JP7516966B2 - Box selection device, information processing method, and information processing program - Google Patents

Description

The present invention relates to a box selection device, an information processing method, and an information processing program.

Conventionally, a box selection device such as that described in Patent Document 1 is known. The box selection device selects a combination of cargo and container based on data such as the volume of the target cargo. This box selection device selects all combinations of containers in which the total capacity of the containers is equal to or greater than the total volume of the cargo.

JP 2016-88630 A

The box selection device described above selects all combinations of containers (boxes) that satisfy the conditions. However, when this method is adopted, the number of combinations increases sharply as the number of types of boxes increases, and it becomes necessary to calculate each time whether all items can be loaded for all combinations. In this case, there is a problem that adopting this method becomes difficult from the viewpoint of calculation costs, etc. Furthermore, such calculation costs also increase with an increase in the number of types of items. In other words, there is a problem that it becomes difficult to apply the box selection device described above to sites where there are many types of items or many options for boxes. To address this issue, when training the neural network calculation unit using information in which parts are listed in advance based on packaging records, it is required to perform training so as to improve the accuracy of box selection.

The object of the present invention is to provide a box selection device, an information processing method, and an information processing program that can train a neural network calculation unit to improve the accuracy of box selection.

A box selection device according to one aspect of the present invention is a box selection device that selects boxes for packaging items based on previously learned learning information, and the control unit of the box selection device includes a storage unit that stores the learning information and box information, an input information receiving unit that creates input information related to the items, an input information conversion unit that converts the input information, a neural network calculation unit that inputs the converted input information to an input layer and calculates an evaluation value related to the boxes by neural network calculation based on previously learned learning information based on packaging performance, and a selection unit that selects boxes based on the evaluation value output from the output layer of the neural network calculation unit, and the input information before conversion includes information regarding the number of items in a predetermined list of items, and the input information conversion unit includes an importance calculation unit that calculates the importance of each item in box selection in learning by neural network calculation based on the packaging performance, and an importance reflection processing unit that weights the number of items in the input information based on the importance.

In the box selection device, the neural network calculation unit inputs input information about the item to the input layer, and calculates an evaluation value about the box by neural network calculation based on the learning information. Then, the selection unit can select a box suitable for the item to be packed based on the evaluation value output from the output layer of the neural network calculation unit. Here, the learning information is learned in advance based on the packing record. That is, the neural network calculation unit can output an evaluation value for selecting a box suitable for the item to be packed by only learning based on the past packing record, without considering complex conditions such as the shape of the item and the environmental information of the packing destination. Here, the input information conversion unit can convert the input information input to the neural network calculation unit. The input information before conversion has information about the number of each item in a predetermined item list. For example, when the types of items in the item list are enormous, it may be difficult to obtain the packing record for every combination of items. In response to this, the input information conversion unit has an importance calculation unit that calculates the importance of each item in the box selection in the learning by the neural network calculation based on the packing record. The importance calculation unit can express the importance of parts that have a large influence on the selection and parts that have a small influence on the selection when selecting a box. Furthermore, the input information conversion unit has an importance reflection processing unit that weights the number of each item in the input information based on the importance. This allows the importance reflection processing unit to convert the input information into a form that reflects the magnitude of the influence on the selection. This allows the neural network calculation unit to learn in a manner that grasps the parts that should be emphasized when selecting a box. As described above, the neural network calculation unit can be trained to improve the accuracy of box selection.

The importance may be set based on the Euclidean distance between the vector of each item and the vector of all items. The vector of a part that has a high importance in box selection due to its part-specific characteristics has a large Euclidean distance from the vector of all items. Therefore, by setting the importance based on the Euclidean distance, the importance calculation unit can appropriately calculate the importance.

The importance calculation unit may calculate the importance based on the deviation between the proportion of each item packed in each box and a probability distribution indicating the proportion of all items packed in each box. For example, the probability distribution of a part with small part-specific characteristics in box selection will be similar to the probability distribution of all items, and the deviation will be small. On the other hand, the probability distribution of a part with large part-specific characteristics in box selection will be a unique distribution, and will therefore deviate greatly from the probability distribution of all items. In this way, the importance calculation unit can appropriately calculate the importance based on the deviation of the probability distribution.

