but it is realized, for example, by an agent-based model in which autonomously acting actors (agents) are given behavioral rules and complex phenomena are expressed through their interactions. Here, the behavioral rules of the agents are constructed based on log data.

[0081] Next, the processing performed by the task data creation unit 101, which is a function of the upper level management device 1, will be described.

[0082] The host management device 1 acquires an order list and inventory data in the warehouse from the external system 3. The host management device 1 may also acquire shipping times, transport truck operation data, etc. from the external system 3. The host management device 1 stores the acquired data in the database 121.

[0083] Order lists are typically obtained in batches determined based on shipping time and truck constraints. Any reordering or reassignment of orders is performed within this batch.

[0084] Depending on the progress of work at each picking station P, interference with AGV shelf calls or waiting for work at the picking station P may occur. To avoid this, the upper management device 1 may change the execution order (the order in which the AGV shelves 8 are called up) within the range of assigned work, even after assigning work to the picking station P. If this is still difficult to avoid, or if a major change in the plan is necessary from the perspective of work leveling, the upper management device 1 may change the picking station P to which the work is assigned for the assigned work that has not yet had the AGV shelf 8 called up.

[0085] The range or timing of tasks for which the assigned picking station P can be changed varies depending on the number of trays 10 provided in the picking station P or the management method thereof. When trays 10 are replenished at each picking station P with order IDs and tray IDs linked in advance, the order of tasks corresponding to the replenished trays 10 can be changed, but the assignment cannot be changed to another picking station P. When order IDs and tray IDs are linked at each picking station P and can be changed up until just before the task begins, the upper management device 1 can change the plan over a wider range.

[0086] The work data creation unit 101 creates attributes X1, X2, ..., X related to the items and picking work based on the acquired order list and the inventory data stored in the database 121. n Here, n is an arbitrary positive integer representing the number of attributes. That is, the work data creation unit 101 creates picking work data related to the picking work in which the user 113 picks items based on the order list and inventory data.

[0087] The attributes of an item are data including the weight and volume of the item, the shelf ID where it is stored, its storage position within the shelf, its storage state (whether the cardboard box is opened or unopened), etc. Furthermore, the attributes of a picking operation are data including the number of items to be picked, the position of the tray 10 into which the items are to be inserted, whether a step stool is used, etc. The position of the tray 10 into which the items are to be inserted or whether a step stool is used is assigned at the time of confirmation.

[0088] FIG. 6 is a diagram illustrating an example of a process in which the work data creation unit 101 creates picking work data.

[0089] In the example of Figure 6, there are tasks with task numbers (task IDs) from "1" to "D", and attributes X relating to items and picking tasks are assigned to each task. For example, X1 is the number of items to be picked, X2 is the storage position within the AGV shelf 8, and X n indicates the tray position where the item is to be inserted.

[0090] Furthermore, when creating the picking work data, the work data creation unit 101 may rearrange the picking work so that picking work that can be performed simultaneously in one shelf call is grouped together, and may group multiple picking work together and assign them to a picking station P. In this case, the work data creation unit 101 links the shelf to be called for each group of work units.

[0091] The work data creation unit 101 may rearrange and organize the picking work under the assumption that the user 113 can improve the efficiency of the work in the following two ways. A If multiple items are listed in the order information, the same item I will be A A plurality of items may be picked and placed in a plurality of trays 10. The second is a method of picking different items I described in a certain order information. A ,I B If different items I are stored together on the called AGV shelf 8, A ,I B may be picked up together from the same AGV shelf 8. Also, the above two methods of grouping may be combined.

[0092] Next, a process of shaping picking operation data by the data shaping unit 102, which is a function of the upper level management device 1, will be described. FIG. 7 is a diagram for explaining an example of processing in which the data reforming unit 102 reforms picking operation data.

[0093] The formatted picking work data shown in FIG. 7 is generated by the work data generation unit 101, and the attributes Y1, Y2, . . . , Y3 of the user 113 are used for the work 1 to D of the picking work data. m , intervention data Z1, Z2, …, Z l 10 shows data obtained by applying preprocessing to each data string, with the workload W and the processing capacity T of the user 113 obtained from the work results assigned.

[0094] For example, the data reforming unit 102 may convert attribute data Y1, Y2, . . . , Y3 of the user 113 into attribute data Y4, Y5, Y6, Y7, Y8, Y9, Y10, Y11, Y12, Y13, Y14, Y15, Y16, Y17, Y18, Y19, Y20, Y21, Y22, Y23, Y24, Y25, Y26 m , intervention data Z1, Z2, …, Z lThe picking task data is shaped, including a process of adding information such as the workload W and the processing capacity T of the user 113 obtained from the task results, and pre-processing of each data string. The shaped picking task data is stored in the database 121. Here, m and l are arbitrary positive integers representing the number of variables expressing the attributes of the user 113 and the number of intervention measures, respectively. The data X relating to the picking task is divided into X'1, X'2, ..., X' to distinguish between pre-processing and post-processing. n’ where n'≒n.

