Trained Model Generation Device, Fault Prediction Device, Fault Prediction System, Fault Prediction Program, and Trained Model

Abstract

To accurately predict a failure of a ticket examination appliance.SOLUTION: A learned model generating device comprises: a model data acquisition unit 14 for receiving model operation information 20 that is operation information on a ticket examination appliance and model environment information 21 that is environment information on the ticket examination appliance, which are acquired over time in a predetermined first period in each of a plurality of ticket examination appliances, and for receiving failure information when a failure occurs in each of the ticket examination appliances during the first period; a feature quantity calculating unit 18 for generating a first feature quantity 25 by calculating the model operation information 20 and the model environment information 21; and a model generating unit 17 for generating a first learned model 23 that outputs failure prediction information 36 on each of the ticket examination appliances when the operation information and the environment information are input by machine learning using the first feature quantity 25 as input data and failure information as teacher data.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] The present invention relates to a generation device for a learned model used for predicting failures of ticket inspection devices such as ticket gates, and bidding devices such as ticket vending machines and ticket collection machines, a failure prediction device that predicts failures of ticket inspection devices using the learned model, a failure prediction system, a failure prediction program, and a learned model.

Background Art

[0002] Conventionally, maintenance of ticket inspection devices has been carried out by performing repairs whenever a failure occurs and performing maintenance inspections of each ticket inspection device every 1 to 6 months. In addition, in order to prevent failures from occurring during the use of ticket inspection devices as much as possible and to perform maintenance inspections efficiently, the maintenance inspection cycle has sometimes been reviewed at any time. For example, as shown in Patent Document 1, by continuously referring to the operating status such as the values of various sensors provided in the ticket gate and the history of failures, and artificially judging the operating status of each ticket gate, ticket gates that are likely to fail are extracted, and the maintenance inspection cycle has been reviewed, such as advancing the timing of maintenance inspection of the ticket gate.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the conventional method of rechecking the maintenance inspection time of the ticket gate machine artificially based on the values of various sensors and the like described above, when a single component fails, the failure of the ticket gate machine can be predicted in advance from the values of various sensors and the like. However, in reality, it has been difficult to accurately determine the failure time of the ticket gate machine. For example, in the outgoing ticket gate equipment, the operating accuracy of the outgoing ticket gate equipment may decrease due to the decrease in the operating accuracy of a plurality of components, and the failure of the outgoing ticket gate equipment may often occur. Although it may be possible to artificially organize the values of various sensors by statistical methods and predict a certain degree of complex failure, in reality, the factors leading to the failure of the outgoing ticket gate equipment are very diverse, and it has been virtually impossible to artificially predict the failure of the outgoing ticket gate equipment caused by the combined operation of such a plurality of components.

[0005] An object of the present invention is to accurately predict the failure of the outgoing ticket gate equipment.

Means for Solving the Problems

[0006] In order to achieve the above object, a learned model generation device according to an embodiment of the present invention is a learned model generation device that generates a learned model used for predicting the failure of the ticket gate equipment, and receives model operation information that is the operation information of the ticket gate equipment and model environment information that is the environment information of the ticket gate equipment, which are obtained over time in a predetermined first period for each of the plurality of ticket gate equipment, and receives failure information when a failure occurs in each of the ticket gate equipment during the first period. A model data acquisition unit, a feature quantity calculation unit that generates a first feature quantity by calculating the model operation information and the model environment information, and a first learned model that outputs failure prediction information for each of the ticket gate equipment when the operation information and the environment information are input by machine learning using the first feature quantity as input data and the failure information as teacher data. And a model generation unit for generating.

[0007] With such a configuration, a first pre-trained model for use in predicting failures of the bidding device can be generated based on the operating information and environmental information that affect the failures of the bidding device and the failure information. The bidding device is composed of various components, and the states of these components are combined in a complex manner to cause the failure of the bidding device. It is very difficult to comprehensively judge the states of these complex components. However, machine learning can be used to recognize the relationships between the states of many components and failures. Therefore, it is possible to recognize signs of the bidding device failing beyond human judgment, and the possibility of discovering signs of the bidding device failing can be improved even in situations where it is difficult to make a judgment manually. As a result, a first pre-trained model capable of accurately predicting failures can be generated. By using this first pre-trained model to perform failure prediction, it becomes possible to accurately predict failures of the bidding device.

[0008] Further, a feature extraction unit is provided that analyzes the first pre-trained model and extracts a second feature amount from among the first feature amounts under a predetermined condition considering the degree of influence for outputting the failure prediction information. The model generation unit generates a second pre-trained model that outputs the failure prediction information of each of the bidding devices when the operating information and the environmental information are input by machine learning using the second feature amount as input data and the failure information as teacher data. It is preferable that the failure prediction of the bidding device is performed using the second pre-trained model.

[0009] With such a configuration, by generating a second pre-trained model using the second feature amount that has a great influence on failure prediction, a second pre-trained model with higher failure prediction accuracy can be generated. By using this second pre-trained model to perform failure prediction, it becomes possible to more accurately predict failures of the bidding device.

[0010] Further, the bidding device may include a plurality of blocks, the operation information for model generation may be acquired for each block, the first feature quantity may be generated for each block, the first trained model may be generated for each block, the second feature quantity may be generated for each block, and the second trained model may be generated for each block.

[0011] With such a configuration, it is possible to generate a second trained model capable of predicting a failure for each block even for a bidding device having a complex configuration including a plurality of blocks. By performing failure prediction using this second trained model, it is possible to perform failure prediction for each block, and it becomes possible to perform failure prediction of the bidding device in more detail and with higher accuracy.

[0012] A failure prediction device according to an embodiment of the present invention is a failure prediction device that performs failure prediction of a bidding device using a trained model, and includes a first feature quantity generated by calculating operation information for model generation, which is operation information of the bidding device acquired over time in a predetermined first period for each of the plurality of bidding devices, and environmental information for model generation, which is environmental information of the bidding device, as input data, and failure information when a failure occurs in each of the bidding devices during the first period as teacher data. A model acquisition unit that acquires the trained model generated to output failure prediction information of each of the bidding devices when the operation information and the environmental information are input by machine learning; a prediction data acquisition unit that receives prediction operation information, which is the operation information, and prediction environmental information, which is the environmental information, acquired over time in a predetermined second period for each of the plurality of bidding devices; and a failure prediction unit that inputs the prediction operation information and the prediction environmental information to the trained model acquired via the model acquisition unit and outputs the failure prediction information of each of the bidding devices.

[0013] With such a configuration, a learned model for predicting failures of the bidding device, which is generated based on the operation information and environmental information that affect the failures of the bidding device and the failure information, can be obtained, and failure prediction can be performed using this learned model. Therefore, it is possible to recognize signs of a failure of the bidding device beyond human judgment, and even in situations where it is difficult to make a judgment manually, the possibility of discovering signs of a failure of the bidding device is improved. As a result, failure prediction can be performed using a learned model that enables accurate failure prediction, and accurate failure prediction of the bidding device can be performed.

[0014] A failure prediction system according to an embodiment of the present invention includes the learned model generation device, a prediction data acquisition unit that receives the operation information for prediction, which is the operation information, and the environmental information for prediction, which is the environmental information, that are obtained over time in a predetermined second period for each of the plurality of bidding devices, and a failure prediction unit that inputs the operation information for prediction and the environmental information for prediction into the first learned model generated by the learned model generation device to output the failure prediction information for each of the bidding devices.

[0015] With such a configuration, a first learned model for predicting failures of the bidding device, which is generated based on the operation information and environmental information that affect the failures of the bidding device and the failure information, can be obtained, and failure prediction can be performed using this first learned model. Therefore, it is possible to recognize signs of a failure of the bidding device beyond human judgment, and even in situations where it is difficult to make a judgment manually, the possibility of discovering signs of a failure of the bidding device is improved. As a result, failure prediction can be performed using a first learned model that enables accurate failure prediction, and accurate failure prediction of the bidding device can be performed.

[0016] In addition, a failure prediction system according to an embodiment of the present invention includes the learned model generation device, a prediction data acquisition unit that receives prediction operation information, which is the operation information, and prediction environment information, which is the environment information, that are obtained over time in a predetermined second period for each of the plurality of bidding devices, and a failure prediction unit that inputs the prediction operation information and the prediction environment information into the second learned model generated by the learned model generation device to output the failure prediction information for each of the bidding devices.

