Article list division device, article list division method, and article list division program

Abstract

To provide an article list division device, an article list division method, and an article list division program that divide articles by considering a relationship between the articles so that an appropriate box can be selected.SOLUTION: An article division list generation unit 13 generates, on the basis of an evaluation value output from an output layer 23 of a neural network operation unit 12, an article division list. A neural network calculation unit 12 can output an evaluation value for an article simply by performing learning based on a past packaging result. Therefore, the article list division device 100 can easily generate a division list with reduced calculation cost. The division list generation unit 13 generates a division list based on an output of probability that a plurality of articles are placed in a same box and probability that they are placed in different boxes. The division list generation unit 13 can generate a division list of articles by taking into consideration the relationship between the articles and whether the articles are placed in a same box or different boxes.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an item list division device, an item list division method, and an item list division program. [Background technology]

[0002] Conventionally, a device such as that described in Patent Document 1 is known. This device selects combinations of cargo and containers based on data such as the volume of the cargo. This device selects all combinations of containers whose total container capacity is equal to or greater than the total volume of the cargo. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2016-88630 Summary of the Invention [Problem to be solved by the invention]

[0004] In logistics operations, multiple items from an item list are sorted and packed into multiple selected boxes. However, sorting items into appropriate boxes while taking into account characteristics such as the shape and weight of each item is a task that requires skill, and support is required, such as creating a division list that allows for appropriate item sorting based on the item list and presenting the division list to the worker. On the other hand, when multiple items are divided, if attention is paid to the relationships between the parts, parts that are not necessarily appropriate combinations may end up in the same box. Therefore, it is necessary to divide the items by taking into account the relationships between the parts so that appropriate boxes can be selected.

[0005] An object of the present invention is to provide an item list division device, an item list division method, and an item list division program that divide items taking into account the relationships between the items so that appropriate boxes can be selected. [Means for solving the problem]

[0006] An item list division device according to one embodiment of the present invention is an item list division device that divides an item list containing information about items to be packed into multiple boxes based on pre-learned learning information, and generates a division list containing information about the items for each box.The control unit of the item list division device has a memory unit that stores the learning information, an input information receiving unit that creates input information about the item list, a neural network calculation unit that inputs the input information to an input layer and calculates evaluation values ​​for the items by neural network calculation based on learning information that has been pre-learned based on packing performance, and a division list generation unit that generates a division list based on the evaluation values ​​output from the output layer of the neural network calculation unit.The division list generation unit creates a division list based on the output of the probability that multiple items will be placed in the same box and the probability that they will be placed in different boxes.

[0007] In the item list division device, the neural network calculation unit inputs input information to an input layer and calculates evaluation values ​​for items through neural network calculation based on learning information previously learned based on packing records. The division list generation unit can then generate a division list based on evaluation values ​​output from the output layer of the neural network calculation unit. Here, the learning information is previously learned based on packing records. In other words, the neural network calculation unit can output evaluation values ​​for items simply by learning based on past packing records, without considering complex conditions such as the shape and weight of the items. Therefore, the item list division device can easily generate a division list with reduced calculation costs. Furthermore, the division list generation unit creates a division list based on the output of the probability that multiple items will be placed in the same box and the probability that they will be placed in different boxes. In this case, the division list generation unit can generate a division list for items by taking into account the relationships between the items and whether the items will be placed in the same box or different boxes. As described above, items can be divided by taking into account the relationships between the items so that appropriate boxes can be selected.

[0008] The neural network operation includes a self-attention structure, and during learning, the input information input to the input layer is divided into items, and the relationships between the likelihood of items being in the same box and the likelihood of items being in different boxes are matched. In this case, the relationships between items can be learned, which can improve the matching rate of the divided lists.

[0009] The division list generation unit may prioritize the pattern with a higher probability when a combination of the relationships between the first item, the second item, and the third item includes a pattern in which the first item and the second item are placed in the same box and a pattern in which the first item and the second item are placed in different boxes. When multiple patterns are possible for a combination of three or more items, a pattern based on probability can be adopted.