[0095] In the example of FIG. 7, the number of items, weight, age of the user 113, etc. are treated as continuous values, and when the range of values ​​that can be taken by the variables differs, the data shaping unit 102 performs scaling processing. For example, processing is performed to normalize the value of each variable to a specified range. For example, the processing to normalize to the range of [0, 1] is performed using the following formula 1. Here, x (i) is the i-th sample of the data x, and x min and x max are the minimum and maximum values ​​of data x, respectively, and x norm (i) is the normalized variable.

[0096] x norm (i) =(x (i) -x min ) / (x max -x min ) (Formula 1)

[0097] In the example of Figure 7, attributes related to items and picking work are converted into dummy variables (binary values ​​of 0 or 1) or into continuous values, and scaling is performed on the data in each cell. Therefore, in the example of Figure 7, to clearly indicate that the attributes are related to items and picking work that have been shaped, they are indicated by the above-mentioned X'.

[0098] The storage location of an item and the destination tray 10 are managed using letters or numbers, for example, as shown in Figure 6. Information regarding intervention measures is also managed as Measure 1, Measure 2, ..., Measure 1. The size of the letters or numbers has no meaning, so to make the information meaningful for data processing, they are converted into dummy variables (binary values ​​of 0 or 1). For intervention measures, implementing Measure 1 is represented as Z1 = 1, and not implementing Measure 1 is represented as Z1 = 0. Note that the storage location of an item and the destination tray 10 can also be converted into continuous values ​​such as the height of the storage location or the distance to the tray 10, based on the equipment information of the facility.

[0099] Attribute data Y1, Y2, ..., Y of user 113 m For example, this includes worker ID, age, sex, height, exercise experience, years of employment, employment type, the duration of work that has elapsed since the start of work at the current time, fatigue level, etc.

[0100] Here, the fatigue level may be estimated based on, for example, the pulse and body movements measured by a wristband sensor worn by the worker. Alternatively, the fatigue level may be estimated based on the captured images of the worker's work movements captured by a camera installed at the picking station P. Alternatively, the fatigue level may be estimated based on the performance of the worker's processing capacity, which is collected as a work log.

[0101] Intervention data Z1, Z2, …, Z l These include the presentation of information via the user interface 112, changes to the work assignments, and constraints on the workload W. For example, when intervention measure data indicating a target work time is associated with Z1, implementing this intervention measure data is represented as Z1=1, and not implementing it is represented as Z1=0.

[0102] The intervention data for information presentation includes information related to the target work time and the estimated time for completing the work. For example, if you want to increase the pace of a worker, you can present a target work time faster than the current time or the estimated time for completing the work at the current pace. If you want to slow down the pace of a worker, you can present a target work time slower than the current time.

[0103] Furthermore, the intervention measure data regarding information presentation includes presenting information to support the picking work that the worker should perform, presenting characters, playing music, presenting the amount of incentive, presenting the ranking of the workers, and the like.

[0104] The intervention measure data regarding changes in assigned work quantifies the load of the picking work based on the work attributes, and when the user 113 is working at a slow pace or has reduced processing ability due to fatigue, a less demanding work is assigned to that user 113, and when the user 113 is working at a faster pace than expected or is rushing too quickly, a more demanding work is assigned.

[0105] The workload W is a numerical value that represents the physical load when performing, for example, picking work. The workload is calculated based on the attributes X1, X2, ..., X n and attribute data Y1, Y2, ..., Y of user 113 m In addition, data with preset values ​​can be prepared and the corresponding values ​​can be read in. In addition, attributes X1, X2, ..., X related to the item and picking work can be n , attribute data Y1, Y2, ..., Y of user 113 m A model for estimating the load may be prepared by analyzing the relationship between the actual values ​​of the fatigue level and the actual values ​​of the fatigue level.

[0106] The workload W will be higher, for example, if items stored in a high place cannot be picked without using a step ladder. The farther the tray 10 into which the items are to be placed is located from the picking position, the higher the workload W will be. Alternatively, the greater the number or weight of the items, the higher the workload W will be.

[0107] Conversely, for example, if the items are stored at about waist height and can be picked without using a step stool, the workload W will be lower. Also, the closer the tray 10 into which the items are to be placed is located from the picking position, the lower the workload W will be. Alternatively, the smaller the number or weight of the items, the lower the workload W will be.

[0108] The constraint on the workload W included in the intervention measure data refers to a measure that limits the allocation to work with a small workload W or work with a large workload W. If you want to slow down the work pace of a specific worker, you can set a constraint on the workload W using a measure such that W<0.3, for example.

[0109] The processing capacity T of the user 113 is, for example, the work time required for the picking work.

[0110] The work time may be calculated, for example, by using a digital assortment system as the user interface 112, from the difference between the arrival time of the AGV 7 and the time when a switch located at the position of the destination tray 10 is pressed. Alternatively, the upper management device 1 may determine the start and end times of the picking work from images captured by a camera installed at the picking station P, and calculate the work time.

