JP2022169915A - Learned model generating device, failure predicting device, failure predicting system, failure predicting program, and learned model - Google Patents

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

The present invention relates to a device for generating a learned model used for predicting failures of ticket examination equipment such as ticket gates and ticket issuing equipment such as ticket vending machines and fare adjustment machines, and failure prediction of exiting ticket examination equipment using the learned model. It relates to a prediction device, a failure prediction system, a failure prediction program, and a learned model.

In the past, the maintenance of the ticket gate equipment involved repairing each time a failure occurred, and performing maintenance and inspection of each ticket gate equipment every 1 to 6 months. In addition, in order to prevent failures from occurring during use of the ticket gate equipment and to perform maintenance and inspections efficiently, the maintenance and inspection cycle is sometimes reviewed as needed. For example, as shown in Patent Document 1, the operation status of each ticket gate is determined artificially by continuously referring to the operation status and failure history such as the values of various sensors provided in the ticket gate. The ticket gates that are likely to break down are extracted, and the maintenance and inspection cycle of the ticket gates is reviewed, such as advancing the timing of maintenance and inspection of the ticket gates.

JP-A-2004-38747

However, in the conventional method of manually reviewing the timing of maintenance and inspection of ticket gates based on the values of various sensors, in the event that a single part fails, it is possible to adjust the ticket gates in advance based on the values of various sensors. Although failures can be predicted, in reality, it is difficult to sufficiently determine the failure timing of ticket gates. For example, when the operating accuracy of a plurality of parts of a ticket gate device is lowered, the operating accuracy of the ticket gate device is compounded, which often leads to failure of the ticket gate device. In some cases, it is possible to artificially organize the values of various sensors using statistical methods and predict complex failures to some extent, but in reality, there are a wide variety of factors that lead to ticket gate equipment failures. However, it has been practically impossible to artificially predict the failure of the ticket gate device that is caused comprehensively by the operation of such a plurality of parts.

SUMMARY OF THE INVENTION An object of the present invention is to accurately predict a failure of an exit ticket gate device.

In order to achieve the above object, a trained model generation device according to one embodiment of the present invention is a trained model generation device for generating a trained model used for predicting failures of bidders, comprising a plurality of Receiving model operation information, which is operation information of the bid device, and model environment information, which is environment information of the bid device, which are acquired over time in each of the bid devices in a predetermined first period; A model data acquisition unit that receives failure information when a failure occurs in each of the bid devices during a first period, and generates a first feature amount by calculating the model operation information and the model environment information. and a machine learning using the first feature value as input data and the failure information as teaching data, when the operation information and the environment information are input, failure prediction of each of the bid devices and a model generator that generates a first trained model that outputs information.

With such a configuration, it is possible to generate the first trained model to be used for failure prediction of the ticketing device based on the operation information and environment information that affect the ticketing device failure, and the failure information. A ticketing machine is composed of various parts, and a complicated combination of states of these parts leads to failure of the ticketing machine. While it is extremely difficult to comprehensively judge the states of these complex parts, machine learning can recognize the relationship between the states of many parts and failures. Therefore, it is possible to recognize a sign of the failure of the ticketing device beyond human judgment, and even in a situation where it is difficult to make a decision manually, the possibility of discovering the sign of the failure of the ticketing device is improved. As a result, it is possible to generate a first trained model capable of accurately predicting failures. By performing failure prediction using this first learned model, it is possible to accurately perform failure prediction of the bid equipment.

Further, a feature quantity extraction unit that analyzes the first trained model and extracts a second feature quantity from the first feature quantity under a predetermined condition considering the degree of influence for outputting the failure prediction information. and the model generating unit performs machine learning using the second feature as input data and the failure information as teacher data, and when the operation information and the environment information are input, the failure of each of the bid devices It is preferable that a second trained model for outputting prediction information is generated, and the failure prediction of the bid device is performed using the second trained model.

With such a configuration, it is possible to generate a second trained model with higher failure prediction accuracy by generating the second trained model using the second feature amount that greatly affects failure prediction. By performing failure prediction using this second trained model, it becomes possible to perform more accurate failure prediction of the bid equipment.

Further, the bidding device includes a plurality of blocks, the model operation information is obtained for each block, the first feature amount is generated for each block, and the first trained model is generated for each block. , a second feature may be generated for each block, and the second trained model may be generated for each block.