The information processing method according to one aspect of the present invention is an information processing method used in a box selection method having a storage step of storing previously learned learning information and box information, an input information receiving step of creating input information related to items, an input information conversion step of converting the input information, a neural network calculation step of inputting the converted input information into an input layer and calculating an evaluation value related to the box by neural network calculation based on the previously learned learning information based on the packing record, and a selection step of selecting a box based on the evaluation value output from the output layer in the neural network calculation step, in which the input information before conversion has information regarding the number of each item in a predetermined list of items, and has an importance calculation step of calculating the importance of each item in box selection in the learning by the neural network calculation based on the packing record, and an importance reflection processing step of weighting the number of each item in the input information based on the importance.

An information processing program according to one aspect of the present invention is an information processing program used in a box selection program that causes a computer system to execute a storage step of storing previously learned learning information and box information, an input information receiving step of creating input information related to items, an input information conversion step of converting the input information, a neural network calculation step of inputting the converted input information into an input layer and calculating an evaluation value related to the box by neural network calculation based on the previously learned learning information based on the packing record, and a selection step of selecting a box based on the evaluation value output from the output layer in the neural network calculation step, in which the input information before conversion has information regarding the number of each item in a predetermined list of items, and causes the computer system to execute an importance calculation step of calculating the importance of each item in box selection in the learning by the neural network calculation based on the packing record, and an importance reflection processing step of weighting the number of each item in the input information based on the importance.

These information processing methods and programs can achieve the same effects as the box selection device described above.

The present invention provides a box selection device, an information processing method, and an information processing program that can train a neural network calculation unit to improve the accuracy of box selection.

1 is a schematic configuration diagram showing a box selection device according to an embodiment of the present invention; 1 is a schematic diagram showing a processing image when the box selection device selects a shipping box using a neural network. FIG. 1 is a schematic diagram showing a processing image when a neural network calculation unit of a box selection device learns using past packing results as training data. FIG. 11 is a schematic diagram showing a specific example of the processing contents of the box selection device; FIG. 11 is a diagram for explaining the processing content of an importance calculation unit; FIG. 11 is a diagram for explaining the processing contents of an importance reflection processing unit; FIG. 10 is a flowchart showing the contents of a control process during learning by a control unit of the box selection device. 10 is a flowchart showing the contents of a control process performed by a control unit of the box selection device.

Embodiments of the present invention will be described in detail below with reference to the drawings. In the drawings, identical or equivalent elements are given the same reference numerals, and duplicate descriptions will be omitted.

Figure 1 is a schematic diagram showing a box selection device 100 according to one embodiment of the present invention. The box selection device 100 is a device that selects which box to pack multiple items into when packing them into boxes at a logistics site such as a warehouse. After a box is selected, the position and order in which the items are packed in the box are automatically determined by another method. In this embodiment, parts are exemplified as the items to be packed, and a shipping box is exemplified as the box. However, the types of items and boxes are not particularly limited.

As shown in FIG. 1, the box selection device 100 includes an information input unit 2, an information output unit 3, and a control unit 10. The box selection device 100 is a device that selects a shipping box in which to pack parts based on previously learned learning information. Note that in order to explain each component of the box selection device 100, FIGS. 2 and 3 may be referred to. FIG. 2 is a schematic diagram showing a processing image when the box selection device 100 selects a shipping box using a neural network. FIG. 3 is a schematic diagram showing a processing image when the neural network calculation unit 12 of the box selection device 100 learns using past packaging results as training data.

The information input unit 2 is an interface for inputting various information. For example, the information input unit 2 is composed of 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 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.

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 selection unit 13, a memory unit 14, and an input information conversion unit 31. 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 that includes a combination of these. 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.

The input information receiving unit 11 creates input information related to the parts. The input information receiving unit 11 acquires information input by the information input unit 2 and creates input information using the acquired information. The input information receiving unit 11 acquires identification information associated with each of the multiple parts, information related to the number of parts, and the like. The input information is information that has been adjusted to make it easier to perform neural network calculations on the information related to the parts. As shown in FIG. 2, the input information receiving unit 11 acquires a parts list of the parts to be packed, i.e., a list showing which parts are to be packed and how many of them, in the pre-packing stage. Also, as shown in FIG. 3, the input information receiving unit 11 acquires a parts list of the parts that have actually been packed and information on which shipping box was used for packing during learning. The specific contents of the identification information associated with the parts and the specific contents of the input information will be described later. In this embodiment, the input information created by the input information receiving unit 11 is converted by the input information conversion unit 31 and then input to the neural network calculation unit 12.