[0111] The host management device 1 may determine whether there was an interruption or an abnormality during work from the camera image as work performance, and add that information to the work log as flag data. The host management device 1 may also perform various data processing or analysis excluding the flagged data.

[0112] The preprocessing for each data sequence is, for example, preprocessing such as variable transformation or scaling.

[0113] Next, the processing performed by the processing capacity prediction unit 103 and the work planning unit 104, which are functions of the upper management device 1, will be described.

[0114] FIG. 8 is a diagram for explaining an example of the processing performed by the processing capacity predicting unit 103 and the work planning unit 104. In FIG.

[0115] In the example of FIG. 8, it is assumed that tasks 1 to D are assigned to one of users a to c.

[0116] First, the processing capacity prediction unit 103 predicts the processing capacity of each user 113 for the picking work to be assigned based on the formatted picking work data. Then, the work planning unit 104 formulates an optimal work assignment combination as a work plan from among the feasible assignment combinations while satisfying various constraints. Figure 8 shows an example in which each work is assigned to user a to user c.

[0117] The prediction of processing capacity may involve, for example, extracting from past work logs the performance results of a picking operation that has similar conditions to the next picking operation that is scheduled to be performed, and using the performance results as the predicted value.

[0118] Alternatively, a function (prediction model) may be constructed using machine learning, with attribute data related to the work to be performed as explanatory variables and processing capacity as the objective variable. Note that processing capacity may be a probability distribution of work time, in which case the mean value of the probability distribution and the value at one side α point may be used as a representative value of processing capacity for the work time.

[0119] A processing capacity prediction model may be constructed individually for each user 113, or multiple models may be constructed classified by the attributes of the user 113. Furthermore, when predicting the processing capacity of a user 113 who has no past work history, the capacity may be calculated from work history with similar attributes to the user 113 in a past work log. Alternatively, a model for a user 113 with similar attributes may be selected from already constructed prediction models, and operation may be started.

[0120] The various constraints considered by the work planning unit 104 are, first, that the plan is feasible based on the operating status of the AGV shelves 8 and AGVs 7, and that the latest time at which each picking station P finishes processing all assigned work does not exceed the process deadline, which is determined based on the shipping time. Among these, the optimal work plan is one that equalizes the work completion times of each picking station P and minimizes the total required time for the longest picking station P. Therefore, intervention measures for each user 113 are selected in the hope of shortening the total required time for the picking stations P, and that, in the long run, they will increase the average processing capacity compared to a case where no intervention measures are taken.

[0121] Possible intervention measures for the user 113 include presenting information via the user interface 112, such as presenting a target work time or an estimated time for completing a task, presenting information to support the picking task, presenting a character, playing music, presenting an incentive amount, presenting a worker ranking, etc. For example, when it is desired to increase the processing capacity of the user 113, a target work time that is faster than the current time or an expected time for completing a task based on the current processing capacity is presented. Another possible measure is to change the content of the assigned work, and when processing capacity is reduced due to fatigue or the like, work with a small workload W is assigned to restore processing capacity.

[0122] If there is a user 113 with outstanding processing ability within the entire process and it is expected that the leveling of work completion times will be disrupted, a target work time that is later than the current time may be presented, or a work with a large workload W may be assigned to the user.

[0123] The intervention measures for each user 113 do not need to be uniform for all users, but may be implemented individually according to the attributes of each user that are deemed appropriate at the time. For this reason, a measure may be implemented to present a target time to one user 113, while a measure may be implemented to limit the allocation of work to picking work, which is light on the workload, to another user 113. It may also be determined that it is best not to implement any measures.

[0124] Next, the operation of the upper management device 1 in the picking system 100 according to the embodiment will be described.

[0125] FIG. 9 is a block diagram for explaining the process of determining an intervention measure for the user 113. As shown in FIG.

[0126] Here, the focus is on the functions required for the relevant processing, and functions related to the processing of determining the work plan of the picking system 100 of the embodiment are omitted. Also, the physical system in Fig. 9 refers to the real space where the user 113 to be intervened exists. In the picking system 100 of this embodiment, the physical system is a logistics warehouse where the user 113, picking station P, AGV 7, AGV shelf 8, etc. exist.

[0127] The policy decision unit 105 outputs an intervention policy required to bring about a target behavioral change in the user 113 who is the intervention target. In general, it is considered that the effect of an intervention policy differs depending on the attributes of the user 113, and the policy decision unit 105 is assumed to have a function that can select an appropriate policy from candidate policies for each user 113, taking this into consideration. That is, the policy decision unit 105 outputs an intervention policy required to bring about a target behavioral change in the user 113 who is the intervention target. m and has a function f that selects and outputs one of intervention measures Z from intervention measure 1, intervention measure 2, ..., intervention measure l according to the content. That is, Z is expressed as in the following equation 2.

[0128] Z=f(Y1,Y2,…,Y m ) (Formula 2)

[0129] The output intervention plan is stored in the database 121 and is sent to the user 113 via the communication interface 15, the network 2, and the user interface 112.