With such a configuration, it is possible to generate a second trained model capable of performing failure prediction for each block even for a bid device having a complicated configuration including a plurality of blocks. By performing failure prediction using this second trained model, failure prediction can be performed for each block, and more detailed and accurate failure prediction can be performed for the devices to be bid.

A failure prediction device according to an embodiment of the present invention is a failure prediction device that performs failure prediction for a bidder using a learned model, wherein each of the plurality of bidders is chronologically determined in a predetermined first period. Using as input data a first feature amount generated by calculating the model operation information, which is the operation information of the bidding device, and the model environment information, which is the environment information of the bidding device, which are acquired, the first period When the operation information and the environment information are input, failure prediction information for each of the bid devices is output by machine learning using failure information when a failure occurs in each of the bid devices as teacher data. a model acquisition unit that acquires the learned model generated in the above; and the predictive operation information and the environment information that are the operation information acquired over time in each of the plurality of the bidding devices over a predetermined second period. a prediction data acquisition unit that receives the prediction environment information, and inputs the prediction operation information and the prediction environment information to the learned model acquired via the model acquisition unit, and obtains the and a failure prediction unit for outputting the failure prediction information of the bid device.

With such a configuration, a trained model for use in predicting the failure of the tendering device generated from the operating information and environmental information that affects the failure of the tendering device and the failure information is acquired, and the learned model is used to predict the failure of the tendering device. Predictions can be made. Therefore, it is possible to recognize a sign of the failure of the ticketing device beyond human judgment, and even in a situation where it is difficult to make a decision manually, the possibility of discovering the sign of the failure of the ticketing device is improved. As a result, it is possible to perform failure prediction using a learned model capable of accurately predicting failures, and to accurately predict failures of devices to be bid.

A failure prediction system according to one embodiment of the present invention includes: the learned model generation device; a prediction data acquisition unit that receives information and prediction environment information that is the environment information; a failure prediction unit for inputting and outputting the failure prediction information of each of the bid devices.

With such a configuration, the first trained model to be used for failure prediction of the tendering device generated from the operation information and environmental information that affect the failure of the tendering device and the failure information is acquired, and the first trained model is generated from the failure information. can be used for failure prediction. Therefore, it is possible to recognize a sign of the failure of the ticketing device beyond human judgment, and even in a situation where it is difficult to make a decision manually, the possibility of discovering the sign of the failure of the ticketing device is improved. As a result, it is possible to perform failure prediction using the first learned model capable of performing failure prediction with high accuracy, and to perform failure prediction of the bid device with high accuracy.

Further, in the failure prediction system according to one embodiment of the present invention, the learned model generation device and the prediction, which is the operation information obtained over time in each of the plurality of bidder devices in a predetermined second period, a prediction data acquisition unit that receives the prediction environment information that is the operation information and the environment information; a failure prediction unit for inputting information and outputting the failure prediction information for each of the bid devices.

With such a configuration, by acquiring the second trained model generated using the second feature quantity that greatly affects failure prediction, failure prediction is performed using the second trained model with higher accuracy. Therefore, it is possible to more accurately predict the failure of the bidding device.

Further, the learned model generation device, the prediction operation information which is the operation information obtained over time in each of the blocks of the plurality of the ticketing devices in a predetermined second period, and the plurality of the ticketing devices and the second trained model generated by the trained model generation device, the prediction data acquisition unit receiving the environment information for prediction, which is the environment information acquired over time in a predetermined second period, respectively. a failure prediction section for inputting the operation information for prediction and the environment information for prediction to and outputting the failure prediction information for each block of the bid device.

With such a configuration, it is possible to perform failure prediction for each block even for a ticketing device having a complicated configuration including a plurality of blocks, so that failure prediction of the ticketing device can be performed with higher accuracy.

A failure prediction program according to one embodiment of the present invention is a failure prediction program for predicting failures of bid devices using a learned model, wherein each of the plurality of bid devices is chronologically determined in a predetermined first period. Using as input data a first feature amount generated by calculating the model operation information, which is the operation information of the bidding device, and the model environment information, which is the environment information of the bidding device, which are acquired, the first period When the operation information and the environment information are input, failure prediction information for each of the bid devices is output by machine learning using failure information when a failure occurs in each of the bid devices as teacher data. and the predictive operation information and the environment information, which are the operation information, obtained over time in each of the plurality of the bidding devices over a predetermined second period. a step of receiving environment information for prediction; and a step of inputting the operation information for prediction and the environment information for prediction to the acquired learned model and causing the failure prediction information of each of the devices to be bid to be output. let the computer do it.