The neural network calculation unit 12 calculates an evaluation value for the shipping box by neural network calculation based on learning information previously learned based on packaging records. The neural network calculation unit 12 acquires learning information by learning using teacher data based on past packaging records (see Figure 3). In addition, the neural network calculation unit 12 uses the learning information to calculate an evaluation value for selecting which shipping box is suitable for the parts list to be packed at the actual work site (see Figure 2).

The neural network calculation unit 12 has a learning unit 16. The learning unit 16 generates a machine learning model as learning information. In the learning stage (see FIG. 3), the learning unit 16 generates the best neural network estimated to have the highest prediction accuracy as a model. The learning unit 16 reads out teacher data (samples of past packaging performance) from the memory unit 14 or the like, and executes machine learning while sequentially inputting each sample to the neural network to be trained. Note that such a neural network can be configured, for example, by a calculation unit or a memory unit that imitates a neuron model. For example, the learning unit 16 may generate a model by executing 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 contents of the neural network calculation unit 12 will be described later.

The selection unit 13 selects a shipping box based on the evaluation value output from the output layer of the neural network calculation unit 12. The selection unit 13 acquires the evaluation value and uses the evaluation value to select the most suitable shipping box for packaging the parts. The selection unit 13 selects the most suitable shipping box from a shipping box list (box information) stored in advance in the storage unit 14. The specific processing content of the selection unit 13 will be described later.

The memory unit 14 is a part that stores various information in the box selection device 100. The memory unit 14 stores at least learning information (neural network model) and box information. The box information is information about multiple transport boxes used at the work site, and is stored in the memory unit 14 in advance at a stage prior to the box selection work. The memory unit 14 stores the box information in the form of a transport box list (see Figure 2) that includes information such as the size and shape of each of the multiple transport boxes.

The input information conversion unit 31 converts the input information created by the input information reception unit 11. The input information conversion unit 31 has an importance calculation unit 32 and an importance reflection processing unit 33. The importance calculation unit 32 calculates the importance of each part in box selection in learning by neural network calculation based on the packaging performance. The importance reflection processing unit 33 weights the number of each part in the input information based on the importance.

Next, a specific example of the processing of the neural network calculation unit 12 will be described with reference to FIG. 2 and FIG. 3. First, with reference to FIG. 3, a case will be described in which the neural network calculation unit 12 of the box selection device 100 learns using past packing records as teacher data. Here, data is used that indicates that "transport box Z" was used in a parts list of "2 parts a," "1 part b," "2 parts c," and "1 part d" as the packing record. Note that the parts list in this case may be referred to as "parts list PL1." Also, the transport boxes that can be used for packing are selected from a transport box list in which six types of transport boxes, "transport boxes U, V, W, X, Y, and Z" (see FIG. 2), are registered. The transport box list may be referred to as "transport box list BL1."

As shown in FIG. 3, the neural network calculation unit 12 extracts the packed parts list PL1 from the packing performance data, and extracts that "shipping box Z" was selected as the packed shipping box for that parts list. The learning unit 16 (see FIG. 1) of the neural network calculation unit 12 appropriately updates the neural network model M so that when that parts list PL1 is extracted, an evaluation value is output that will select shipping box Z. The neural network calculation unit 12 learns using multiple types of parts lists and training data based on numerous packing performance records regarding the shipping boxes selected for that parts list.

The importance calculation unit 32 calculates the importance of each item in box selection for learning by neural network calculation based on the packaging performance. The importance calculation unit 32 calculates the importance of each part using information on the number of parts in the parts list included in the packaging performance and information on the selected shipping box. The extracted parts list PL1 is input to the neural network calculation unit 12 with the importance reflected by the importance reflection processing unit 33.

Next, referring to FIG. 2, a case where the box selection device 100 selects a transport box will be described. As shown in FIG. 2, the box selection device 100 acquires a transport box list BL1 in advance. When selecting a transport box, the box selection device 100 acquires the above-mentioned parts list PL1 as information indicating the parts to be packed. At this time, the input information related to the parts list PL1 is input to the neural network calculation unit 12 in a state where the importance is reflected by the importance reflection processing unit 33. The neural network calculation unit 12 calculates an evaluation value related to the transport box using the neural network model M updated by learning shown in FIG. 3. As a result of learning, the neural network calculation unit 12 outputs an evaluation value that makes the transport box Z most likely to be selected from the transport box list BL1. Therefore, the selection unit 13 selects the transport box Z based on the evaluation value. The box selection device 100 outputs a message that the transport box Z has been selected.