[0017] With such a configuration, by acquiring a second learned model generated using a second feature amount that has a great influence on failure prediction, failure prediction can be performed using a second learned model with higher accuracy, and the failure prediction of the bidding device can be performed more accurately.

[0018] Alternatively, it may include the learned model generation device, a prediction data acquisition unit that receives prediction operation information, which is the operation information, obtained over time in a predetermined second period for each of the blocks of the plurality of bidding devices, and prediction environment information, which is the environment information, obtained over time in a predetermined second period for each of the plurality of bidding devices, and a failure prediction unit that inputs the prediction operation information and the prediction environment information into the second learned model generated by the learned model generation device to output the failure prediction information for each block of the bidding device.

[0019] With such a configuration, even for a bidding device with a complex configuration including a plurality of blocks, failure prediction can be performed for each block, and the failure prediction of the bidding device can be performed more accurately.

[0020] A failure prediction program according to an embodiment of the present invention is a failure prediction program that predicts failures of bidding devices using a learned model. The failure prediction program includes: a step of obtaining a learned model generated by calculating a first feature amount, which is operation information for a model and environmental information for a model that are operation information of the bidding devices and environmental information of the bidding devices, respectively, obtained over time in a predetermined first period for each of the plurality of bidding devices, and using the first feature amount as input data and failure information when a failure occurs in each of the bidding devices during the first period as teacher data, and generating, by machine learning, the learned model that outputs failure prediction information for each of the bidding devices when the operation information and the environmental information are input; a step of receiving prediction operation information, which is the operation information, and prediction environmental information, which is the environmental information, obtained over time in a predetermined second period for each of the plurality of bidding devices; and a step of inputting the prediction operation information and the prediction environmental information into the obtained learned model to cause the learned model to output the failure prediction information for each of the bidding devices, and causing a computer to execute these steps.

[0021] With such a configuration, it is possible to obtain a learned model for failure prediction of bidding devices generated from operation information and environmental information that affect the failures of the bidding devices and failure information, and perform failure prediction using this learned model. Therefore, it is possible to recognize signs of a failure of a bidding device beyond artificial judgment, and the possibility of discovering signs of a failure of a bidding device can be improved even in a situation where it is difficult to make an artificial judgment. As a result, failure prediction can be performed using a learned model capable of accurately performing failure prediction, and the failure of the bidding device can be accurately predicted.

[0022] Further, a failure prediction program according to an embodiment of the present invention is a failure prediction program that performs failure prediction of bidding equipment using a learned model, and includes model operation information that is operation information of the bidding equipment and model environment information that is environment information of the bidding equipment, which are acquired over time in a predetermined first period for each of the plurality of bidding equipment, and receives failure information when a failure occurs in each of the bidding equipment during the first period, a step of generating a first feature amount by calculating the model operation information and the model environment information, a step of generating a first learned model that outputs failure prediction information for each of the bidding equipment when the operation information and the environment information are input by machine learning using the first feature amount as input data and the failure information as teacher data, a step of receiving prediction operation information that is the operation information and prediction environment information that is the environment information, which are acquired over time in a predetermined second period for each of the plurality of bidding equipment, and a step of inputting the prediction operation information and the prediction environment information into the first learned model to output the failure prediction information for each of the bidding equipment, and causing a computer to execute these steps.

[0023] With such a configuration, it is possible to generate a first learned model for failure prediction of bidding equipment based on operation information and environment information that affect the failure of the bidding equipment and failure information. Therefore, it is possible to recognize signs of failure of bidding equipment beyond artificial judgment, and the possibility of discovering signs of failure of bidding equipment can be improved even in situations where it is difficult to make an artificial judgment. As a result, it is possible to generate a first learned model capable of accurately performing failure prediction. By performing failure prediction using this first learned model, it is possible to accurately predict the failure of bidding equipment.

[0024] Also, a failure prediction program according to an embodiment of the present invention is a failure prediction program that performs failure prediction of bidding devices using a learned model, and receives model operation information that is operation information of the bidding devices and model environment information that is environment information of the bidding devices, which are obtained over time in a predetermined first period for each of the plurality of bidding devices, and receives failure information when a failure occurs in each of the bidding devices during the first period. A step of generating a first feature amount by calculating the model operation information and the model environment information, a step of generating a first learned model that outputs failure prediction information of each of the bidding devices when the operation information and the environment information are input by machine learning using the first feature amount as input data and the failure information as teacher data, a step of analyzing the first learned model and extracting a second feature amount from the first feature amounts under a predetermined condition considering the influence degree for outputting the failure prediction information, a step of generating a second learned model that outputs the failure prediction information of each of the bidding devices when the operation information and the environment information are input by machine learning using the second feature amount as input data and the failure information as teacher data, a step of receiving prediction operation information that is the operation information and prediction environment information that is the environment information, which are obtained over time in a predetermined second period for each of the plurality of bidding devices, and a step of inputting the prediction operation information and the prediction environment information into the second learned model to output the failure prediction information of each of the bidding devices are executed by a computer.

[0025] With such a configuration, by generating a second learned model using the second feature amount that has a great influence on failure prediction, a second learned model with higher failure prediction accuracy can be generated. By performing failure prediction using this second learned model, it becomes possible to more accurately predict the failure of the bidding devices.

[0026] In addition, a learned model according to an embodiment of the present invention is a learned model for causing a computer to function so as to output failure prediction information of the bidding device based on prediction operation information and prediction environment information acquired from the bidding device. The learned model is composed of a decision tree including a plurality of branch points arranged in a tree structure. The prediction operation information is operation information acquired over time at the bidding device for a predetermined period, and the prediction environment information is environment information acquired over time at the bidding device for a predetermined period. When the prediction operation information and the prediction environment information are input, a plurality of feature amounts are calculated from the prediction operation information and the prediction environment information, and the corresponding feature amounts are calculated at each of the branch points to calculate evaluation values corresponding to the feature amounts, and at the same time, the progress direction is determined and the process proceeds to the next branch point. This process is repeated, and the failure prediction information is calculated and output by adding up the evaluation values calculated at the branch points that have been traversed, causing the computer to function.

[0027] With such a configuration, the learned model can perform failure prediction of the bidding device using operation information and environment information that affect the failure of the bidding device. Therefore, it is possible to recognize a sign of the bidding device failing beyond artificial judgment, and even in a situation where it is difficult to make an artificial judgment, the possibility of discovering a sign of the bidding device failing is improved. As a result, it becomes possible to perform failure prediction with high accuracy.

[0028] Also, a plurality of the decision trees may be combined and configured.

[0029] With such a configuration, it is possible to generate a learned model according to the situation of the bidding device that is complexly related, and it becomes possible to perform failure prediction with higher accuracy.

[0030] In addition, the failure prediction information may include the time when each of the bidding devices is predicted to fail.

[0031] With such a configuration, it is possible to accurately grasp the period during which the bidding device is expected to fail, and perform maintenance and inspection of the bidding device at an appropriate time.

[0032] In addition, a plurality of the first learned models may be generated, and the failure prediction information output by each of the first learned models may indicate the probability that the bidding device will fail during different periods.

[0033] With such a configuration, it is possible to output a plurality of pieces of failure prediction information with different periods until the bidding device fails by using a plurality of learned models. Therefore, it is possible to more accurately grasp the period during which the bidding device is expected to fail, and perform maintenance and inspection of the bidding device at a more appropriate time.

[0034] In addition, a plurality of the second learned models may be generated, and the failure prediction information output by each of the second learned models may indicate the probability that the bidding device will fail during different periods.

[0035] With such a configuration, it is possible to output a plurality of pieces of failure prediction information with different periods until the bidding device fails by using a plurality of learned models. Therefore, it is possible to more accurately grasp the period during which the bidding device is expected to fail, and perform maintenance and inspection of the bidding device at a more appropriate time.

[0036] In addition, the failure prediction information may include the part of the bidding device predicted to fail.

[0037] With such a configuration, by preferentially inspecting the part of the bidding device predicted to fail, it is possible to efficiently perform maintenance and inspection of the bidding device.

[0038] In addition, it may be provided with a maintenance plan unit that generates a maintenance plan for each of the bidding devices and changes each of the maintenance plans based on the failure prediction information.

[0039] With such a configuration, the maintenance and inspection of the bidding equipment can be carried out more appropriately and efficiently, and it is possible to suppress the failure of the bidding equipment during operation.