With such a configuration, a trained model for use in predicting the failure of the tendering device generated from the operating information and environmental information that affects the failure of the tendering device and the failure information is acquired, and the learned model is used to predict the failure of the tendering device. Predictions can be made. Therefore, it is possible to recognize a sign of the failure of the ticketing device beyond human judgment, and even in a situation where it is difficult to make a decision manually, the possibility of discovering the sign of the failure of the ticketing device is improved. As a result, it is possible to perform failure prediction using a learned model capable of accurately predicting failures, and to accurately predict failures of devices to be bid.

Further, a failure prediction program according to an embodiment of the present invention is a failure prediction program for predicting failures of bid devices using a learned model, wherein each of the plurality of bid devices ages in a predetermined first period. model operation information, which is the operation information of the tendering device, and model environment information, which is the environment information of the tendering device, which are acquired in a systematic manner; a step of generating a first feature quantity by calculating the model operating information and the model environment information; and using the first feature quantity as input data, the failure information a step of generating a first trained model for outputting failure prediction information for each of the bid devices when the operating information and the environment information are input by machine learning using as teacher data; a step of receiving the operation information for prediction as the operation information and the environment information for prediction as the environment information, which are acquired over time in a predetermined second period in each of the above; a step of inputting the operation information and the environment information for prediction and causing the computer to output the failure prediction information for each of the bid devices.

With such a configuration, it is possible to generate the first trained model to be used for failure prediction of the ticketing device based on the operation information and environment information that affect the ticketing device failure, and the failure information. Therefore, it is possible to recognize a sign of the failure of the ticketing device beyond human judgment, and even in a situation where it is difficult to make a decision manually, the possibility of discovering the sign of the failure of the ticketing device is improved. As a result, it is possible to generate a first trained model capable of accurately predicting failures. By performing failure prediction using this first learned model, it is possible to accurately perform failure prediction of the bid device.

Further, a failure prediction program according to an embodiment of the present invention is a failure prediction program for predicting failures of bid devices using a learned model, wherein each of the plurality of bid devices ages in a predetermined first period. model operation information, which is the operation information of the tendering device, and model environment information, which is the environment information of the tendering device, which are acquired in a systematic manner; a step of generating a first feature quantity by calculating the model operating information and the model environment information; and using the first feature quantity as input data, the failure information a step of generating a first trained model that outputs failure prediction information of each of the bid devices when the operating information and the environmental information are input by machine learning using as teacher data; a step of analyzing a model and extracting a second feature value from the first feature value under a predetermined condition considering the degree of influence for outputting the failure prediction information; generating a second trained model that outputs the failure prediction information of each of the bid devices when the operation information and the environment information are input by machine learning using the failure information as teacher data; a step of receiving prediction-use operating information as said operating information and prediction-use environment information as said environment information obtained over time in a predetermined second period in each of said bidding devices; inputting the predictive operation information and the predictive environment information to the model, and causing a computer to output the failure prediction information of each of the bid devices.

With such a configuration, it is possible to generate a second trained model with higher failure prediction accuracy by generating the second trained model using the second feature amount that greatly affects failure prediction. By performing failure prediction using this second trained model, it becomes possible to perform more accurate failure prediction of the bid equipment.

Further, the trained model according to one embodiment of the present invention functions the computer to output failure prediction information of the ticketing device based on the predictive operation information and the predictive environment information acquired from the ticketing device. The model is a trained model for making the bifurcation, which is composed of a decision tree consisting of a plurality of branch points arranged in a tree structure, and the prediction operation information is the operation information acquired over time in a predetermined period in the bidding device. wherein the predictive environment information is environmental information acquired over time in a predetermined period in the bidding device, the predictive operating information and the predictive environment information are input, and the predictive operating information and the predictive A plurality of feature quantities are calculated from the usage environment information, the feature quantity corresponding to each of the branch points is calculated, an evaluation value corresponding to the feature quantity is calculated, and the direction of travel is determined. The computer is made to function so as to calculate and output the failure prediction information by repeating proceeding to branch points and adding up the evaluation values calculated at the proceeding branch points.

With such a configuration, the trained model can predict the failure of the ticketing machine using the operating information and environmental information that affect the ticketing machine failure. Therefore, it is possible to recognize a sign of the failure of the ticketing device beyond human judgment, and even in a situation where it is difficult to make a decision manually, the possibility of discovering the sign of the failure of the ticketing device is improved. As a result, failure prediction can be performed with high accuracy.

Also, a plurality of the decision trees may be combined.