Next, referring to FIG. 4, the processing contents of the box selection device 100 and the detailed configuration and processing contents of the neural network calculation unit 12 will be described in more detail. FIG. 4 is a schematic diagram showing a specific example of the processing contents of the box selection device 100. First, the processing when the input information reception unit 11 creates input information will be described.

The input information before conversion has information regarding the number of each part in a predetermined parts list. The parts list is set in advance by extracting all parts that can be packed in a shipping box from the packing record. Therefore, the input information before conversion has an information table 60 having rows (or columns) corresponding to the number of types of parts included in the parts list, i.e., the number of types of parts to be packed. That is, if there are N types of parts, the information table 60 has N rows of storage sections 61. Note that there is only one storage section 61 in each row (i.e., one column). When the input information receiving unit 11 obtains the parts list to be selected for a shipping box, it inputs the number of parts related to each type into the information table 60. For example, in the example shown in FIG. 6B, a parts list of "2 parts b, 1 part d, 3 parts e, 2 parts f..." is obtained. Therefore, the storage section 61 in the second row of the information table 60 corresponding to part b stores a "2", the storage section 61 in the fourth row corresponding to part d stores a "1", the storage section 61 in the fifth row corresponding to part e stores a "3", and the storage section 61 in the sixth row corresponding to part f stores a "2". The storage sections 61 corresponding to 0 parts store a "0".

When there is an "unknown part x" whose details are not yet known, such as a new part, the input information receiving unit 11 may perform the following preprocessing. For example, the input information receiving unit 11 may add a row indicating the "unknown part" to the information table 60, and input the number of all unknown parts x in that row. The input information receiving unit 11 may discard the row corresponding to the "unknown part x".

ｐ_ｉｊ：各部品がそれぞれの輸送箱に梱包された割合
ｄ_ｉｊ：部品ｉが輸送箱ｊに梱包された回数
ｉ ：部品の種類を示す番号
ｊ ：輸送箱の種類を示す番号
ｎ ：輸送箱の種類の総数

ｓ_ｊ ：全部品がそれぞれの輸送箱に梱包された割合
ｄ_ｉｊ：部品ｉが輸送箱ｊに梱包された回数
ｉ ：部品の種類を示す番号
ｍ ：部品の種類の総数
ｊ ：輸送箱の種類を示す番号
ｎ ：輸送箱の種類の総数
First, the importance calculation unit 32 will be described. As shown in FIG. 5A, the packing record includes a plurality of pieces of information related to the combination of the parts list and the selected transport boxes. Each parts list is composed of a combination of N types of parts. Also, the transport boxes are selected from M types. In this case, the importance calculation unit 32 calculates the ratio of each of the N types of parts packed in each of the M types of transport boxes. For example, the importance calculation unit 32 extracts a parts list including part a, and grasps the type of transport box selected in the extracted parts list and the number of times it has been selected. The importance calculation unit 32 grasps the total number of parts in the parts list including part a and the number of times each type of transport box has been selected, and calculates the ratio of part a packed in each type of transport box. In this way, the importance calculation unit 32 can calculate a vector of part a (see B1 in FIG. 5A) having the ratio of the number of transport box types as an element. For example, the importance calculation unit 32 calculates the ratio of each part packed in each transport box using the following formula (1). The importance calculation unit 32 also calculates the percentage of all items packed into each transport box based on the packing record using the following formula (2): This allows the importance calculation unit 32 to calculate a vector of all parts (see B2 in FIG. 5A) having the percentage of the number of transport box types as an element.

p _ij : Proportion of each part packed in each transport box d _ij : Number of times part i was packed in transport box j i : Number indicating the type of part j : Number indicating the type of transport box n : Total number of types of transport boxes

s _j : Proportion of all parts packed in each transport box d _ij : Number of times part i was packed in transport box j i : Number indicating the type of part m : Total number of part types j : Number indicating the type of transport box n : Total number of types of transport boxes

For example, suppose there are 100 parts lists including part a, and for these, transport box B has been selected 50 times and transport box C has been selected 50 times. In this case, the importance calculation unit 32 calculates that the rate at which part a is packed in transport box B is 50%, and the rate at which it is packed in transport box C is 50%. In this case, the probability distribution of "transport box B: 50%, transport box C: 50%, other types of transport boxes: 0%" is calculated as the vector for part a. Similarly, the probability distribution of "transport box A: 100%, other types of transport boxes: 0%" is calculated as the vector for part b. In this way, the importance calculation unit 32 calculates vectors for all N types of parts.