[0040] Furthermore, a learned model generation program according to an embodiment of the present invention is a learned model generation program for generating a learned model used for predicting the failure of bidding equipment using the learned model, and includes: receiving model operation information, which is operation information of the bidding equipment, and model environment information, which is environment information of the bidding equipment, that are obtained over time in a predetermined first period for each of the plurality of bidding equipment, and receiving failure information in the case where a failure occurs in each of the bidding equipment during the first period; generating a first feature quantity by calculating the model operation information and the model environment information; and generating a first learned model that outputs failure prediction information of each of the bidding equipment when the operation information and the environment information are input by machine learning using the first feature quantity as input data and the failure information as teacher data, and causing a computer to execute these steps.

[0041] With such a configuration, it is possible to generate a learned model for predicting the failure of the bidding equipment based on the operation information and environment information that affect the failure of the bidding equipment and the failure information. The bidding equipment is composed of various parts, and the states of these parts are combined in a complex manner to cause the failure of the bidding equipment. It is very difficult to comprehensively judge the states of these complex parts, but machine learning can recognize the relationship between the states of many parts and failures. Therefore, by recognizing the signs of the failure of the bidding equipment beyond artificial judgment, it is possible to generate a learned model capable of performing highly accurate failure prediction. By performing failure prediction using this learned model, it becomes possible to accurately predict the failure of the bidding equipment.

Brief Description of the Drawings

[0042]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Embodiments for Carrying Out the Invention

[0043] First, with reference to FIG. 1, the ticket checking machine 1a which is an example of the ticket checking device, the ticket vending machine 1b and the ticket cashing machine 1c which are examples of the bidding machines, and their maintenance and repair will be described. Note that the ticket checking device and the bidding machine are collectively referred to as the ticket checking and bidding device 1.

[0044] The ticket gate 1a is installed at railway stations and the like, and manages the entry and exit of the station premises. The ticket gate 1a reads the information recorded on the ticket, determines the propriety of entry and exit according to the information, and manages the entry and exit of the station premises. As tickets, ordinary tickets, commuter tickets, SF cards (Stored Fare Cards), IC cards, etc. are used. The ticket gate 1a includes an insertion slot (not shown), a reading unit (not shown), an antenna unit (not shown), an analysis unit (not shown), an operation control unit (not shown), a storage unit (not shown), a conveyance unit (not shown), an ejection port (not shown), etc. The insertion slot is where ordinary tickets, commuter tickets, SF cards, etc. are inserted. The reading unit reads the information recorded by magnetism or the like. The antenna unit exchanges information with an IC card or the like. The analysis unit analyzes the information acquired by the reading unit or the antenna unit. The operation control unit controls the necessary operations according to the analysis result of the analysis unit. The storage unit stores ordinary tickets and the like for which management has ended at the time of exit or the like. The conveyance unit conveys the ordinary tickets and the like inserted from the insertion slot to the ejection port or the storage unit via the reading unit. The ejection port ejects ordinary tickets and the like that are necessary at the time of entry and at the time of subsequent exit or the like.

[0045] The ticket vending machine 1b is installed at railway stations and the like, and is used for selling tickets and the like. The fare adjustment machine 1c automatically calculates the fare deficiency to the station when a ticket is inserted, collects or subtracts the fare deficiency in coins, banknotes, cards, etc., and also performs the exit process or gatekeeping of the ticket, issues a magnetized exit permit, etc., and automatically pays out change.

[0046] As shown in FIG. 2, the ticket vending machine 1b and the ticket collector 1c include a ticket issuing block 41, a bill block 42, a coin block 43, a card block 44, a customer service section 45, a control section 46, and the like. The ticket issuing block 41 is a block that issues tickets. It cuts the thermal paper into a specified size and prints the amount, characters, etc. along the ticket surface format. Also, the ticket issuing block 41 writes magnetic information on the back of the tickets used for the automatic ticket gate. The bill block 42 discriminates and stores the inserted bills and discharges the change bills. The coin block 43 discriminates and stores the inserted coins and discharges the change coins. The card block 44 is a block that handles various cards. Multiple types of cards are inserted from a single insertion slot, and processing is performed according to the type of the inserted card. The customer service section 45 accepts the insertion of coins, bills, and cards, accepts the input operation of the destination buttons (amount, station name, etc.), and discharges tickets, change, and cards. The control section 46 controls various operations of the ticket vending machine 1b and performs ticket issuing processing, closing processing, etc. based on the input from the operation section such as the customer service section 45 or the instruction from the upper-level device.

[0047] In addition, operating sensors (not shown) for acquiring these operation information 6 are provided at each part constituting each of the ticket gate 1a, the ticket vending machine 1b, and the ticket collector 1c. Also, the ticket gate 1a, the ticket vending machine 1b, and the ticket collector 1c are equipped with environmental sensors (not shown) for acquiring environmental information 7 such as temperature, humidity, illuminance, atmospheric pressure, and weather inside or around them.

[0048] When the outgoing ticket gate device 1 breaks down, an on-call is made for the station staff or the like to contact the maintenance company or the like about the breakdown. In the on-call, information for identifying the faulty outgoing ticket gate device 1 and the situation of the breakdown are communicated. The person in charge of the maintenance company or the like who receives the on-call goes to the corresponding outgoing ticket gate device 1 and performs repairs. Also, in order to suppress the occurrence of breakdowns during operation, each outgoing ticket gate device 1 is regularly maintained and inspected based on a maintenance plan.

[0049] Next, with reference to FIGS. 1 to 3, an overview of the overall configuration of the failure prediction system for the outgoing ticket gate device 1 will be described.

[0050] One or more ticket gates 1 are connected to the Internet line 2. Further, a learned model generation device 3, a failure prediction execution unit 4, and a server 5 are connected to the Internet line 2.

[0051] The server 5 acquires and stores the operation information 6 and the environment information 7 over time from each ticket gate 1 via the Internet line 2. Further, the server 5 acquires and stores the on-call data 8 of the on-call performed in each ticket gate 1 over time via the Internet line 2. Specifically, the on-call data 8 may be directly transmitted from the ticket gate 1 to the server 5, but may be once aggregated at the call center and then transmitted from the call center to the server 5 or the learned model generation device 3.

[0052] The learned model generation device 3 first acquires, via the Internet line 2 from the server 5, the operation information 6, the environment information 7, and the on-call data 8 corresponding to each ticket gate 1, which were acquired during a predetermined period (corresponding to the first period) (step #1 in FIG. 3).

[0053] Next, the learned model generation device 3 calculates, for each ticket gate 1, the operation information 6 as the operation information 20 for the model (see FIG. 5) and the environment information 7 as the environment information 21 for the model (see FIG. 5), and calculates a feature amount (corresponding to the first feature amount) for generating a learned model. Also, the learned model generation device 3 extracts failure information from the on-call data 8 for each ticket gate 1. In the following description, the extracted failure information is also simply referred to as the on-call data 8 (step #3 in FIG. 3).

[0054] Next, the trained model generation device 3 uses the calculated feature amounts as input data for machine learning and the extracted failure information as teacher data, associates the feature amounts corresponding to the same ticket gate device 1 with the failure information, and performs machine learning by the AI (artificial intelligence) 9. By inputting the operation information 6 for prediction (corresponding to the operation information for prediction) and the environment information 7 for prediction (corresponding to the environment information for prediction), a trained model 10 that outputs the failure prediction information 11 of the ticket gate device 1 is generated (step #4 in FIG. 3). Here, the AI 9 may be provided on the Internet line 2 or may be provided in the trained model generation device 3. The specific configuration of the trained model generation device 3 will be described in detail later.

[0055] Thereafter, the server 5 acquires and stores the operation information 6 and the environment information 7 used for performing failure prediction from each ticket gate device 1 via the Internet line 2 (step #5 in FIG. 3).

[0056] Next, the failure prediction execution unit 4 acquires the operation information 6 and the environment information 7 stored in the server 5 via the Internet line 2. Then, for each ticket gate device 1, the failure prediction execution unit 4 uses the operation information 6 continuously acquired during a predetermined period (second period) as the operation information for prediction, and uses the environment information 7 continuously acquired during a predetermined period (second period) as the environment information for prediction, inputs the operation information for prediction and the environment information for prediction into the trained model 10, and outputs the failure prediction information 11 for each ticket gate device 1 (step #6 in FIG. 3). The specific configuration of the failure prediction execution unit 4 will be described in detail later.