With such a configuration, it is possible to generate a learned model according to the situation of bid devices that are intricately related to each other, and to perform failure prediction with higher accuracy.

Further, the failure prediction information may include a time when each of the bid devices is expected to fail.

With such a configuration, it is possible to accurately grasp the period during which the ticketing device is expected to malfunction, and to perform maintenance and inspection of the ticketing device at an appropriate time.

Further, a plurality of the first trained models may be generated, and the failure prediction information output by each of the first trained models may indicate the probability of failure of the bid equipment during different periods.

With such a configuration, it is possible to output a plurality of pieces of failure prediction information with different periods until failure of the bidder, using a plurality of learned models. Therefore, it is possible to more accurately grasp the period during which the ticketing device is expected to be out of order, and it is possible to perform maintenance and inspection of the ticketing device at a more appropriate time.

Further, a plurality of the second trained models may be generated, and the failure prediction information output by each of the second trained models may indicate the probability of failure of the bid equipment during different periods.

With such a configuration, it is possible to output a plurality of pieces of failure prediction information with different periods until failure of the bidder, using a plurality of learned models. Therefore, it is possible to more accurately grasp the period during which the ticketing device is expected to be out of order, and it is possible to perform maintenance and inspection of the ticketing device at a more appropriate time.

Further, the failure prediction information may include a portion of the bid device that is predicted to fail.

With such a configuration, it is possible to efficiently perform maintenance and inspection of the ticket issuing machine by preferentially inspecting parts of the ticket issuing machine that are expected to fail.

Further, a maintenance planning unit may be provided that generates a maintenance plan for each of the tendered devices and changes each of the maintenance plans based on the failure prediction information.

With such a configuration, maintenance and inspection of the ticketing device can be performed more appropriately and efficiently, and failure of the ticketing device during operation can be suppressed.

Further, a trained model generation program according to an embodiment of the present invention is a trained model generation program for generating a trained model used for predicting a failure of a bid device using a trained model, the program comprising a plurality of Receiving model operation information, which is operation information of the bid device, and model environment information, which is environment information of the bid device, which are acquired over time in each of the bid devices in a predetermined first period; a step of receiving failure information when a failure occurs in each of the bid devices during a first period; a step of generating a first feature quantity by calculating the model operating information and the model environment information; First learning for outputting 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 value as input data and the failure information as teacher data. and generating the finished model.

With such a configuration, it is possible to generate a trained model to be used for failure prediction of a ticketing machine based on operation information and environment information that affect a ticketing machine failure, and failure information. A ticketing machine is composed of various parts, and a complicated combination of states of these parts leads to failure of the ticketing machine. While it is extremely difficult to comprehensively judge the states of these complex parts, machine learning can recognize the relationship between the states of many parts and failures. Therefore, by recognizing signs of failure of the bidder beyond human judgment, it is possible to generate a trained model capable of highly accurate failure prediction. By performing failure prediction using this learned model, it becomes possible to accurately perform failure prediction of the bid device.

1 is a schematic diagram illustrating the configuration of a failure prediction system for ticket gate equipment; FIG. It is a figure which illustrates the structure of a bid apparatus. It is a figure which shows the outline of the flow of the process which the failure prediction system of an exit ticket gate apparatus implements. 4 is a diagram illustrating a data flow related to failure prediction of each block according to the first embodiment; FIG. FIG. 3 is a diagram illustrating configurations of a trained model generation device and a failure prediction execution unit; FIG. 4 is a diagram showing an outline of the flow of processing for generating a learned model; 4 is a diagram for explaining the configuration of feature amounts in Embodiment 1. FIG. 4 is a diagram illustrating the configuration of on-call data according to Embodiment 1. FIG. FIG. 3 is a diagram showing an outline of the flow of processing for failure prediction; FIG. 11 is a diagram for explaining the configuration of feature amounts in Embodiment 2; FIG. 11 is a diagram illustrating the configuration of on-call data in Embodiment 2; It is a figure which compares an inspection cost and the number of on-call with a prior art.

First, referring to FIG. 1, a ticket examination machine 1a as an example of ticket examination equipment, a ticket vending machine 1b and a fare adjustment machine 1c as examples of ticket issuing equipment, and their maintenance and repair will be described. The ticket gate device and the ticket issuing device are collectively referred to as the exit ticket gate device 1 .