[0130] As a first function, the model evaluation unit 106 compares the results (actual measurements) obtained from the user 113 of the physical system with the control target (command value) input from the external system 3 regarding the intervention effect obtained by executing the intervention measure, and calculates an evaluation result regarding the control performance of the measure decision unit 105. Here, as described above, for example, in the case of the picking system 100 of this embodiment, the intervention effect is the increase in the processing capacity of the user 113 (the amount of reduction in picking work time) when the intervention measure is executed, with the case where the intervention measure is not executed as a reference. Alternatively, it may be a binary value indicating whether the processing capacity of the user 113 increases (whether the picking work time decreases) when the intervention measure is executed, with the case where the intervention measure is not executed as a reference. The control target (command value) of the intervention effect may be a value specified individually for each user 113, or a single value may be specified for the entire physical system. In the latter case, it is assumed that the model evaluation unit 106 also has a function of converting the intervention effect for each user 113 and the intervention effect for the entire system into each other.

[0131] That is, the first function of the model evaluation unit 106 is to calculate the actual value U act and the control target (command value) U ref The function g outputs an evaluation result R relating to the control performance from the above. That is, the evaluation result R is expressed as in the following Equation 3.

[0132] R=g(U act ,U ref ) (Formula 3)

[0133] For example, regarding the intervention effect, if the user 113 changes his / her behavior as a target when the intervention measure is implemented, U=1, and if the user does not change his / her behavior as a target when the intervention measure is implemented, U=0, then there is an XNOR gate as a simple function g. act and the control target (command value) U ref If they match, R=1 is output, otherwise R=0 is output, and the policy decision unit 105 adjusts the model parameters based on this value. In addition, the intervention effect U is treated as a continuous value, and the actual measured value U act and the control target (command value) U refIt is also possible to set a function g such that the closer the values ​​of are, the larger the R (reward) becomes.

[0134] Based on the evaluation result R received from the model evaluation unit 106, the policy decision unit 105 continues the current intervention policy if the control target has been achieved, and if not, tries a different intervention policy and obtains another evaluation result R. By repeating this feedback, the model used by the policy decision unit 105 is updated so that the actual measured value of the intervention effect approaches the control target (command value).

[0135] The policy decision unit 105 is realized, for example, by a bandit algorithm, which is a type of reinforcement learning. In the bandit algorithm, the intervention policy described above is called an arm, and the result obtained by selecting a certain arm is used as a reward, and the arm selection parameters are adjusted to maximize the final cumulative reward. In other words, in this case, the model evaluation unit 106 corresponds to a function that compares the result (actual value) obtained from the user 113 of the physical system with the control target (command value) input from the external system 3 and calculates the reward for the selected arm. In a typical bandit algorithm, the intervention effect obtained from the user 113 is often used as the reward as is, but the presence of the model evaluation unit 106 allows the administrator to arbitrarily specify the control target (command value) of the intervention effect.

[0136] The feedback loop (hereinafter referred to as Loop 1) formed between the aforementioned policy decision unit 105, the physical system, and the model evaluation unit 106 is the basic flow for adjusting the parameters of the model that determines the intervention policy. However, if the user 113's response to the intervention policy changes from moment to moment, the parameter adjustment cannot keep up with the changes, and it may be difficult to achieve the control goal of the intervention effect. Specifically, reinforcement learning (bandit algorithm), which is one of the implementation methods, usually requires a large number of trials, but it is difficult to ensure a sufficient number of trials using only the feedback of Loop 1. In addition, repeated trials may actually have an unintended effect on the user 113. For example, if the user 113 is repeatedly administered a large number of times, the user 113's response may change as the user 113 becomes accustomed to or bored with the intervention.

[0137] Therefore, the upper management device 1 has a function of adjusting the parameters of a model that determines an intervention measure using another feedback loop via the measure effect prediction unit 107.

[0138] The policy effect prediction unit 107 calculates a predicted value of the intervention effect for the intervention measure to be executed. As described above, this prediction is assumed to be executed taking into consideration that the effect of the intervention measure differs depending on the attributes of the user 113. That is, the policy effect prediction unit 107 calculates a predicted value of the intervention effect for the intervention measure to be executed Z and the attributes Y1, Y2, ..., Y m and the predicted value of the intervention effect U pred That is, the predicted value U pred is expressed as the following equation 4.

[0139] U pred =h(Z,Y1,Y2,…,Y m ) (Equation 4)

[0140] The above formula corresponds to constructing a trained prediction model using user attribute data and data related to the intervention measures as explanatory variables and the intervention effect as a target variable.

[0141] The policy effect prediction unit 107 is realized by, for example, counterfactual machine learning based on a supervised learning algorithm. Based on the log data obtained from the physical system, the data reforming unit 102 calculates the actual value U of the intervention effect. act Using the modified data with the addition of , as training data, a prediction model equivalent to the aforementioned function h is constructed. Here, when some intervention measure is implemented for user 113, since there is only one intervention measure that can be implemented under the same conditions, there exists data (counterfactual scenario) that could have been observed but was not actually observed. The aforementioned counterfactual machine learning is a general term for a methodology for estimating intervention results based on this counterfactual scenario, and for example, methods called Meta-Learner and Causal-Tree have been devised.