ｓ_ｊ：全部品がそれぞれの輸送箱に梱包された割合
ｐ_ｉｊ：各部品がそれぞれの輸送箱に梱包された割合
ｉ ：部品の種類を示す番号
ｍ ：部品の種類の総数
ｊ ：輸送箱の種類を示す番号
ｎ ：輸送箱の種類の総数
Next, the importance calculation unit 32 calculates the Euclidean distance between the vector of each part and the vector of all parts, and sets the value as the importance of each part. The importance calculation unit 32 calculates the importance of each part using the following formula (3). The use of Euclidean distance as the importance will be described with reference to FIG. 5B. FIG. 5B is a graph showing the ratio of parts x and y packed in transport boxes A to G, respectively, in terms of probability. Since the probability distribution of part x is completely consistent with the probability distribution of all parts, it can be seen that a transport box is selected for part x without any correlation between the part x and a specific transport box. In contrast, the probability distribution of part y is significantly different from the probability distribution of all parts, and is biased toward transport box A. From this, it can be seen that part y has a part-specific characteristic with respect to the selection of a transport box. The Euclidean distance of part x is small, and the Euclidean distance of part y is large. In the selection of a transport box, it can be considered that a part that has a part-specific characteristic and has a high importance has a large Euclidean distance.

_sj : Proportion of all parts packed in each shipping box _pij : Proportion of each part packed in each shipping box i: Number indicating the type of part m: Total number of part types j: Number indicating the type of shipping box n: Total number of types of shipping boxes

Next, the importance reflection processing unit 33 will be described. After acquiring an information table showing the importance of each part as shown on the left side of FIG. 6(a), the importance reflection processing unit 33 converts it into an information table 70 of the weight of each part as shown on the right side of FIG. 6(a) by normalizing it. When the vector of a part matches perfectly with the vectors of all parts, the weight is 0, and the weight of the part with the highest importance is 1 (here, importance 3 is the maximum).

Next, as shown in FIG. 6(b) and FIG. 4, the importance reflection processing unit 33 multiplies the number of each part stored in the information table 60 of the input information by the weight of each part in the information table 70. In this way, the importance reflection processing unit 33 can create an information table 80 in which the number of parts after the importance reflection is stored. The importance reflection processing unit 33 may prepare a storage section for unknown parts to which the unknown part x belongs in the information tables 70, 80, and may consolidate the number of unknown parts in the storage section for unknown parts. An arbitrary weight may be set for the unknown parts. In this case, the information tables 70, 80 have N+1 rows. Alternatively, the neural network calculation unit 12 may be re-trained using the learning data to which the information of the unknown parts has been added. Alternatively, the number of unknown parts may be discarded.

After storing the number of each part after reflecting the importance in the information table 80 and converting the input information, the importance reflection processing unit 33 inputs the information on the number stored in each row to the neural network calculation unit 12. Note that even during learning, the importance reflection processing unit 33 reflects the importance of each part in the input information before inputting the input information related to the packaging performance to the neural network calculation unit 12.

The neural network calculation unit 12 has an input layer 21, an intermediate layer 22, and an output layer 23 as a model of a neural network. The input layer 21 is a layer for inputting parameters for performing neural network calculations. 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. The input layer 21 has nodes ND according to the number of input parameters. In addition, the intermediate layer 22 has an arbitrary number of nodes ND, and the output layer 23 has nodes ND according to the number of output parameters. For example, the input layer 21 outputs an input vector x = ( _x1 , _x2 , ... _xm ) having m parameters as components to the intermediate layer 22 as it is. The intermediate layer 22 converts the total input into an output by 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 by the weights and activation functions of each layer. This output is an output vector y=(y ₁ , y ₂ , ..., _yn ) of the neural network with n parameters as components. This output vector y indicates an evaluation value. During learning (see FIG. 3), if the correct answer of the output vector for one input vector x is d=(d ₁ , d ₂ , ..., _dn ), the weight w of each layer is updated so that the output vector y approaches the correct answer d. This is information processing called learning. The number of nodes ND in each of the layers 21 to 23 is not particularly limited, and each of the layers 21 to 23 may have nodes ND according to the number of parameters input thereto.