[0010] An item list division method according to one aspect of the present invention is a method for dividing an item list containing information about items to be packed into multiple boxes based on pre-learned learning information, and generating a division list containing information about the items for each box.The method includes a storage step for storing the learning information, an input information receiving step for creating input information about the item list, a neural network calculation step for inputting the input information into an input layer and calculating evaluation values ​​for the items by neural network calculation based on learning information pre-learned based on packing performance, and a division list generation step for generating a division list based on the evaluation values ​​output from the output layer in the neural network calculation step.In the division list generation step, a division list is created based on the output of the probability that multiple items will be placed in the same box and the probability that they will be placed in different boxes.

[0011] An item list division program according to one embodiment of the present invention is an item list division program that divides an item list containing information about items to be packed into multiple boxes based on pre-learned learning information, and generates a division list containing information about the items for each box.The program has a computer system execute the following steps: a storage step for storing the learning information; an input information receiving step for creating input information about the item list; a neural network calculation step for inputting the input information into an input layer and calculating evaluation values ​​for the items by neural network calculation based on learning information pre-learned based on packing performance; and a division list generation step for generating a division list based on the evaluation values ​​output from the output layer in the neural network calculation step.In the division list generation step, a division list is created based on the output of the probability that multiple items will be placed in the same box and the probability that they will be placed in different boxes.

[0012] According to these item list division methods and item list division programs, it is possible to obtain the same effects and advantages as those of the above-mentioned item list division device. [Effects of the Invention]

[0013] According to the present invention, it is possible to provide an item list division device, an item list division method, and an item list division program that divide items taking into account the relationships between the items so that an appropriate box can be selected. [Brief explanation of the drawings]

[0014] [Figure 1] 1 is a schematic configuration diagram showing an item list division device according to an embodiment of the present invention; [Figure 2] FIG. 1 is a schematic diagram showing an image of work at a work site. [Figure 3] FIG. 10 is a schematic diagram showing a processing image when the item list division device generates divided lists using a neural network. [Figure 4] FIG. 10 is a schematic diagram illustrating a process image when a neural network calculation unit of the item list division device performs learning using past packing results as training data. [Figure 5] FIG. 2 is a schematic diagram showing a specific example of processing content of an item list division device. [Figure 6] FIG. 10 is a diagram illustrating processing performed by an input information receiving unit when converting a parts list. [Figure 7] FIG. 10 is a schematic diagram showing the contents of pre-processing in the intermediate layer. [Figure 8] This is an image of the Self Attention structure. [Figure 9] This is a table showing the probability that each combination of items will end up in the same box or in different boxes. [Figure 10] FIG. 10 is a diagram illustrating processing details of a division list generation unit. [Figure 11] FIG. 10 is a diagram for explaining learning content. [Figure 12] 10 is a flowchart showing the contents of a control process performed by a control unit of the item list division device. DETAILED DESCRIPTION OF THE INVENTION

[0015] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the drawings, the same or equivalent elements are designated by the same reference numerals, and redundant description will be omitted.

[0016] FIG. 1 is a schematic diagram showing the configuration of an item list division device 100 according to one embodiment of the present invention. The item list division device 100 is a device that divides an item list containing information about items to be packed into multiple boxes, and generates a division list containing information about the items for each box. In this embodiment, parts are exemplified as items to be packed, and shipping boxes are exemplified as boxes. However, the types of items and boxes are not particularly limited.

[0017] Specifically, as shown in FIG. 2, at a logistics site such as a warehouse, multiple parts are collected on a pallet PT or the like and transported to a work site where they are sorted and packed into shipping boxes. Meanwhile, division lists BL1 and BL2 are generated based on a parts list PL of the collected parts. Multiple appropriate shipping boxes BX1 and BX2 are selected from a large number of types of shipping boxes. At the work site, workers pack the collected multiple parts into shipping boxes BX1 and BX2 based on the division lists BL1 and BL2. Thus, workers can appropriately divide multiple items based on the division lists. Note that the type of box to select for the parts list is determined automatically by another method. The position and order in which the parts should be packed into the shipping boxes are also determined automatically by another method.