[0057] Furthermore, a maintenance plan unit 12 connected to the Internet line 2 may be provided. The maintenance plan unit 12 formulates the maintenance plan for each ticket gate device 1 and changes the maintenance plan so as to increase the priority of maintenance of the ticket gate device 1 with a high possibility of failure based on the failure prediction information 11 as necessary (step #7 in FIG. 3). The specific configuration of the maintenance plan unit 12 will be described in detail later.

[0058] As described above, in this embodiment, the operation information 6 and the environment information 7 of each ticket gate device 1 are collected for a predetermined period, and these are calculated to obtain information suitable for predicting a failure, which is used as input data. The input data is labeled with the actually generated on-call data 8 and used as teacher data, and a failure prediction is performed using the learned model 10 that has been machine-learned. Therefore, it is possible to comprehensively judge a large amount of information to perform a failure prediction, and it is possible to accurately perform a failure prediction of the ticket gate device 1. In particular, since there are very many factors causing the failure of the ticket gate device 1 and they act on each other in a complex manner, it was difficult to judge a failure from the information obtained from the ticket gate device 1 artificially. However, since the learned model 10 is generated using machine learning, it is possible to comprehensively judge more information to perform a failure prediction. Also, although it is not impossible to use a program to select artificially necessary information and predict a failure from their correlation relationships, ultimately, it is inevitable to select the necessary information and derive their correlation relationships artificially, which is not realistic considering the complexity of the ticket gate device 1. Therefore, compared with the case of using such a program, by generating the learned model 10 as described above and performing a failure prediction, it is possible to perform a failure prediction more easily and with higher accuracy. Also, in the ticket gate device 1, the influence on the failure changes depending on the environment such as the meteorological situation such as air pressure and humidity. For example, the ease of paper jamming of tickets changes depending on the environment, or the influence on the operation accuracy of electronic devices changes. Therefore, by including the environment information 7 in the input data for machine learning, it is possible to generate a learned model 10 that takes into account the complex interaction between the operation information 6 and the environment information 7. And since it is possible to review the maintenance plan based on a highly accurate failure prediction, it is possible to reduce the possibility that the ticket gate device 1 during actual operation fails.

[0059] 〔Embodiment 1〕 Next, as Embodiment 1, specific failure predictions of the ticket vending machine 1b and the ticket validator 1c, which are the ticket gate devices 1, will be described with reference to FIGS. 1 to 8. In the following description, mainly, among the ticket vending machine 1b and the ticket validator 1c, the ticket vending machine 1b will be described as an example.

[0060] 〔Trained Model Generation Device〕 First, with reference to FIGS. 1 to 3, specific configuration examples of the trained model generation device 3 and specific examples of the trained model generation flow will be described using FIGS. 4 to 8.

[0061] The trained model generation device 3 includes a data communication unit 13, a model data acquisition unit 14, a storage unit 15, a generation control unit 16, a model generation unit 17, a feature quantity calculation unit 18, and a feature quantity extraction unit 19.

[0062] The data communication unit 13 transmits and receives data to and from the ticket vending machine 1b, the AI 9, etc. via the Internet line 2. The model data acquisition unit 14 acquires the model operation information 20 and the model environment information 21 of the ticket vending machine 1b received by the data communication unit 13 and the on-call data 8, and stores them in the storage unit 15 (step #1 in FIG. 6).

[0063] Here, the ticket vending machine 1b and the ticket checker 1c have a more complex internal structure than the ticket barrier 1a. Also, as described above, the ticket vending machine 1b and the ticket checker 1c include a ticket issuing block 41, a bill block 42, a coin block 43, and a card block 44, and the configurations of these blocks are each complex, and the causes of failures are also independent. Therefore, although failure prediction may be performed for the entire ticket vending machine 1b and the ticket checker 1c, it is more appropriate to perform failure prediction independently for each of the ticket issuing block 41, the bill block 42, the coin block 43, and the card block 44.

[0064] Therefore, the model operation information 20 is acquired for each of the four blocks. The model operation information 20 is information collected over time for a predetermined period (first period) for each of the ticket issuing block 41, the bill block 42, the coin block 43, and the card block 44 from the operation information 6 obtained from the ticket vending machine 1b. That is, the model operation information 20 for the ticket issuing block 41, the model operation information 20 for the bill block 42, the model operation information 20 for the coin block 43, and the model operation information 20 for the card block 44 are acquired.

[0065] The specified period can be set arbitrarily. For example, it can be set to one month, which can be the one-month period before the month in which failure prediction is performed, or the one-month period immediately before failure prediction. The operation information 20 for model is acquired from each of the four blocks of each ticket vending machine 1b, and is information associated with each block of the ticket vending machine 1b. For example, in each block, the operation information 20 for model consists of a total of 2000 pieces of information (feature quantities). Also, the environmental information 21 for model is acquired from the ticket vending machine 1b and is information associated with each ticket vending machine 1b. For example, it consists of a total of 2000 pieces of information (feature quantities). Note that the environmental information 21 for model may also be acquired for each block in the same manner as the operation information 20 for model.

[0066] In addition to the operation information 20 for model, the environmental information 21 for model, and the on-call data 8, the storage unit 15 stores various data such as the learning input data 22 described later, the first learned model 23, and the second learned model 24. The generation control unit 16 includes a processor such as a CPU or a GPU, and controls the operations of the learned model generation device 3 such as the model generation unit 17 and the feature quantity calculation unit 18.

[0067] Based on the control of the generation control unit 16, the feature quantity calculation unit 18 calculates the first feature quantity 25 for each block by calculating at least a part of the operation information 20 for model among the acquired operation information 20 for model and the environmental information 21 for model (step #2 in FIG. 6).

[0068] For example, as shown in FIG. 7, the first feature quantity 25 calculates the utilization rate of the IC card per day from the number of uses of each of the IC card, ordinary ticket, commuter ticket, and SF card among the operation information 6. Further, as the first feature quantity 25, the total or average value of the utilization rate of the IC card for three days, or the total or average value for seven days may be calculated. In addition, the differential value of the number of times the ordinary ticket is jammed during a predetermined period or the like is calculated as the first feature quantity 25. Also, the first feature quantity 25 may be the failure content or the like in each block or in the functional units within each block.

[0069] In this way, the first feature quantity 25 of each block is composed of the operation information 20 for the model and the environmental information 21 for the model, with the calculated items and their values added thereto. As a result, the first feature quantity 25 consists of, for example, a total of 20,000 pieces of information (feature quantities) in each block. The first feature quantity 25 of each block is information that may affect the failure of each block of the ticket vending machine 1b. By generating a large amount of this information and making a comprehensive judgment, it becomes possible to accurately predict the failure of each block of the ticket vending machine 1b.

[0070] The model generation unit 17 includes a processor such as a CPU or a GPU, and based on the control of the generation control unit 16, for each block, input data is the first feature quantity 25, and teacher data is the on-call data 8 (failure information). The learning input data 22 is input to the AI 9 for machine learning to generate the first learned model 23 (step #3 in FIG. 6). The on-call data 8 is information including the on-call information performed when each block of the ticket vending machine 1b fails and the results of repairs performed along with the on-call, and is accumulated for each on-call. Therefore, the on-call data 8 includes various information. As shown in FIG. 8, the failure information extracted from the on-call data 8 includes at least the ID for identifying the target ticket vending machine 1b, the information of the target block, the call date and time when the on-call was made, the on-call content, the failed part determined from the repair result, and the cause of the failure.

[0071] Based on the control of the generation control unit 16, the feature quantity extraction unit 19 inputs the first feature quantity 25 to the first learned model 23 for analysis for each block to obtain the degree of influence of each feature quantity on failure prediction. Then, based on the control of the generation control unit 16, the feature quantity extraction unit 19 extracts, for example, 100 feature quantities in descending order of the degree of influence on failure prediction from the 20,000 feature quantities for each block, and sets the extracted 100 feature quantities as the second feature quantity 26 (step #4 in FIG. 6).

[0072] Then, based on the control of the generation control unit 16, the model generation unit 17 inputs the learning input data 27, where the input data is the second feature amount 26 and the teacher data is the on-call data 8, to the AI 9 for each block for machine learning, and generates the second learned model 24 (step #5 in FIG. 6).

[0073] Note that the first learned model 23 and the second learned model 24 are models used to predict whether the ticket vending machine 1b will fail within a predetermined prediction period, for example, within 7 days. Also, the first learned model 23 and the second learned model 24 are, for example, decision trees. The first learned model 23 and the second learned model 24 having such a decision tree structure evaluate a predetermined feature amount (the first feature amount 25 or the second feature amount 26) at each branch point of the decision tree, and an evaluation value (failure probability) corresponding to the evaluation result is given for each branch point. Then, the evaluation values are summed along the branches of the decision tree to obtain the failure prediction information 36. Note that the first learned model 23 and the second learned model 24 may be ensemble models such as XGBoost, Random Forest, LightGBM, CatBoost, AdaBoost, etc., in which a plurality of decision trees are provided in association with each other.