In this embodiment, the number of nodes ND in the input layer 21 is set to be the same as the number of rows (N+1) in the information table 80. The intermediate layer 22 is composed of general layers in neural networks, such as a fully connected layer, a convolutional layer, and a pooling layer, and is appropriately configured according to the work site. The neural network calculation unit 12 calculates the evaluation value using a proportional function. Specifically, the output layer 23 calculates the evaluation value using a proportional function. The proportional function is shown in the following formula (4). The output layer 23 outputs the dimensions of each side of the transport box as the evaluation value. That is, the long side, short side, and height of the transport box are assigned to each node ND of the output layer 23, and the dimensions of each side are output.

yk = zk ... (4)

yk: Output of the kth node in the output layer zk: Value of the kth node in the output layer

The selection unit 13 obtains the dimensions of each side as the evaluation value output from the output layer 23. Then, the selection unit 13 compares the dimensions of each side of the selectable transport boxes (transport boxes in the transport box list BL1) with the dimensions related to the evaluation value to calculate an error. Then, the selection unit 13 selects a transport box that has a minimum error with the dimensions related to the evaluation value. Here, the selection unit 13 may use the sum of squares of the errors of each side (see the following formula (5)) as an evaluation function during selection.

Next, the box selection method according to this embodiment will be described with reference to Figures 7 and 8. Figure 7 is a flowchart showing the contents of the control process performed by the control unit 10 of the box selection device 100 during learning. Figure 8 is a flowchart showing the contents of the control process performed by the control unit 10 of the box selection device 100 during selection of a shipping box. The box selection method is a method for selecting a shipping box in which to pack parts based on previously learned learning information. Note that the detailed processing content in each step is the same as that described in the description of each component of the box selection device 100, and therefore will not be described here.

As shown in FIG. 7, the importance calculation unit 32 executes an importance calculation step S100 in which the importance of each part in box selection in learning by neural network calculation is calculated based on the packing record. The importance reflection processing unit 33 executes an importance reflection processing step S110 in which the number of each part in the input information is weighted based on the importance. The neural network calculation unit 12 executes a learning step S120 in which learning is performed based on the converted input information in which the importance of each part is reflected. The processing of S110 may be executed each time the input information receiving unit 11 inputs a parts list included in the packing record to the neural network calculation unit 12, or may be executed collectively for input information based on all parts lists included in the packing record before the learning step S120 is executed.

In the box selection method, as shown in FIG. 8, a storage step S10 is executed in which the learning information and the box information of the transport box are stored in the storage unit 14. Note that the storage step S10 may be executed in advance before the process of FIG. 8 is executed. Here, as described in FIG. 3, a neural network model is generated or updated as the learning information.

Next, the input information receiving unit 11 executes an input information receiving step S20 in which the input information is obtained and input information is created based on the obtained information. Next, the input information converting unit 31 executes an input information converting step S25 in which the input information is converted. Here, the same processing as the importance reflection processing step S110 in FIG. 7 is performed. Next, the neural network computing unit 12 inputs the input information converted in the input information converting step S25 to the input layer 21, and executes a neural network computing step S30 in which an evaluation value for the shipping box is calculated by neural network computing based on the learning information stored in the memory unit 14. Next, the selection unit 13 executes a selection step S40 in which a shipping box is selected based on the evaluation value output from the output layer 23 in the neural network computing step S30. With the above, the control processing shown in FIG. 7 is completed. When selecting a shipping box for another parts list, the processing may be repeated from the input information receiving step S20.

Next, we will explain the operation and effects of the box selection device 100 according to this embodiment.