[0142] The upper management device 1 uses the current policy decision unit 105 to execute intervention measures on the physical system and collect log data, while constructing a model to predict the intervention effect to be used by the policy effect prediction unit 107. Here, the model is constructed, for example, as described above, by learning a prediction model from the formatted work log using user attribute data and data related to the intervention measure as explanatory variables and the actual measured value of the intervention effect as the objective variable. After construction, a feedback loop (hereinafter referred to as loop 2) formed between the policy decision unit 105, policy effect prediction unit 107, and model evaluation unit 106 is formed offline, and the parameters of the model that determines the intervention measure are adjusted using the policy effect prediction unit 107 as a virtual intervention target.

[0143] Specifically, the attributes Y1, Y2, ..., Y of the user 113 m The policy decision unit 105 decides an intervention policy Z according to the selected intervention policy and inputs the intervention policy Z to the policy effect prediction unit 107. The policy effect prediction unit 107 calculates the predicted value U pred The model evaluation unit 106 outputs the predicted value U of the intervention effect. pred and the control target (command value) U ref The policy decision unit 105 receives the evaluation result R and adjusts the model parameters. Then, the attributes Y1, Y2, ..., Y mSelect a hypothetical intervention and perform it in the same way. The predicted value of the intervention effect, U pred and the control target (command value) U ref This series of processes is repeated until the error falls below a pre-specified threshold S1 (e.g., a control error of 20%). In other words, the model parameter adjustment is repeated until the difference between the predicted value of the intervention effect and the command value of the intervention effect becomes smaller than a predetermined threshold.

[0144] The use of this loop 2 solves the problem of the policy decision unit 105 having difficulty tracking the physical system. The reasons why loop 2 is thought to be effective include that it can be executed faster than loop 1, the required number of trials can be ensured, and it can try combinations (counterfactuals) of users 113 and intervention policies that did not exist in the log data. For example, although there is no log data for a new user 113, the predicted value of the intervention policy can be calculated using the log data of other users 113.

[0145] In addition, for example, depending on the characteristics (fatigue, etc.) of the user 113 and changes in distribution (people replacement), the attributes Y1, Y2, ..., Y m By changing the value of the parameter, the policy decision unit 105 can be made more robust against the passage of time in the physical system.

[0146] That is, the attribute data of the user 113 used when calculating the predicted value of the intervention effect may be selected according to expected future changes in the user 113 in the intervention area. For example, when the attributes Y1, Y2, ..., Ym of the user 113 include dynamic information such as fatigue level, working hours, and the number of times each intervention measure is implemented, which change over time or with the implementation of measures, data in which these values ​​are changed may be generated as training data and used to implement the virtual intervention. Also, when the attributes Y1, Y2, ..., Ym of the user 113 include static information such as age, gender, height, exercise experience, years of work, and employment status, which change with future personnel changes, data in which these values ​​are changed may be generated as training data and used to implement the virtual intervention.

[0147] Here, as a second function of the model evaluation unit 106, the model evaluation unit 106 calculates the predicted value U pred and the measured value U act and calculates an evaluation result (prediction error) regarding the prediction performance of the policy effect prediction unit 107. Then, for example, if the prediction error becomes larger than a pre-specified threshold S2 (for example, a prediction error of 20%), the model evaluation unit 106 transmits an update flag to the model update unit 108. Upon receiving the update flag, the model update unit 108 updates the policy effect prediction model using the latest log data.

[0148] In this way, a policy decision unit 105 suited to the target physical system is constructed, but it is considered difficult to maintain the performance of a model once constructed semi-permanently. In other words, as time passes, a discrepancy occurs between the physical system and the policy effect prediction model, and as a result, the performance of the policy decision model, whose parameters are adjusted based on the policy effect prediction model, deteriorates. For this reason, it is necessary to periodically update the model while continuing to monitor the evaluation result R (the difference between the actual measured value of the intervention effect and the command value) output by the model evaluation unit 106.

[0149] Specifically, first, the model evaluation unit 106 evaluates the actual value U of the intervention effect using the first function. actand the command value U ref If the difference exceeds the threshold S1, the model evaluation unit 106 performs the second function of calculating the actual value U of the intervention effect. act and the predicted value U calculated by the policy effect prediction unit 107 pred and calculates the prediction error of the policy effect prediction unit 107. If the prediction error exceeds the threshold S2, the model evaluation unit 106 sends an update flag to the model update unit 108 and updates the policy effect prediction model. If the prediction error is below the threshold S2, the model update for policy effect prediction is skipped. Next, under the latest policy effect prediction unit 107, the predicted value U of the intervention effect is calculated. pred and the control target (command value) U ref Loop 2 is repeated until the error falls below threshold S1, and the parameters of the policy decision model are adjusted.