[0018] As shown in FIG. 1, the item list division device 100 includes an information input unit 2, an information output unit 3, and a control unit 10. The item list division device 100 is a device that divides a parts list based on previously learned learning information to generate divided lists. Note that FIGS. 3 and 4 may be referenced to explain each component of the item list division device 100. FIG. 3 is a schematic diagram showing a processing image when the item list division device 100 divides an item list using a neural network. FIG. 4 is a schematic diagram showing a processing image when the neural network calculation unit 12 of the item list division device 100 performs learning using past packing results as training data.

[0019] The information input unit 2 is an interface for inputting various types of information. For example, the information input unit 2 is configured with a keyboard, a mouse, a touch panel, etc. Alternatively, the information input unit 2 may input information via a storage medium or communication. The information output unit 3 is an interface for outputting various types of information. For example, the information output unit 3 may output information via a monitor, a speaker, etc. Alternatively, the information output unit 3 may output information via a storage medium or communication.

[0020] The control unit 10 is composed of a CPU, RAM, ROM, etc. The control unit 10 has an input information receiving unit 11, a neural network calculation unit 12, a division list generation unit 13, and a memory unit 14. The control unit 10 can be configured as one or more processors that operate according to a program, one or more dedicated hardware circuits such as ASICs, or a circuit including a combination thereof. The processor has a CPU and memories such as RAM and ROM. Various programs for processing information are stored in the memory, and the CPU reads out the various programs and performs calculations.

[0021] The input information receiving unit 11 creates input information related to a parts list. The input information receiving unit 11 acquires information input by the information input unit 2 and uses the acquired information to create input information. The input information receiving unit 11 acquires identification information associated with each of the multiple parts in the parts list, information related to the number of parts, etc. The input information is information related to the parts that has been adjusted to facilitate neural network calculations. As shown in FIG. 3, the input information receiving unit 11 acquires a parts list of the parts to be packed, i.e., a list indicating which parts and how many of each part are to be packed, in the pre-packing stage. Furthermore, as shown in FIG. 4, during learning, the input information receiving unit 11 acquires a parts list of the parts that have actually been packed and information regarding the relationship between whether each part in the parts list will be packed in the same box or in different boxes.

[0022] The neural network calculation unit 12 calculates an evaluation value for an item by neural network calculation based on learning information previously learned based on packaging performance. The neural network calculation unit 12 acquires learning information by learning using training data based on past packaging performance (see Figure 4). Furthermore, the neural network calculation unit 12 uses the learning information to calculate an evaluation value for a single item when it is packed into each of multiple candidate shipping boxes at an actual work site.

[0023] The neural network calculation unit 12 includes a learning unit 16. The learning unit 16 generates a machine learning model as learning information. In the learning stage (see FIG. 4), the learning unit 16 generates the best neural network estimated to have the highest prediction accuracy as a model. The learning unit 16 reads training data (samples of past packaging results) from the memory unit 14 or the like, and performs machine learning by sequentially inputting each sample into the neural network to be trained. Note that such a neural network can be configured, for example, by a calculation unit and a memory unit that mimic a neuron model. For example, the learning unit 16 may generate a model by performing deep learning using a multilayer neural network. The type of machine learning is not limited to deep learning, and the learning unit 16 may generate a model using other methods. Note that the specific configuration and processing content of the neural network calculation unit 12 will be described later.

[0024] The division list generation unit 13 generates a division list based on the evaluation values ​​output from the output layer of the neural network operation unit 12. The division list generation unit 13 creates a division list based on the output of the probability that multiple items will be placed in the same box and the probability that they will be placed in different boxes. The probability that certain items will be placed in the same box and the probability that they will not be placed in different boxes are output as evaluation values. Therefore, the division list generation unit 13 generates a division list based on these probabilities. Note that the specific processing content of the division list generation unit 13 will be described later.