[0074]

[0075] ​Furthermore, for each block, the first learned model 23 is used to analyze the first feature quantity 25, and feature quantities with a large impact on failures are extracted as the second feature quantity 26. These second feature quantities 26 are used as input data to generate a second learned model 24. As a result, failure prediction can be performed using the second learned model 24 with higher failure prediction accuracy, and failure prediction for each block of the ticket vending machine 1b can be performed with higher accuracy.

[0076] 〔Failure Prediction Execution Unit〕 Next, specific configuration examples of the failure prediction execution unit 4 and specific examples of the failure prediction flow will be described with reference to FIGS. 1 to 3 and using FIGS. 4, 5, and 9.

[0077] The failure prediction execution unit 4 includes a data communication unit 28, a model acquisition unit 29, a prediction data acquisition unit 30, a storage unit 31, a prediction control unit 32, and a failure prediction unit 33.

[0078] The data communication unit 28 transmits and receives data to and from the ticket vending machine 1b, the learned model generation device 3, etc. via the Internet line 2. The model acquisition unit 29 acquires, for each block, the first learned model 23 or the second learned model 24, which is a learned model used for failure prediction, from the learned model generation device 3 via the data communication unit 28 and stores it in the storage unit 31 (step #1 in FIG. 9). The prediction data acquisition unit 30 acquires the prediction operation information 34 and the prediction environment information 35 of the ticket vending machine 1b received by the data communication unit 28 and stores them in the storage unit 31 (step #2 in FIG. 9). The prediction operation information 34 and the prediction environment information 35 are information obtained by collecting the operation information 6 acquired from each block of the ticket vending machine 1b and the environment information 7 of the ticket vending machine 1b over a predetermined period (the second period) over time. The predetermined period can be set arbitrarily, for example, the day before failure prediction is performed. The prediction control unit 32 includes a processor such as a CPU and controls the operations of the failure prediction unit 33, etc. Note that the prediction environment information 35 may also be acquired for each block in the same manner as the prediction operation information 34.

[0079] Based on the control of the prediction control unit 32, the failure prediction unit 33 inputs the prediction operation information 34 and the prediction environment information 35 for each block into the learned model of the corresponding block to perform failure prediction (step #3 in FIG. 9), stores the failure prediction information 36 for each block in the storage unit 31, and then outputs it (step #4 in FIG. 9). In this embodiment, the learned model is the second learned model 24, but it is also possible to use the first learned model 23.

[0080] The failure prediction information 36 indicates the probability that each block of each ticket vending machine 1b will fail within the prediction period. When the prediction period is 7 days, the probability that each block of the ticket vending machine 1b will fail within 7 days is indicated by a value between 0 and 1 for each block of each ticket vending machine 1b. The closer this value is to 1, the higher the probability of failure. When the second learned model 24 is composed of a decision tree, the failure prediction information 36 is derived by calculating the second feature quantity 26 from the prediction operation information 34 and the prediction environment information 35, evaluating the second feature quantity 26 at each branch point, and adding up the evaluation values every time a branch point is passed.

[0081] In addition, the failure prediction performed using the second learned model 24 was verified under the following conditions. The second learned model 24 was generated with the first period being 4 months and the first feature quantity 25 being 19303, and then failure prediction was performed for 1 month. As a result, the model accuracy was AUC (Area under an ROC curve): 75.31%, Precision (rate of not making false alarms): 10%, and Recall (rate of not missing): 70%. From this, according to this embodiment, it can be seen that failures can be predicted with higher accuracy compared to the case of performing a maintenance plan based on conventional artificial judgment, and the probability that the ticket vending machine 1b will fail during actual operation when the failure prediction of this embodiment is performed is significantly reduced compared to the prior art.

[0082] Furthermore, as the failure prediction information 36, in each block of each ticket vending machine 1b, the failure part predicted to fail may be output together with its failure probability. For example, in each block of each ticket vending machine 1b, the top three parts with a high probability of failure are output.

[0083] In failure prediction, the evaluation value of each second feature amount 26 is calculated. The failure probability for each part of each ticket vending machine 1b can be obtained by adding up the evaluation values of the second feature amounts 26 that are determined in advance to have an impact on the failure of each part among the evaluation values of each second feature amount 26. As the failure prediction information 36, for each block of the ticket vending machine 1b, the part with a high probability of failure is output together with the probability of failure.

[0084] In this way, in bidding devices such as the ticket vending machine 1b and the cash register 1c, by generating a learned model for each block of the ticket issuing block 41, the banknote block 42, the coin block 43, and the card block 44 and performing failure prediction, even for a bidding device with a complex internal structure, failure prediction specialized for each block can be performed, and the prediction accuracy can be improved.

[0085] In addition, since failure prediction is performed using the learned models (the first learned model 23 and the second learned model 24) that are machine-learned using the feature amounts (the first feature amount 25 and the second feature amount 26) that are presumed to have a great influence on the failure, the failure prediction for each block of bidding devices such as the ticket vending machine 1b and the cash register 1c can be performed accurately. In particular, by performing failure prediction using the second learned model 24 that is machine-learned using the second feature amount 26 extracted in consideration of the magnitude of the influence on the failure prediction, the failure prediction for each block of the bidding device can be performed more accurately. As a result, based on the failure prediction information 36, the deadline for maintenance of the block of the bidding device with a high probability of failure becomes clear, and it becomes possible to review the cycle and timing of the maintenance plan and perform maintenance and inspection more appropriately.

[0086] In addition, by outputting the parts with a high probability of failure for each block of the bidding equipment, the parts that should be preferentially maintained and inspected are clarified, and it becomes possible to perform maintenance and inspection efficiently.

[0087] 〔Embodiment 2〕 Next, as Embodiment 2, specific failure prediction of the ticket gate machine 1a, which is the ticket checking and issuing equipment 1, will be described with reference to FIGS. 1, 3, 5, 6, 9 to 11.

[0088] 〔Trained Model Generation Device〕 First, a specific configuration example of the trained model generation device 3 and a specific example of the trained model generation flow will be described with reference to FIGS. 5, 6, 10, and 11 while referring to FIGS. 1 and 3.

[0089] The trained model generation device 3 includes a data communication unit 13, a model data acquisition unit 14, a storage unit 15, a generation control unit 16, a model generation unit 17, a feature quantity calculation unit 18, and a feature quantity extraction unit 19.

[0090] The data communication unit 13 transmits and receives data to and from the ticket gate machine 1a, AI 9, etc. via the Internet line 2. The model data acquisition unit 14 acquires the model operation information 20 and the model environment information 21 of the ticket gate machine 1a received by the data communication unit 13 and the on-call data 8, and stores them in the storage unit 15 (step #1 in FIG. 6). The model operation information 20 and the model environment information 21 are information obtained by collecting the operation information 6 and the environment information 7 acquired from the ticket gate machine 1a over a predetermined period (the first period) over time. The predetermined period can be arbitrarily set. For example, it can be set to one month, or it can be one month before the month when the failure prediction is performed, or it can be one month immediately before the failure prediction is performed. The model operation information 20 and the model environment information 21 are information acquired from each ticket gate machine 1a and attached to the ticket gate machine 1a, and are composed of, for example, a total of 2000 pieces of information (feature quantities). In addition to the model operation information 20, the model environment information 21, and the on-call data 8, the storage unit 15 also stores various data such as the learning input data 22, the first learned model 23, and the second learned model 24, which will be described later. The generation control unit 16 includes a processor such as a CPU or a GPU, and controls the operations of the learned model generation device 3 such as the model generation unit 17 and the feature quantity calculation unit 18.