[0150] Here, the host management device 1 may be equipped with a simulator 109 that simulates behavioral changes of a user 113 in a physical system. The simulator 109 is realized, for example, by an agent-based model in which autonomously acting actors (agents) are given behavioral rules and their interactions represent complex phenomena. The agent's behavioral rules for intervention measures are constructed based on log data obtained from the physical system. Before applying the intervention measure decision model, for which parameter adjustment has been completed, to the physical system, the host management device 1 may execute the intervention measure on the simulator 109 to verify in advance whether the parameter adjustment has been performed appropriately.

[0151] The policy effect prediction unit 107 and the simulator 109 have similar roles in that they aim to predict (simulate) the response of the user 113 in the digital space before actually executing an intervention measure for the user 113 of the physical system, and improve the fit of the policy decision unit 105. However, there is no distinction here, as the policy effect prediction unit 107 is intended to focus on predicting the intervention effect for various combinations of user 113 attributes and intervention measures, while the simulator 109 is intended to focus on simulating situations such as the passage of time in the physical system and interactions between users 113 according to one scenario. are.

[0152] Furthermore, because this is information processing targeting uncertain human behavior, care must be taken to ensure that model update decisions are not overly influenced by momentary deviations from the target (when the number of log data samples referenced is very small, or the intervention period is very short). For this reason, it is desirable to carry out the model update in a situation where there is a certain number of log data samples, or when the intervention period has been secured to a certain extent.

[0153] In the case of the picking system 100 of the embodiment, the order list is usually acquired in batch units determined based on shipping time and transport truck constraints, and therefore, for example, various update processes may be performed within this batch unit.

[0154] When the operation of the picking system 100 has just started, or when there have been major changes to the workers (users 113) or the contents of the picking work, it may be difficult to keep the prediction error of the policy effect prediction model within the threshold S2 using only the log data obtained from the target picking system 100. In such cases, log data may be supplemented from another logistics warehouse (external system 3) operated by the same business operator and used to build the policy effect prediction model.

[0155] FIG. 10 is a flowchart for explaining an example of the operation of the upper management device 1 in the retrieval work of an item in the picking system 100 according to the embodiment.

[0156] The operation of this flowchart is realized by the processor 11 of the upper management device 1 reading and executing a program stored in the internal memory of the processor 11, the ROM 12, or the RAM 13.

[0157] In step ST101, during the retrieval work of an item, the host management device 1 (processor 11) acquires a control target and an order list from the external system 3 via the communication interface 15. For example, the control target is a command value, which is a control target value for the intervention effect of an intervention measure to be implemented for the user 113. As described above, the command value is specified individually for each user 113, or one value is specified for the entire physical system. The order list is composed of, for example, orders that associate items to be picked with delivery destinations of the items. Here, the order list is assumed to be a batch of orders composed of multiple orders. The orders may also be order slips or the like. The host management device 1 may acquire the control target and the order list from an external storage device via an interface for the external storage device, or may accept the control target and the order list entered by an operator using an operation unit or the like.

[0158] In step ST102, the processor 11 creates picking operation data. The processor 11 functioning as the operation data creation unit 101 creates the picking operation data based on the order list and the inventory list through the above-described processing.

[0159] In step ST103, the processor 11 determines an intervention measure. The processor 11 functioning as the measure determination unit 105 determines the intervention measure through the above-described process. For example, the processor 11 determines an intervention measure required to cause the intervention target user 113 to undergo a target behavioral change.

[0160] In step ST104, the processor 11 assigns picking work to each picking station P. The processor 11 functioning as the data reforming unit 102 reforms the picking data by the above-described process. Furthermore, the processor 11 functioning as the processing capacity prediction unit 103 predicts the processing capacity of each user 113 based on the reformed picking work data by the above-described process. Furthermore, the processor 11 functioning as the work planner 104 assigns the optimal allocation work to each picking station P while satisfying various constraints as picking work data.

[0161] In step ST105, the processor 11 executes an intervention measure. The processor 11 functioning as the measure decision unit 105 executes the intervention measure decided by the above-described process.

[0162] In step ST106, the processor 11 calls the AGV shelf 8. The processor 11 operating as the work planner 104 controls the AGV 7 based on the picking work data, and transports the predetermined AGV shelf 8 to the predetermined picking station P.

[0163] In step ST107, the user 113 picks an item. When the processor 11 calls the AGV shelf 8 to the picking station P, the processor 11 causes the user 113 present at the picking station P to perform the picking work corresponding to the AGV shelf 8. Furthermore, while the user 113 is picking the item, the processor 11 may determine the start time and end time of the picking work based on images captured by a camera installed at the picking station P.

[0164] In step ST108, processor 11 generates a work log. As described above, processor 11 generates a work log based on work completion information received from user 113 via user interface 112. For example, processor 11 creates a work log indicating the work content performed each time picking work (sorting work) of a specified item is completed, and stores (updates) the created work log in database 121.