[0025] The storage unit 14 is a part that stores various information in the item list division device 100. The storage unit 14 stores at least learning information (neural network model). The storage unit 14 also stores an item list division program P1 that executes an item list division method described below.

[0026] Next, a specific example of the processing of the neural network operation unit 12 will be described with reference to Figures 3 and 4. First, with reference to Figure 4, a case will be described in which the neural network operation unit 12 of the item list division device 100 performs learning using past packing records as training data. Here, data is used that shows that, as the packing record, a parts list PL of "8 parts A," "1 part B," "1 part C," "2 parts D," and "3 parts E" has been divided into a division list BL1 of "2 parts D" and a division list BL2 of "8 parts A," "1 part B," "1 part C," and "3 parts E."

[0027] As shown in FIG. 4, the neural network calculation unit 12 extracts the packed parts list PL from the packing record data and extracts information indicating the relationship between the items in the parts list PL and whether they will be packed in the same box or different boxes. The learning unit 16 (see FIG. 1) of the neural network calculation unit 12 appropriately updates the neural network model M so that when the parts list PL is input to the model M, the model M outputs the probability that the parts will and will not be packed in the same box. The neural network calculation unit 12 performs learning using training data based on multiple types of parts lists and numerous packing records related to division lists for the parts lists. The neural network calculation unit 12 prepares probabilities such as those shown in FIG. 9 as prediction values ​​for the parts that will and will not be packed in the same box through repeated learning. Specific learning methods will be described later.

[0028] Next, referring to FIG. 3, a case where the item list division device 100 divides a parts list to generate division lists will be described. First, the item list division device 100 acquires the aforementioned parts list PL, which includes information on the parts to be packed. The neural network calculation unit 12 calculates an evaluation value for each part in the parts list PL using the neural network model M updated by learning shown in FIG. 4. As a result of learning, the neural network calculation unit 12 outputs an evaluation value that indicates that "two parts D" will be placed in the same box, and outputs an evaluation value that indicates that "eight parts A," "one part B," "one part C," and "three parts E" will be placed in the same box. Therefore, "two parts D" are assigned to shipping box BX1 corresponding to division list BL1, and "eight parts A," "one part B," "one part C," and "three parts E" are assigned to shipping box BX2 corresponding to division list BL2. The item list division device 100 outputs a message indicating that division lists BL1 and BL2 have been generated.

[0029] Next, the detailed configuration of the neural network operation unit 12 and the processing content thereof will be described with reference to Fig. 5. Fig. 5 is a schematic diagram showing a specific example of the processing content of the item list division device 100. First, the processing when the input information receiving unit 11 creates input information will be described. In this embodiment, the input information converts the input item list into a format that can be input to the neural network operation unit 12.

[0030] Specifically, as shown in FIG. 6, the input information receiving unit 11 acquires a parts list in a format in which the number of parts for each type is written. In this parts list, one line is allocated for each type of part. The input information receiving unit 11 expands each part into lines for the number of parts. Thus, if there are three instances of "part a," three lines of "part a" are allocated. Furthermore, the input information receiving unit 11 converts the part names (e.g., product numbers) stored in each line into IDs that identify the parts. Here, "part a" is converted to "part ID: 1," "part b" is converted to "part ID: 2," and "part c" is converted to "part ID: 3."

[0031] Returning to FIG. 5, the input information receiving unit 11 converts the information on the parts in the parts list to create input information shown in a table that stores IDs that identify the parts, and then inputs the ID information stored in each row to the neural network calculation unit 12.