[0091] Based on the control of the generation control unit 16, the feature quantity calculation unit 18 calculates at least a part of the operation information 20 for the model, among the operation information 20 for the model and the environment information 21 for the model, to calculate the first feature quantity 25 (step #2 in FIG. 6). For example, as shown in FIG. 10, from the respective usage frequencies of the IC card, ordinary ticket, commuter ticket, and SF card among the operation information 6, the utilization rate of the IC card per day is calculated as the first feature quantity 25. Further, as the first feature quantity 25, the total and average value of the utilization rate of the IC card for three days, the total and average value for seven days may be calculated. In addition, the differential value of the number of times the ordinary ticket is full within a predetermined period, etc. is calculated as the first feature quantity 25. In this way, the first feature quantity 25 is composed of the operation information 20 for the model and the environment information 21 for the model, with the calculated items and their values added thereto. As a result, the first feature quantity 25 consists of, for example, a total of 20,000 pieces of information (feature quantities). The first feature quantity 25 is information that may affect the failure of the ticket gate 1a, and by generating a large amount of this information and making a comprehensive judgment, the failure of the ticket gate 1a can be accurately predicted.

[0092] The model generation unit 17 includes a processor such as a CPU or GPU, and based on the control of the generation control unit 16, inputs the learning input data 22 with the input data being the first feature quantity 25 and the teacher data being the on-call data 8 (failure information) to the AI 9 for machine learning to generate the first learned model 23 (step #3 in FIG. 6). The on-call data 8 is information including the on-call information performed when the ticket gate 1a fails and the result of the repair performed along with the on-call, and is accumulated for each on-call. Therefore, the on-call data 8 includes various information, and as shown in FIG. 11, the failure information extracted from the on-call data 8 includes at least the ID for identifying the target ticket gate 1a, the call date and time when the on-call was made, the on-call content, the failed part determined as a result of the repair, and the cause of the failure.

[0093] Based on the control of the generation control unit 16, the feature extraction unit 19 inputs the first feature quantity 25 into the first trained model 23 for analysis, and obtains the influence degree of each feature quantity on the failure prediction. Then, based on the control of the generation control unit 16, the feature extraction unit 19 extracts, for example, 100 feature quantities in order from those with a large influence degree on the failure prediction from 20,000 feature quantities, and sets these extracted 100 feature quantities as the second feature quantity 26 (step #4 in FIG. 6).

[0094] Then, based on the control of the generation control unit 16, the model generation unit 17 inputs the learning input data 27 with the input data being the second feature quantity 26 and the teacher data being the on-call data 8 into the AI 9 for machine learning to generate the second trained model 24 (step #5 in FIG. 6).

[0095] Note that the first trained model 23 and the second trained model 24 are models used to perform a failure prediction on whether the ticket gate machine 1a will fail within a predetermined prediction period, for example, within 7 days. Also, the first trained model 23 and the second trained model 24 are, for example, decision trees. The first trained model 23 and the second trained model 24 with such a decision tree structure evaluate a predetermined feature quantity (the first feature quantity 25 or the second feature quantity 26) at each branch point of the decision tree, and an evaluation value (failure probability) corresponding to the evaluation result is given for each branch point. Then, the evaluation values are summed up along the branches of the decision tree to obtain the failure prediction information 36. Note that the first trained model 23 and the second trained model 24 may be ensemble models such as XGBoost, Random Forest, LightGBM, CatBoost, AdaBoost, etc., in which a plurality of decision trees are provided in association with each other.

[0096] In this way, characteristic quantities that may affect the failure of the ticket gate 1a are selected from the operation information 6 and the environment information 7, and these are calculated so that the degree of influence on the failure becomes clear, and characteristic quantities are added to generate the first characteristic quantity 25. Then, the first learned model 23 is generated using these first characteristic quantities 25, which are presumed to have a large influence on the failure, as input data. Therefore, it is possible to perform failure prediction in which the interaction of each characteristic quantity is properly reflected, and the failure prediction of the ticket gate 1a can be performed with high accuracy.

[0097] Furthermore, the first characteristic quantity 25 is analyzed using the first learned model 23, characteristic quantities with a large degree of influence on the failure are extracted as the second characteristic quantity 26, and the second learned model 24 is generated using these second characteristic quantities 26 as input data. As a result, failure prediction can be performed using the second learned model 24 with higher failure prediction accuracy, and the failure prediction of the ticket gate 1a can be performed with even higher accuracy.

[0098] 〔Failure Prediction Execution Unit〕 Next, a specific configuration example of the failure prediction execution unit 4 and a specific example of the failure prediction flow will be described with reference to FIGS. 5 and 9 while referring to FIGS. 1 and 3.

[0099] The failure prediction execution unit 4 includes a data communication unit 28, a model acquisition unit 29, a prediction data acquisition unit 30, a storage unit 31, a prediction control unit 32, and a failure prediction unit 33.

[0100] The data communication unit 28 transmits and receives data to and from the ticket gate machine 1a, the learned model generation device 3, etc. via the Internet line 2. The model acquisition unit 29 acquires, via the data communication unit 28, the first learned model 23 or the second learned model 24, which is a learned model used for failure prediction, from the learned model generation device 3, and stores it in the storage unit 31 (step #1 in FIG. 9). The prediction data acquisition unit 30 acquires the prediction operation information 34 and the prediction environment information 35 of the ticket gate machine 1a received by the data communication unit 28, and stores them in the storage unit 31 (step #2 in FIG. 9). The prediction operation information 34 and the prediction environment information 35 are information obtained by collecting the operation information 6 and the environment information 7 acquired from the ticket gate machine 1a over a predetermined period (the second period) over time. The predetermined period can be arbitrarily set, for example, the day before failure prediction is performed. The prediction control unit 32 includes a processor such as a CPU and controls the operations of the failure prediction unit 33 and the like.

[0101] Based on the control of the prediction control unit 32, the failure prediction unit 33 inputs the prediction operation information 34 and the prediction environment information 35 into the learned model to perform failure prediction (step #3 in FIG. 9), stores the failure prediction information 36 in the storage unit 31, and then outputs it (step #4 in FIG. 9). In this embodiment, the learned model is the second learned model 24, but it is also possible to use the first learned model 23.

[0102] The failure prediction information 36 indicates the probability of each of the ticket gate machines 1a failing within the prediction period. When the prediction period is 7 days, the probability of the ticket gate machine 1a failing within 7 days is indicated by a value between 0 and 1 for each ticket gate machine 1a. The closer this value is to 1, the higher the probability of failure. When the second learned model 24 is configured as a decision tree, the failure prediction information 36 is derived by calculating the second feature amount 26 from the prediction operation information 34 and the prediction environment information 35, evaluating the second feature amount 26 at each branch point, and adding up the evaluation values every time a branch point is passed.

[0103] Note that the failure prediction performed using the second pre-trained model 24 was verified under the following conditions. The second pre-trained model 24 was generated with the first period being four months and the first feature quantity 25 being 19,303, and then failure prediction was performed for one month. As a result, the model accuracy was AUC (Area under an ROC curve): 75.31%, Precision (rate of not idling): 10%, and Recall (rate of not missing): 70%. From this, according to this embodiment, it can be predicted more accurately than in the case of performing a maintenance plan based on conventional manual judgment, and it can be seen that the probability of the ticket gate machine 1a failing during actual operation when the failure prediction of this embodiment is performed is significantly reduced compared to the conventional case.

[0104] Furthermore, as the failure prediction information 36, in each ticket gate machine 1a, the failure part predicted to fail may be output together with its failure probability. For example, in each ticket gate machine 1a, the top three parts with a high probability of failure are output.

[0105] In failure prediction, the evaluation value of each second feature quantity 26 is calculated. The failure probability for each part of each ticket gate machine 1a can be obtained by adding up the evaluation values of the second feature quantities 26 that are determined in advance to have an impact on the failure of each part among the evaluation values of each second feature quantity 26. As the failure prediction information 36, for each ticket gate machine 1a, the part with a high probability of failure is output together with the probability of failure.

[0106] In this way, by using the trained models (the first trained model 23 and the second trained model 24) that are machine-learned using the feature quantities (the first feature quantity 25 and the second feature quantity 26) that are presumed to have a great influence on the failure, it is possible to accurately perform the failure prediction of the ticket gate machine 1a. In particular, by performing the failure prediction using the second trained model 24 that is machine-learned using the second feature quantity 26 extracted in consideration of the magnitude of the influence on the failure prediction, it is possible to more accurately perform the failure prediction of the ticket gate machine 1a. As a result, based on the failure prediction information 36, the deadline for maintenance of the ticket gate machine 1a with a high probability of failure becomes clear, and it becomes possible to review the cycle and timing of the maintenance plan and perform maintenance and inspection more appropriately.

[0107] Also, by outputting the parts with a high probability of failure for each ticket gate machine 1a, the parts that should be mainly maintained and inspected become clear, and it becomes possible to efficiently perform maintenance and inspection.