[0165] In step ST109, processor 11 determines whether or not there is any unprocessed allocation work remaining at each picking station P. If there is any unprocessed work remaining, the process proceeds to step ST110. On the other hand, if there is no unprocessed allocation work remaining, the process ends.

[0166] In step ST110, the processor 11 calculates the deviation of the intervention effect. The processor 11 compares the work logs of the user 113 before and after the intervention measure is executed, and calculates the actual measured value of the intervention effect. The processor 11 functioning as the model evaluation unit 106 calculates the difference (deviation) between the actual measured value of the intervention effect and the command value. Furthermore, the processor 11 calculates the prediction error of the intervention effect. The processor 11 functioning as the measure effect prediction unit 107 calculates the predicted value of the intervention effect for the user 113 based on the information (work log) of the user 113 who is the intervention target and information about the executed intervention measure. The processor 11 functioning as the model evaluation unit 106 calculates the prediction error of the intervention effect from the actual measured value and predicted value of the intervention effect.

[0167] In step ST111, the processor 11 determines whether the deviation is within the expected range. The processor 11 determines whether the deviation calculated in step ST110 is within a predetermined threshold. If it is determined that the deviation is within the threshold, the process returns to step ST104. On the other hand, if it is determined that the deviation exceeds the threshold, the process proceeds to step ST112.

[0168] In step ST112, the processor 11 determines whether the prediction error is within the expected range. The processor 11 determines whether the prediction error calculated in step ST110 is within a predetermined threshold. If it is determined that the deviation exceeds the threshold, the process proceeds to step ST113. On the other hand, if it is determined that the deviation is within the threshold, the process proceeds to step ST114.

[0169] In step ST113, the processor 11 updates the policy effect prediction model. The processor 11, functioning as the model update unit 108, updates the policy effect prediction model as described above. That is, the processor 11 updates the prediction model that performs the function h described above using the shaped work logs as training data. The work logs used to update the model are appropriately selected so as to contribute to reducing the prediction error of the updated prediction model below a predetermined threshold. For example, the processor 11 uses work logs collected after the previous model update. The processor 11 may also extract data at a predetermined rate from work logs before and after the previous model update as training data. Furthermore, the processor 11 may add log data obtained from other logistics warehouses (external systems 3) to the training data.

[0170] In step ST114, the processor 11 updates the policy decision-making model. The processor 11 functioning as the model update unit 108 adjusts the parameters of the policy decision-making model as described above, and updates the policy decision-making model. The processor 11, under the control of the latest policy effect prediction unit 107, repeats virtual intervention in the policy effect prediction unit 107 until the deviation between the predicted value of the intervention effect and the control target (command value) falls below a predetermined threshold, and adjusts the parameters of the policy decision-making model. That is, the processing described as loop 2 above is performed. Of course, the processing described as loop 2 may be performed as processing in parallel with each step. At this time, the processor 11 functioning as the simulator 109 may verify whether the parameters of the policy decision-making model have been appropriately adjusted, as described above.

[0171] (Effects of the embodiment) According to the embodiment described above, even if human characteristics change from moment to moment, it is possible to update the parameters of the model that determines the intervention measures so that the target intervention effect can be achieved. This makes it possible to stably achieve infrastructure efficiency improvement through behavioral modification.

[0172] (Variation) The above-described embodiment can be applied to systems other than the picking system 100 in a logistics warehouse. For example, the embodiment can also be applied to guiding people at train stations and surrounding commercial facilities. For example, with the spread of COVID-19, the number of people using public transportation and commercial facilities has decreased, and it has become an issue to promote the use of transportation and commercial facilities while avoiding congestion, and to revitalize the city.

[0173] For example, if the user 113 in the above embodiment is a user of a means of transportation or a commercial facility, the user interface 112 through which the user 113 receives the intervention measures may be replaced with a smartphone or digital signage. The intervention results may be input by the user 113 via a smartphone application, or detected from the smartphone's location information or an image sensor installed in the facility.

[0174] Possible intervention measures to be implemented for the user 113 include, for example, delivering congestion information on public transport and commercial facilities, delivering recommended store information, delivering store coupons, changing public transport fares depending on the time of day and route (dynamic pricing), etc. These intervention measures can encourage the user 113 to change their destination or travel time, or to take detours.

[0175] The intervention effect may be, for example, an increase in the amount of consumption by the user 113 when the intervention measure is implemented, with the case where the intervention measure is not implemented as a reference. Alternatively, it may be a binary value, such as whether the amount of consumption by the user 113 increases when the intervention measure is implemented, or whether the user changes their destination or travel time or makes a detour, with the case where the intervention measure is not implemented as a reference.

[0176] (Other embodiments) The program according to this embodiment may be transferred in a state where it is stored in an electronic device (computer) serving as the host management device 1, or may be transferred in a state where it is not stored in an electronic device. In the latter case, the program may be transferred via a network, or may be transferred in a state where it is stored in a storage medium. The storage medium is a non-transitory, tangible medium. The storage medium is a medium that can be read by a computer serving as the host management device 1 (a computer-readable medium). The storage medium may be in any form, such as an optical disk (e.g., a CD-ROM), a magnetic disk, or a semiconductor memory (e.g., a memory card), as long as it is capable of storing a program and is computer-readable.