In the box selection device 100, the neural network calculation unit 12 inputs input information about the parts to the input layer 21, and calculates an evaluation value about the box by neural network calculation based on the learning information. Then, the selection unit 13 can select a box suitable for the parts to be packed based on the evaluation value output from the output layer 23 of the neural network calculation unit 12. Here, the learning information is learned in advance based on the packing record. In other words, the neural network calculation unit 12 can output an evaluation value for selecting a box suitable for the parts to be packed by omitting to consider complex conditions such as the shape of the parts and the environmental information of the packing destination, and only by learning based on the past packing record. Here, the input information conversion unit 31 can convert the input information input to the neural network calculation unit 12. The input information before conversion has information about the number of each part in a predetermined parts list. For example, when the types of parts in the parts list become enormous, it may be difficult to obtain the packing record for every combination of parts. In response to this, the input information conversion unit 31 has an importance calculation unit 32 that calculates the importance of each part in box selection in learning by neural network calculation based on the packaging record. The importance calculation unit 32 can express the importance of parts that have a large influence on selection and parts that have a small influence when selecting boxes. The input information conversion unit 31 also has an importance reflection processing unit 33 that weights the number of each part in the input information based on the importance. This allows the importance reflection processing unit 33 to convert the input information into a form that reflects the magnitude of the influence on selection. This allows the neural network calculation unit 12 to learn in a manner that grasps the parts that should be emphasized when selecting boxes. As described above, the neural network calculation unit 12 can learn so as to improve the accuracy of box selection.

Furthermore, when learning is performed by adding additional data such as part shapes, as in conventional methods, the number of part types in the parts list becomes enormous, and it takes a huge amount of effort to obtain the additional data. On the other hand, the box selection device 100 according to this embodiment can improve the box selection accuracy without using such additional data.

The importance may be set based on the Euclidean distance between the vector of each part and the vector of all parts. The vector of a part that has a high importance in box selection due to its unique characteristics has a large Euclidean distance between the vector of all parts. Therefore, by setting the importance based on the Euclidean distance, the importance calculation unit 32 can appropriately calculate the importance.

The importance calculation unit 32 may calculate the importance based on the deviation between the proportion of each part packed in each box and a probability distribution indicating the proportion of all parts packed in each box. For example, the probability distribution of a part with small part-specific characteristics in box selection will be similar to the probability distribution of all parts, and the deviation will be small. On the other hand, the probability distribution of a part with large part-specific characteristics in box selection will be a unique distribution, and will therefore deviate greatly from the probability distribution of all parts. In this way, the importance calculation unit 32 can appropriately calculate the importance based on the deviation of the probability distribution.

The information processing method according to this embodiment is an information processing method used in a box selection method having a storage step S10 for storing previously learned learning information and box information, an input information reception step S20 for creating input information related to parts, an input information conversion step S25 for converting the input information, a neural network calculation step S30 for inputting the converted input information into an input layer and calculating an evaluation value related to the box by neural network calculation based on the previously learned learning information based on the packaging record, and a selection step S40 for selecting a box based on the evaluation value output from the output layer in the neural network calculation step S30, in which the input information before conversion has information on the number of each part in a predetermined parts list, and has an importance calculation step S100 for calculating the importance of each part in box selection in learning by neural network calculation based on the packaging record, and an importance reflection processing step S110 for weighting the number of each part in the input information based on the importance.

The information processing program according to this embodiment is an information processing program used in a box selection program that causes a computer system to execute a storage step S10 for storing previously learned learning information and box information, an input information reception step S20 for creating input information related to parts, an input information conversion step S25 for converting the input information, a neural network calculation step S30 for inputting the converted input information into an input layer and calculating an evaluation value related to the box by neural network calculation based on the previously learned learning information based on the packaging record, and a selection step S40 for selecting a box based on the evaluation value output from the output layer in the neural network calculation step S30. The input information before conversion has information regarding the number of each part in a predetermined parts list, and causes the computer system to execute an importance calculation step S100 for calculating the importance of each part in box selection in learning by neural network calculation based on the packaging record, and an importance reflection processing step S110 for weighting the number of each part in the input information based on the importance.

These information processing methods and programs can achieve the same effects as the box selection device described above.

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

In the above embodiment, the importance calculation unit calculates the importance using Euclidean distance, but the method of calculating the importance is not particularly limited. For example, the importance may be calculated using various methods such as Manhattan distance, KL divergence, histogram intersection, and cosine similarity.