[0108] Note that by performing the failure prediction in the above-described Embodiment 1 and Embodiment 2, as shown in FIG. 12, it can be seen that the inspection cost and the number of on-call cases are improved as compared with the management based on the conventional maintenance plan.

[0109] 〔Maintenance Plan〕 The failure prediction execution unit 4 in Embodiment 1 and Embodiment 2 may further include a maintenance planning unit 37. The maintenance planning unit 37 first determines a maintenance plan for periodically performing maintenance and inspection on each ticket gate device 1 (each ticket gate machine 1a, each ticket vending machine 1b, each payment machine 1c). Then, based on the failure prediction information 36, the maintenance planning unit 37 changes the maintenance plan so that maintenance and inspection are performed on each ticket gate machine 1a, each ticket vending machine 1b, or each payment machine 1c within the expected failure period.

[0110] The maintenance planning unit 37 may be configured to be connected to a display device (not shown). The maintenance planning unit 37 displays the maintenance plan and the failure prediction information 36 on the display device. The person in charge of maintenance, the administrator, etc. can further review the maintenance plan while referring to the maintenance plan and the failure prediction information 36 displayed on the display device.

[0111] In this way, by changing the maintenance plan based on the failure prediction information 36, it is possible to efficiently perform maintenance and inspection, perform maintenance and inspection of the ticket gate device 1 before a failure occurs, and suppress the failure of the operating ticket gate device 1.

[0112] Note that the maintenance plan unit 37 is not limited to the configuration provided in the failure prediction execution unit 4, and may be provided at any location. For example, the maintenance plan unit 37 may be configured to be connected to the Internet line 2 like the maintenance plan unit 12 shown in FIG. 1.

[0113] 〔Another Embodiment〕 (1) In each of the above embodiments, the configurations of the learned model generation device 3 and the failure prediction execution unit 4 are not limited to being divided into functional blocks as shown in FIG. 5, and the functional blocks may be arbitrary as long as the functions of the learned model generation device 3 and the failure prediction execution unit 4 can be realized. For example, the functional blocks of the generation control unit 16, the model generation unit 17, the feature quantity calculation unit 18, and the feature quantity extraction unit 19 are not limited to such a configuration, and may be configured by functional blocks that integrate some or all of the functions of each other, or may be configured by functional blocks obtained by further subdividing each functional block. Also, in each of the above embodiments, some or all of the functions of the learned model generation device 3 and the failure prediction execution unit 4 may be integrated.

[0114] (2) In each of the above embodiments, the learned model generation device 3 may be configured to generate a learned model independently and complete it without transmitting the learned model to the failure prediction execution unit 4. Conversely, the failure prediction execution unit 4 may perform failure prediction using a learned model generated separately instead of using the learned model generated by the learned model generation device 3.

[0115] (3) In each of the above embodiments, part or all of the learned model generation method realized by the learned model generation device 3 and the failure prediction method realized by the failure prediction execution unit 4 are not limited to the above device configuration and can be realized with any configuration. Further, part or all of the learned model generation method realized by the learned model generation device 3 and the failure prediction method realized by the failure prediction execution unit 4 may be realized by a program. The program is stored in an arbitrary recording device such as the storage unit 15 or the storage unit 31, and an arbitrary processor (corresponding to a computer) such as the CPU or GPU provided in the generation control unit 16 executes this program.

[0116] (4) In each of the above embodiments, the on-call data 8 may be configured to include only on-call information and not include the repair result. In this case, the on-call data 8 (failure information) is composed of an ID for identifying the target ticket gate device 1, the call date and time when the on-call was made, and the on-call content. Note that the on-call content may be only information indicating whether the on-call was made. In this case, the on-call data 8 (failure information) is composed of an ID for identifying the target ticket gate device 1, the call date and time when the on-call was made, and the presence or absence of the on-call.

[0117] (5) In each of the above embodiments, the learned model generation device 3 may be configured not to include the feature amount extraction unit 19 and not to generate the second feature amount 26 and the second learned model 24. In this case, the learned model generation device 3 generates only the first feature amount 25 and the first learned model 23.

[0118] (6) In each of the above embodiments, the failure prediction is not limited to the configuration that performs failure prediction only when the prediction period is 7 days, and a failure prediction that outputs the probability of failure during a plurality of different prediction periods may be performed. For example, as the prediction periods, 1 month (30 days), 14 days, 7 days, and 3 days are set, and as the failure prediction, the probability that the ticket gate device 1 or any block fails within 1 month, the probability of failure within 14 days, the probability of failure within 7 days, and the probability of failure within 3 days are output. In this case, the first learned model 23 and the second learned model 24 are generated for each of the ticket gate device 1 or its blocks, four types of learned models for obtaining the probability of failure within 1 month, four types of learned models for obtaining the probability of failure within 14 days, four types of learned models for obtaining the probability of failure within 7 days, and four types of learned models for obtaining the probability of failure within 3 days. Then, the failure prediction execution unit 4 performs failure prediction for each of the ticket gate device 1 or its blocks using the four types of first learned models 23 or the four types of second learned models 24, and outputs four types of failure prediction information 36. For example, the failure prediction is executed every Monday, and the probability of failure within 3 days, the probability of failure within 7 days, the probability of failure within 14 days, and the probability of failure within 1 month are output. Then, for example, a maintenance plan is made by referring to the probability of failure within 3 days and the probability of failure within 7 days, and the probability of failure within 14 days and the probability of failure within 1 month are used as reference information.

[0119] As a result, the probability of failure during a plurality of periods is shown, and the timing for performing maintenance and inspection can be grasped more accurately. Accordingly, the change of the maintenance plan can also be performed more accurately.

[0120] (7) In each of the above embodiments, the first learned model 23 and the second learned model 24 may be updated as the failure prediction is carried out. By inputting the operation information for prediction 34 and the environmental information for prediction 35 used in the failure prediction performed during a predetermined period into the first learned model 23 or the second learned model 24 and performing machine learning, the first learned model 23 or the second learned model 24 is updated. At this time, the result of the failure prediction using these operation information for prediction 34 and environmental information for prediction 35 and the result of maintenance and repair may be added to the teacher data.

[0121] In this way, by updating the learned model as the failure prediction is carried out, the accuracy of the learned model is improved, and it becomes possible to perform a more accurate failure prediction.

[0122] (8) In each of the above embodiments, the ticket gate device 1 may be configured not to include an environmental sensor. In this case, the learned model 10 is generated from the operation information 6 and the on-call data 8 (failure information), and the operation information 6 is input into the learned model 10 during the failure prediction. Further, other information may be used for generating the learned model 10, or any combination of the operation information 6, the environmental information 7, and other information may be used for generating the learned model 10.

[0123] The environmental information 7 may not have a significant impact on the failure prediction. In such a case, by adopting a configuration without an environmental sensor, the ticket gate device 1 can be made into a simpler configuration.

[0124] (9) The ticket gate device 1, the learned model generation device 3, the failure prediction execution unit 4, and the maintenance plan unit 12 are not limited to a configuration connected to the Internet line 2, and any configuration in which data can be transferred between them is acceptable. For example, at least a part of these devices may be connected to a network line such as an intranet other than the Internet line 2, or may be configured to be able to transmit and receive data via a dedicated line such as a LAN, or may be configured to transfer data via a storage medium.

[0125] (10) The failure information used to generate the learned model 10 may be extracted from the on-call data 8, or may be obtained from sources other than the on-call data 8. For example, the administrator of the ticket gate device 1 may judge and create the failure information, and input it into the learned model generation device 3 via the Internet line 2 or the like. Alternatively, the ticket gate device 1 may judge a failure by itself, generate the failure information, and input it into the learned model generation device 3 via the Internet line 2 or the like.

[0126] (11) In the second embodiment, the configuration is not limited to performing failure prediction for the entire ticket gate 1a. Similar to the first embodiment, the ticket gate 1a may be divided into a plurality of blocks, and failure prediction may be performed for each block.

[0127] Conversely, in the ticket vending machine 1b and the ticket collector 1c, similar to the first embodiment, the configuration may be such that failure prediction is performed not for each block but for each ticket vending machine 1b or ticket collector 1c as in the second embodiment.

[0128] Furthermore, in the ticket gate device 1, the block is not limited to being divided into a ticket issuing block 41, a bill block 42, a coin block 43, and a card block 44. It may be divided into arbitrary blocks, and failure prediction may be performed for each block.