[0032] The neural network calculation unit 12 has an input layer 21, an intermediate layer 22, and an output layer 23 as a neural network model. The input layer 21 is a layer for inputting parameters for neural network calculation. The intermediate layer 22 is a layer for performing a predetermined calculation on the input parameters. The output layer 23 is a layer for outputting an evaluation value. Each of the layers 21 to 23 has nodes corresponding to the number of input parameters. For example, the input layer 21 receives an input vector x=(x1, x2, ... x) having m parameters as components. m ) is output to the intermediate layer 22 as is. The intermediate layer 22 converts the total input into an output using the weights and activation functions of each layer, and passes the output to the output layer 23. The output layer 23 also converts the total input into an output using the weights and activation functions of each layer. This output is the output vector y=(y1, y2, ..., y n ) This output vector y indicates the evaluation value. During learning (see Figure 4), the correct output vector for one input vector x is expressed as d = (d1, d2, ..., d n ), the weights w of each layer are updated so that the output vector y approaches the correct answer d. This is the information processing known as learning.

[0033] The processing of the intermediate layer 22 will be described with reference to FIGS. 7 to 9. The intermediate layer 22 has a preprocessing layer (embedding layer) that performs preprocessing for calculations. The preprocessing layer converts each part ID into a feature vector for that part. If a feature vector for each part has been obtained through a separate process, that feature vector may be used, or the feature vector may be obtained through learning. For example, as shown in FIG. 7, an "N-dimensional feature vector" is prepared in advance for "part IDs = 1 to 6." Note that the number of dimensions of the feature vector is not particularly limited, but the higher the number of dimensions, the more detailed the distinction between each part can be. If the input part ID is "part ID = 1, 1, 4, 5, 6," the preprocessing layer of the intermediate layer 22 converts the part ID into a feature vector CV1. The preprocessing layer also assigns order information (positional encoding) to each part. This is necessary when learning the relationships between parts, even if the parts are the same, they must be represented as different parts. Therefore, by assigning order information to the feature vector, the preprocessing layer can treat multiple parts of the same type as different parts. As shown in FIG. 7, "order information" is prepared in advance. The pre-processing layer acquires an ordered feature vector CV2 by adding order information to the feature vector CV1 of "part ID=1,1,4,5,6." Note that, as the feature vector order information, for example, numbers assigned in the order of each part ID input to the middle layer can be used. There are no particular limitations on how the feature vector order information is added, as long as it can represent the same part as being different.

[0034] The intermediate layer 22 has a calculation layer for performing calculations. Here, the neural network calculations include a self-attention structure called "Transformer." The self-attention structure is a structure that calculates all the relationships between each input component and can learn data with high influence. Figure 8 is an image diagram of the self-attention structure. Component IDs are arranged horizontally, and identical component IDs are arranged vertically. The intersection of each component ID shows the magnitude of the influence of the combination. The greater the influence, the larger the circle. "Attention" refers to a structure that displays the relationship between each component. "Self" means that it is between itself (the same component).

[0035] The middle layer 22 acquires the probability of each part being placed in the same box or in different boxes, using an attention structure. The computational layer of the middle layer 22 acquires the relationship between the combination of "part ID = 1, 1, 4, 5, 6" from pre-trained learning results. Specifically, the middle layer 22 acquires a table such as that shown in FIG. 9. In the table shown in FIG. 9, part IDs are arranged horizontally and parts with the same part ID are arranged vertically. The intersecting columns of part IDs indicate the probability that the combination will be placed in the same box or in different boxes. In the table, the first region E1 indicates the probability that parts will be placed in the same box. The second region E2 indicates the probability that parts will be placed in different boxes. For example, the values ​​VL1a and VL1b both represent the combination of the part with "part ID = 4" and the part with "second part ID = 1." The value VL1a indicates the probability that the part with "part ID = 4" and the part with "second part ID = 1" will be placed in the same box. The value VL1b indicates the probability that the part with "part ID = 4" and the part with "second part ID = 1" are placed in different boxes. In this case, the value VL1a has a higher probability.

[0036] The output layer 23 outputs the probability that each part will be placed in the same box and the probability that each part will be placed in a different box, which are acquired by the intermediate layer 22, to the division list generation unit 13 as evaluation values.