[0177] Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be embodied in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, and are also included in the scope of the invention and its equivalents as defined in the claims. [Explanation of symbols]

[0178] 100...Picking system 1...Upper management device 11...Processor 101...Work data creation department 102...Data Formatting Department 103...Processing capacity prediction unit 104...Work Planning Department 105…Policy decision department 106...Model Evaluation Section 107...Policy Effect Forecasting Department 108...Model Update Section 109...Simulator 12...ROM 121...Database 13...RAM 14...Auxiliary storage device 15...Communication interface 2. Network 3. External Systems 7...AGV 70...Tire 71...Processor 72...ROM 73...RAM 74...Auxiliary storage device 75...Communication interface 76...Drive unit 77...Sensor 78...Battery 79…Charging mechanism 8...AGV shelf 10...Tray 112...User Interface 113...User

Claims

1. An information processing method executed by a processor of an information processing device for determining an intervention measure to encourage a user's behavioral change, comprising: Acquiring attribute data of the user, a command value of an intervention effect, and an actual measurement value of the intervention effect obtained from log information of the execution of the intervention measure for the user; Calculating the difference between the actual measurement value of the intervention effect and the command value; adjusting parameters of a model that determines the intervention strategy based on the difference; Calculating a predicted value of an intervention effect based on the attribute data of the user and the information on the intervention measure; Repeating parameter adjustment of the model until a difference between the predicted value of the intervention effect and the command value of the intervention effect becomes smaller than a predetermined threshold; An information processing method comprising:

2. Calculating the predicted value of the intervention effect includes calculating the predicted value of the intervention effect using a prediction model trained with the user attribute data and the data related to the intervention measure as explanatory variables and the intervention effect as a response variable. The information processing method according to claim 1 .

3. updating the prediction model when a difference between the actual measured value of the intervention effect and the predicted value of the intervention effect calculated by the prediction model exceeds a predetermined threshold. The information processing method according to claim 2 .

4. The user attribute data used when calculating the predicted value of the intervention effect is selected according to a change in users expected in the future in the intervention destination area.

3. The information processing method according to claim 1.

5. The predictive model is constructed by counterfactual machine learning. The information processing method according to claim 2 .

6. Executing the intervention measures output by the model for determining the intervention measures in a simulator that simulates behavioral changes of users in an intervention destination area; determining whether parameter adjustment of a model that determines the intervention measure worked properly based on the intervention effect obtained as a result of execution in the simulator; and The information processing method according to claim 1 , further comprising:

7. The simulator is constructed using an agent-based model. The information processing method according to claim 6.

8. The intervention effect is data regarding the amount of reduction in the user's work time or whether the user's work time has been reduced. The information processing method according to claim 1 .

9. The user's attribute data includes at least one of the user's ID, age, sex, height, exercise experience, years of employment, occupation, employment type, fatigue level, personality, hobbies and preferences, or elapsed time.

3. The information processing method according to claim 1.

10. The data regarding the intervention measure includes at least one of a management ID, an implementation content, an implementation timing, or an embodiment of the intervention measure.

3. The information processing method according to claim 1.

11. creating picking operation data relating to a picking operation in which the user picks an item based on an order list acquired from an external system; assigning a picking task to the user based on the picking task data and the intervention measure; The log information is data regarding work results for the picking work assigned to the user. The information processing method according to claim 1 .

12. An information processing device for determining an intervention measure to encourage a user's behavioral change, a communication interface for transmitting and receiving data to and from an external system and a user interface; a storage unit that stores a program; a processor connected to the communication interface and the storage unit; The processor uses a program stored in the storage unit to via the communication interface, attribute data of the user, a command value of an intervention effect, and an actual measurement value of the intervention effect obtained from log information of the execution of the intervention measure for the user; Calculating the difference between the actual measured value of the intervention effect and the command value; adjusting parameters of a model that determines the intervention measure based on the difference; Calculating a predicted value of an intervention effect based on the user attribute data and the information on the intervention measure; repeating parameter adjustment of the model until a difference between the predicted value of the intervention effect and the command value of the intervention effect becomes smaller than a predetermined threshold; An information processing device configured as follows.

13. An information processing program comprising instructions to be executed by a processor of an information processing device, the instructions determining an intervention measure to encourage a user's behavioral change, Acquiring attribute data of the user, a command value of an intervention effect, and an actual measurement value of the intervention effect obtained from log information of the execution of the intervention measure for the user; Calculating the difference between the actual measurement value of the intervention effect and the command value; adjusting parameters of a model that determines the intervention strategy based on the difference; Calculating a predicted value of an intervention effect based on the attribute data of the user and the information on the intervention measure; Repeating parameter adjustment of the model until a difference between the predicted value of the intervention effect and the command value of the intervention effect becomes smaller than a predetermined threshold; An information processing program comprising:

Images