Industrial Applicability

[0129] The present invention can be applied not only to the ticket gate device installed at railway stations and other transportation stations, but also to the learned model generation device, failure prediction device, failure prediction system, failure prediction program, and learned model in the ticket gate device installed in various facilities such as theme parks.

Explanation of Signs

[0130] 3 Learned model generation device 4 Failure prediction execution unit 6 Operation information 7 Environment information 8 On-call data 9 AI 10 Trained Model 11 Fault Prediction Information 12 Maintenance Planning Department 14 Model Data Acquisition Unit 15 Memory Unit 16 Generation Control Unit 17 Model Generation Unit 18 Feature Calculation Unit 19 Feature Extraction Unit 20 Model Operating Information 21 Model Environment Information 22 Learning Input Data 23 First Trained Model (Trained Model) 24 Second Trained Model (Trained Model) 25 First Feature (Feature) 26 Second Feature (Feature) 27 Learning Input Data 29 Model Acquisition Unit 30 Prediction Data Acquisition Unit 31 Memory Unit 32 Prediction Control Unit 33 Fault Prediction Unit 34 Prediction Operating Information 35 Prediction Environment Information 36 Fault Prediction Information 37 Maintenance Planning Department

Claims

A trained model generation device that generates a trained model used for predicting failures of a bidding device having a plurality of blocks including at least two of a ticket issuing block, a banknote block, a coin block, and a card block, A model data acquisition unit that receives model operation information, which is operation information for each of the blocks of the bidding device, and model environment information, which is environment information of the bidding device, acquired over time in a predetermined first period in each of the plurality of bidding devices, and receives failure information when a failure occurs in each of the bidding devices during the first period, A feature quantity calculation unit that generates a first feature quantity for each of the blocks by calculating the model operation information and the model environment information, A trained model generation device comprising: a model generation unit that generates a first trained model for each of the blocks, which outputs failure prediction information for each of the bidding devices when the operation information and the environment information are input, by machine learning using the first feature quantity as input data and the failure information as teacher data.

2. Comprising a feature quantity extraction unit that analyzes the first trained model and extracts a second feature quantity from among the first feature quantities under a predetermined condition considering the degree of influence for outputting the failure prediction information, The model generation unit generates a second trained model for each of the blocks, which outputs the failure prediction information for each of the bidding devices when the operation information and the environment information are input, by machine learning using the second feature quantity as input data and the failure information as teacher data, and the failure prediction of the bidding device is performed using the second trained model. The trained model generation device according to claim 1.

3. A failure prediction device that predicts failures of a bidding device having a plurality of blocks including at least two of a ticket issuing block, a banknote block, a coin block, and a card block, using a trained model, A model acquisition unit that acquires, for each block, a learned model generated by calculating a first feature amount for each block, which is operation information for the bidding device and model environment information, which is environment information of the bidding device, acquired over time in a predetermined first period for each of the plurality of bidding devices, and using the operation information and the environment information as input data and failure information when a failure occurs in each of the bidding devices during the first period as teacher data, and generating, by machine learning, failure prediction information for each of the bidding devices when the operation information and the environment information are input. A prediction data acquisition unit that receives prediction operation information, which is the operation information for each block, and prediction environment information, which is the environment information for each block, acquired over time in a predetermined second period for each of the plurality of bidding devices. A failure prediction device including: a failure prediction unit that inputs the prediction operation information and the prediction environment information into the learned model acquired via the model acquisition unit, and outputs the failure prediction information for each of the bidding devices for each block.

4. The learned model generation device according to claim 1, A prediction data acquisition unit that receives prediction operation information, which is the operation information for each block, and prediction environment information, which is the environment information for each block, acquired over time in a predetermined second period for each of the plurality of bidding devices. A failure prediction system including: a failure prediction unit that inputs the prediction operation information and the prediction environment information into the first learned model generated by the learned model generation device, and outputs the failure prediction information for each of the bidding devices for each block.

5. The learned model generation device according to claim 2, A prediction data acquisition unit that receives prediction operation information, which is the operation information for each block, and prediction environment information, which is the environment information for each block, acquired over time in a predetermined second period for each of the plurality of bidding devices. A failure prediction system including: a failure prediction unit that inputs the prediction operation information and the prediction environment information into the second learned model generated by the learned model generation device, and outputs the failure prediction information for each of the bidding devices for each block.

6. A failure prediction program for predicting failures of a bidding device having a plurality of blocks including at least two of a ticket issuing block, a banknote block, a coin block, and a card block, using a learned model, obtaining the learned model generated for each block so that when the operation information and the environment information for each block are input, failure prediction information for each bidding device is output, by machine learning using, as input data, first feature amounts generated for each block by calculating model operation information, which is operation information for each block of the bidding device, and model environment information, which is environment information of the bidding device, obtained over time in a predetermined first period for each of the plurality of bidding devices, and using, as teacher data, failure information when a failure occurs in each of the bidding devices during the first period; receiving, for each of the plurality of bidding devices, prediction operation information, which is operation information for each block, and prediction environment information, which is environment information, obtained over time in a predetermined second period; causing a computer to execute a step of inputting the prediction operation information and the prediction environment information into the obtained learned model and outputting the failure prediction information for each block of each bidding device.

7. A failure prediction program for predicting failures of a bidding device having a plurality of blocks including at least two of a ticket issuing block, a banknote block, a coin block, and a card block, using a learned model, receiving, for each of the plurality of bidding devices, model operation information, which is operation information for each block of the bidding device, and model environment information, which is environment information of the bidding device, obtained over time in a predetermined first period, and receiving failure information when a failure occurs in each of the bidding devices during the first period; generating first feature amounts for each block by calculating the model operation information and the model environment information; generating, for each block, a first learned model that outputs failure prediction information for each bidding device when the operation information and the environment information are input, by machine learning using the first feature amounts as input data and the failure information as teacher data; A step of receiving, for each of the plurality of the bidding devices, prediction operation information which is the operation information for each block and is obtained over time in a predetermined second period, and prediction environment information which is the environment information; A failure prediction program that causes a computer to execute a step of inputting the prediction operation information and the prediction environment information into the first learned model and outputting the failure prediction information for each of the respective bidding devices for each block.

8. A failure prediction program for predicting failures of a bidding device having a plurality of blocks including at least two of a ticket issuing block, a bill block, a coin block, and a card block, using a learned model, A step of receiving, for each of the plurality of the bidding devices, model operation information which is the operation information for each block of the bidding device and is obtained over time in a predetermined first period, and model environment information which is the environment information of the bidding device, and receiving failure information in the case where a failure has occurred in each of the bidding devices during the first period; A step of generating a first feature amount for each block by calculating the model operation information and the model environment information; A step of generating, for each block, a first learned model that outputs failure prediction information for each of the respective bidding devices when the operation information and the environment information are input, by machine learning using the first feature amount as input data and the failure information as teacher data; A step of analyzing the first learned model and extracting a second feature amount from among the first feature amounts under a predetermined condition considering the degree of influence for outputting the failure prediction information; A step of generating, for each block, a second learned model that outputs the failure prediction information for each of the respective bidding devices when the operation information and the environment information are input, by machine learning using the second feature amount as input data and the failure information as teacher data; A step of receiving, for each of the plurality of the bidding devices, prediction operation information which is the operation information for each block and is obtained over time in a predetermined second period, and prediction environment information which is the environment information; A failure prediction program that causes a computer to execute a step of inputting the prediction operation information and the prediction environment information into the second learned model and outputting the failure prediction information for each of the respective bidding devices for each block. A learned model for causing a computer to function so as to output failure prediction information of the bidding device based on prediction operation information and prediction environment information obtained from a bidding device having a plurality of blocks including at least two of a ticket issuing block, a paper money block, a coin block, and a card block, which is composed of a decision tree composed of a plurality of branch points arranged in a tree structure, wherein the prediction operation information is operation information for each of the blocks obtained over time in a predetermined period in the bidding device, and the prediction environment information is environment information obtained over time in a predetermined period in the bidding device, a learned model for causing a computer to function so that the prediction operation information and the prediction environment information are input, a plurality of feature amounts are calculated from the prediction operation information and the prediction environment information, the corresponding feature amounts are calculated at each of the branch points to calculate evaluation values corresponding to the feature amounts, and the progress direction is determined to repeat proceeding to the next branch point, and by summing up the evaluation values calculated at the proceeded branch points, the failure prediction information is calculated and output for each of the blocks.

10. The learned model according to claim 9, which is configured by combining a plurality of the decision trees.

Images