[0037] The division list generation unit 13 creates a division list based on the output of the probability that multiple parts will be placed in the same box and the probability that they will be placed in different boxes. The division list generation unit 13 acquires the probability shown in FIG. 9 as an evaluation value for whether "part ID = 1, 1, 4, 5, 6" will be placed in the same box or different boxes. Based on the probability in the acquired table, the division list generation unit 13 determines whether each part will be placed in the same box or different boxes and divides the parts in the parts list. For example, for the combination of part "part ID = 4" and part "second part ID = 1," the value VL1a is higher, so the part "part ID = 4" and part "second part ID = 1" are divided so that they are placed in the same box. As shown in FIG. 10(a), the division list generation unit 13 determines the relationship between whether each part will be placed in the same box or different boxes. In Figure 10(a), parts 50A to 50D corresponding to "part ID = 1, 1, 4, 5" are in the same box. Part 50E with "part ID = 6" is in a different box from the other parts. Therefore, as shown in Figure 10(b), the division list generation unit 13 creates a division list BL1 for parts 50A to 50D and a division list BL2 for part 50E.

[0038] In the above description, division was performed based on the relationship between two parts. However, when considering the relationship between three or more parts, the relationship between the two parts may contradict. In response to this, when a combination of the relationships between a first part, a second part, and a third part results in a pattern in which the first part and the second part are placed in the same box and a pattern in which they are placed in different boxes, the division list generation unit 13 prioritizes the pattern with the higher probability. For example, consider parts 50A, 50B, and 50E in the relationship shown in FIG. 10(a). In FIG. 10(a), the relationship PB1 between part 50B and part 50E indicates a relationship in which they are placed in different boxes, but assume that they are a relationship in which they are placed in the same box. In this case, part 50A and part 50B are a relationship in which they are placed in the same box, and part 50B and part 50E are a relationship in which they are placed in the same box. Therefore, a pattern in which parts 50A, 50B, and 50C are placed in the same box is valid. However, a pattern in which parts 50A, 50B, and 50C are placed in different boxes is valid for the relationship PB2 between part 50A and part 50E. In this case, the division list generation unit 13 compares the probability of relationship PB1 with the probability of relationship PB2, and prioritizes the pattern with the higher probability. For example, if relationship PB1 is "29%" and relationship PB2 is "20%," the pattern in which parts 50A, 50B, and 50C are placed in the same box is prioritized.

[0039] Next, a method in which the neural network calculation unit 12 prepares probabilities such as those shown in FIG. 9 through advance learning will be described with reference to FIG. 11. For the purpose of explanation, the columns of the table in FIG. 11(a) are labeled "1A" to "10A" and "1B" to "10B." The columns "iA" and "iB" have the same combination of parts (i = 1 to 10). This combination of parts is referred to as "combination i." The column "iA" indicates the probability that parts in combination i will be placed in the same box. The column "iB" indicates the probability that parts in combination i will be placed in different boxes. A formula FM for calculating "LOSS" is set as shown in FIG. 11(b). In the formula, the "probability of combination 1 being in the same box" is the probability (0.9) shown in the column "1A," and the "probability of combination 1 being in different boxes" is the probability (0.5) shown in the column "1B." In other words, the "probability that combination i is in the same box" in the formula is the probability shown in the "iA" column, and the "probability that combination i is in a different box" is the probability shown in the "iB" column. When calculating a formula (FM) for a certain packing record for "part ID = 1,1,4,5,6", if parts of combination i are in the same box, then "S i,1 =1" "S i,2 =0" is assigned, and if the parts of combination i are in different boxes, "S i,1 =0" "S i,2 =1" is assigned.