ｓ_ｊ：全部品がそれぞれの輸送箱に梱包された割合
ｐ_ｉｊ：各部品がそれぞれの輸送箱に梱包された割合
ｉ ：部品の種類を示す番号
ｍ ：部品の種類の総数
ｊ ：輸送箱の種類を示す番号
ｎ ：輸送箱の種類の総数

ｓ_ｊ：全部品がそれぞれの輸送箱に梱包された割合
ｐ_ｉｊ：各部品がそれぞれの輸送箱に梱包された割合
ｉ ：部品の種類を示す番号
ｍ ：部品の種類の総数
ｊ ：輸送箱の種類を示す番号
ｎ ：輸送箱の種類の総数
For example, when the importance calculation unit calculates the importance by Manhattan distance, the importance calculation unit may calculate the importance by using the following formula (6). When the importance calculation unit calculates the importance by KL divergence, the importance calculation unit may calculate the importance by using the following formula (7).

_sj : Proportion of all parts packed in each shipping box _pij : Proportion of each part packed in each shipping box i: Number indicating the type of part m: Total number of part types j: Number indicating the type of shipping box n: Total number of types of shipping boxes

_sj : Proportion of all parts packed in each shipping box _pij : Proportion of each part packed in each shipping box i: Number indicating the type of part m: Total number of part types j: Number indicating the type of shipping box n: Total number of types of shipping boxes

In addition, the configuration of each layer of the neural network calculation unit is not particularly limited. For example, the intermediate layer can be of any type, such as fully connected or convolutional. In addition, the output layer does not have to output three sides, and can be in the form of one-hot output.

10...control unit, 11...input information receiving unit, 12...neural network calculation unit, 13...selection unit, 14...memory unit, 21...input layer, 23...output layer, 31...input information conversion unit, 32...importance calculation unit, 33...importance reflection processing unit, 100...box selection device.

Claims (4)

A box selection device that selects a box for packaging an item based on previously learned learning information,
The control unit of the box selection device
A storage unit that stores the learning information and box information;
an input information receiving unit for creating input information regarding the item;
an input information conversion unit that converts the input information;
a neural network calculation unit that inputs the converted input information into an input layer and calculates an evaluation value for the box by neural network calculation based on the learning information that has been learned in advance based on packaging records;
a selection unit that selects the box based on the evaluation value output from an output layer of the neural network calculation unit,
the input information before conversion includes information regarding the number of each item in a predetermined list of items;
The input information conversion unit
an importance calculation unit that calculates an importance of each item in box selection in learning by the neural network calculation based on the packaging record;
an importance reflection processing unit that weights the number of each item in the input information based on the importance,
A box selection device, wherein the importance is set based on the Euclidean distance between the vector of each item and the vector of all items. The box selection device according to claim 1, wherein the importance calculation unit calculates the importance based on the deviation between the ratio of each item packed in each box and a probability distribution indicating the ratio of all items packed in each box. a storage step of storing previously learned learning information and box information;
an input information receiving step for creating input information relating to an item;
an input information conversion step of converting the input information;
a neural network calculation step of inputting the converted input information into an input layer and calculating an evaluation value for the box by neural network calculation based on the learning information that has been learned in advance based on packaging records;
and a selection step of selecting the box based on the evaluation value output from an output layer in the neural network operation step,
the input information before conversion includes information regarding the number of each item in a predetermined list of items;
an importance calculation step of calculating an importance in box selection for each item in learning by the neural network calculation based on the packaging performance;
an importance reflection processing step of weighting the number of each item in the input information based on the importance;
An information processing method, wherein the importance is set based on the Euclidean distance between the vector of each item and the vector of all items. a storage step of storing previously learned learning information and box information;
an input information receiving step for creating input information relating to an item;
an input information conversion step of converting the input information;
a neural network calculation step of inputting the converted input information into an input layer and calculating an evaluation value for the box by neural network calculation based on the learning information that has been learned in advance based on packaging records;
a selection step of selecting the box based on the evaluation value output from an output layer in the neural network calculation step,
the input information before conversion includes information regarding the number of each item in a predetermined list of items;
an importance calculation step of calculating an importance in box selection for each item in learning by the neural network calculation based on the packaging performance;
an importance reflection processing step of weighting the number of each item in the input information based on the importance;
An information processing program, wherein the importance is set based on the Euclidean distance between the vector of each item and the vector of all items.

Images