[0040] Next, the procedure for the neural network calculation unit 12 to perform learning using the formula (FM) for "LOSS" described above will be described. First, in the initial stage of learning, the neural network calculation unit 12 sets provisionally set initial value probabilities for each of the columns "1A" to "10B." The neural network calculation unit 12 calculates "LOSS" using formula (FM) using the first packing record for "Part ID = 1, 1, 4, 5, 6." Here, if the packing record shows that the parts of combination i are packed in the same box, the neural network calculation unit 12 adjusts the probability of the parts being packed in the same box for that combination i to increase and the probability of the parts being packed in different boxes to decrease. If the packing record shows that the parts of combination i are packed in different boxes, the neural network calculation unit 12 adjusts the probability of the parts being packed in the same box for that combination i to decrease and the probability of the parts being packed in different boxes to increase. The neural network calculation unit 12 repeats the above calculation using a large number of packing records. When the LOSS converges to a certain value, the neural network calculation unit 12 saves the obtained table as a table to be used in the processing described with reference to FIG.

[0041] Next, an item list division method will be described with reference to Fig. 12. Fig. 12 is a flowchart showing the contents of the control process by the control unit 10 of the item list division device 100. The item list division method is a method for dividing a parts list to generate divided lists based on previously learned learning information. Note that the detailed processing contents in each step are the same as those described in the description of each component of the item list division device 100, and therefore will not be described again.

[0042] As shown in Fig. 12, a storage step S10 is executed to store the learning information in the storage unit 14. Note that the storage step S10 may be executed in advance before the process of Fig. 12 is executed. Here, as described in Fig. 4, a neural network model is generated or updated as the learning information.

[0043] Next, the input information accepting unit 11 executes an input information accepting step S20 in which it acquires information related to the parts list and creates input information based on the acquired information. Next, the neural network computing unit 12 inputs the input information created in the input information accepting step S20 to the input layer 21, and executes a neural network computing step S30 in which it calculates evaluation values ​​for the parts by neural network computation based on the learning information stored in the memory unit 14. Next, the division list generating unit 13 executes a division list generating step S40 in which it generates a division list based on the evaluation values ​​output from the output layer 23 in the neural network computing step S30. In the division list generating step S40, a division list is created based on the output of the probability that multiple items will be placed in the same box and the probability that they will be placed in different boxes. This completes the control process shown in FIG. 12. To generate a division list for another parts list, the process may be repeated from the input information accepting step S20.

[0044] Next, the operation and effect of the item list division device 100 according to this embodiment will be described.

[0045] In the item list division device 100, the neural network calculation unit 12 inputs input information to the input layer 21 and calculates evaluation values ​​for items through neural network calculations based on learning information previously learned based on packing records. The division list generation unit 13 can then generate a division list based on the evaluation values ​​output from the output layer 23 of the neural network calculation unit 12. Here, the learning information is previously learned based on packing records. In other words, the neural network calculation unit 12 can output evaluation values ​​for items simply by learning based on past packing records, without having to consider complex conditions such as the shape and weight of the items. Therefore, the item list division device 100 can easily generate a division list with reduced calculation costs. Furthermore, the division list generation unit 13 creates a division list based on the output of the probability that multiple items will be placed in the same box and the probability that they will be placed in different boxes. In this case, the division list generation unit 13 can generate a division list for items by taking into account the relationships between the items and whether the items will be placed in the same box or different boxes. From the above, items can be divided taking into account the relationships between the items so that an appropriate box can be selected.

[0046] The neural network operation includes a self-attention structure, and during learning, the input information input to the input layer 21 may be divided into items, and the relationships regarding the likelihood that items will be placed in the same box and the likelihood that items will be placed in different boxes may be associated with each other. In this case, the relationships between items can be learned, thereby improving the matching rate of the divided lists.

[0047] When a combination of the relationships between a first item, a second item, and a third item includes a pattern in which the first item and the second item are placed in the same box and a pattern in which the first item and the second item are placed in different boxes, the division list generation unit 13 may prioritize the pattern with the higher probability. When multiple patterns are possible for a combination of three or more items, a pattern based on probability can be adopted.

[0048] The item list division method of this embodiment is an item list division method that divides an item list containing information about items to be packed into multiple boxes based on pre-learned learning information, and generates a division list containing information about the items for each box.The method includes a storage step that stores the learning information, an input information reception step that creates input information about the item list, a neural network calculation step that inputs the input information into the input layer 21 and calculates evaluation values ​​for the items by neural network calculation based on learning information that has been pre-learned based on packing performance, and a division list generation step that generates a division list based on the evaluation values ​​output from the output layer 23 in the neural network calculation step.In the division list generation step, a division list is created based on the output of the probability that multiple items will be placed in the same box and the probability that they will be placed in different boxes.

[0049] The item list division program of this embodiment is an item list division program that divides an item list containing information about items to be packed into multiple boxes based on pre-learned learning information, and generates a division list containing information about the items for each box.The program has a computer system execute the following steps: a storage step for storing the learning information; an input information receiving step for creating input information about the item list; a neural network calculation step for inputting the input information into the input layer 21 and calculating evaluation values ​​for the items by neural network calculation based on learning information pre-learned based on packing performance; and a division list generation step for generating a division list based on the evaluation values ​​output from the output layer 23 in the neural network calculation step.In the division list generation step, a division list is created based on the output of the probability that multiple items will be placed in the same box and the probability that they will be placed in different boxes.

[0050] According to these item list division methods and item list division programs, it is possible to obtain the same effects and advantages as those of the above-mentioned item list division device.

[0051] The present invention is not limited to the above-described embodiments.

[0052] For example, the number of candidate boxes and the number of types of items are not limited to those shown in the above embodiment. [Explanation of symbols]

[0053] 10...control unit, 11...input information receiving unit, 12...neural network calculation unit, 13...division list generation unit, 14...storage unit, 100...item list division device.

Claims

1. An item list division device that divides an item list including information on items to be packed into a plurality of boxes based on previously learned learning information, and generates a division list including information on the items for each of the boxes, A control unit of the item list division device A storage unit that stores the learning information; an input information receiving unit for creating input information related to the item list; a neural network calculation unit that inputs the input information into an input layer and calculates an evaluation value for the item by neural network calculation based on the learning information that has been learned in advance based on packaging records; a division list generation unit that generates the division list based on the evaluation value output from an output layer of the neural network calculation unit, The division list generation unit is an item list division device that creates a division list based on an output of the probability that multiple items will be placed in the same box and the probability that multiple items will be placed in different boxes.

2. The neural network operation includes a self-attention structure; The item list division device according to claim 1, wherein during learning, the input information input to the input layer is divided for each item, and a relationship regarding the possibility that the items will be placed in the same box and the possibility that the items will be placed in different boxes is made to correspond to each other.

3. The item list division device of claim 2, wherein when a combination of the relationships between a first item, a second item, and a third item includes a pattern in which the first item and the second item are placed in the same box and a pattern in which the first item and the second item are placed in different boxes, the divided list generation unit prioritizes the pattern with the higher probability.

4. 1. An item list division method for dividing an item list including information on items to be packed into a plurality of boxes based on previously learned learning information, and generating a division list including information on the items for each of the boxes, comprising: a storage step of storing the learning information; an input information receiving step for creating input information related to the item list; a neural network calculation step of inputting the input information into an input layer and calculating an evaluation value for the item by neural network calculation based on the learning information that has been learned in advance based on packaging records; a division list generating step of generating the division list based on the evaluation value output from an output layer in the neural network operation step, In the division list generating step, a division list is generated based on an output of a probability that a plurality of items will be placed in the same box and a probability that a plurality of items will be placed in different boxes.

5. An item list division program for dividing an item list including information on items to be packed into a plurality of boxes based on previously learned learning information, and generating a division list including information on the items for each of the boxes, a storage step of storing the learning information; an input information receiving step for creating input information related to the item list; a neural network calculation step of inputting the input information into an input layer and calculating an evaluation value for the item by neural network calculation based on the learning information that has been learned in advance based on packaging records; a division list generating step of generating the division list based on the evaluation value output from the output layer in the neural network operation step, In the division list generation step, a division list is created based on the output of the probability that a plurality of items will be placed in the same box and the probability that the items will be placed in different boxes.

Images