The ticket gate 1a is installed in a railway station or the like, and manages entry and exit into the station premises. The ticket gate 1a reads the information recorded on the ticket, judges the propriety of entry and exit according to the information, and manages entry and exit into the station premises. As a ticket, an ordinary ticket, a commuter pass, an SF card (Stored Fare Card), an IC card, or the like is used. The ticket gate 1a includes an inlet (not shown), a reading unit (not shown), an antenna unit (not shown), an analysis unit (not shown), an operation control unit (not shown), and a storage unit (not shown). not shown), a conveying section (not shown), a discharge port (not shown), and the like. An ordinary ticket, a commuter pass, an SF card, etc. are inserted into the slot. The reading unit reads information recorded by magnetism or the like. The antenna section exchanges information with an IC card or the like. The analysis unit analyzes information acquired by the reading unit and the antenna unit. The operation control unit controls necessary operations according to the analysis result of the analysis unit. The storage unit stores ordinary tickets, etc., for which management has been completed at the time of exiting, or the like. The conveying unit conveys an ordinary ticket or the like inserted from the insertion port to the discharge port or the storage unit via the reading unit. The discharge port discharges ordinary tickets or the like that are required when entering the venue or the like and then when exiting the venue or the like.

The ticket vending machine 1b is installed at a railway station or the like and used for ticket sales and the like. When the ticket is inserted, the fare adjustment machine 1c automatically calculates the insufficient fare to the station, collects or subtracts the insufficient fare with coins, banknotes, cards, etc., and processes the exit of the ticket. Alternatively, it collects tickets, issues magnetized participation certificates, etc., and automatically pays out the change.

As shown in FIG. 2, the ticket vending machine 1b and the fare adjustment machine 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. A ticket issuing block 41 is a block for issuing a ticket, and cuts thermal paper into a specified size, and prints the amount of money, characters, etc. according to the ticket surface format. Also, the ticket issuing block 41 writes magnetic information on the reverse side of the ticket to be used in the automatic ticket gate. The banknote block 42 discriminates and stores inserted banknotes, releases bills, and the like. The coin block 43 discriminates and stores inserted coins and releases change coins. The card block 44 is a block that handles various cards. A plurality of types of cards are inserted from one slot, and processing is performed according to the type of the inserted card. The customer service section 45 accepts insertion of coins, banknotes, and cards, accepts input operations of destination buttons (amount of money, station name, etc.), and ejects tickets, change, and cards. The control unit 46 controls various operations of the ticket vending machine 1b, and performs ticket issuing processing, deadline processing, etc. according to inputs from the operation unit such as the customer service unit 45 or instructions from higher-level equipment.

Further, an operation sensor (not shown) for acquiring the operation information 6 is provided in each part constituting the ticket gate 1a, the ticket vending machine 1b, and the fare adjustment machine 1c. In addition, the ticket gate 1a, the ticket vending machine 1b, and the fare adjustment machine 1c are provided with environmental sensors (not shown) for acquiring environmental information 7 such as temperature, humidity, illumination, atmospheric pressure, and weather inside or around them.

When the exit ticket gate device 1 breaks down, an on-call is made by a station employee or the like to report the breakdown to a maintenance company or the like. In the on-call, the information for identifying the failed exit ticket gate device 1 and the status of the failure are communicated. The person in charge of the maintenance company or the like who receives the on-call goes to the relevant ticket gate device 1 and repairs it. Further, each ticket gate device 1 is regularly maintained and inspected based on a maintenance plan in order to prevent failure during operation.

Next, an outline of the overall configuration of the failure prediction system for the exit ticket gate device 1 will be described with reference to FIGS. 1 to 3. FIG.

One or a plurality of exit ticket gate devices 1 are connected to an Internet line 2. - 特許庁Also, a trained model generation device 3 , a failure prediction execution unit 4 , and a server 5 are connected to the Internet line 2 .

The server 5 acquires and stores the operating information 6 and the environment information 7 over time from each exit ticket gate device 1 via the Internet line 2 . Further, the server 5 acquires and stores on-call data 8 of on-calls made at each exit ticket gate device 1 over time via the Internet line 2 . Specifically, the on-call data 8 may be transmitted directly from the exit ticket gate device 1 to the server 5, or may be temporarily aggregated at a call center and transmitted from the call center to the server 5 or the trained model generation device 3.

The learned model generating device 3 firstly generates operation information 6 corresponding to each ticket gate device 1 acquired during a predetermined period (corresponding to the first period) from the server 5 via the Internet line 2. , environmental information 7 and on-call data 8 (step #1 in FIG. 3).

Next, the learned model generating device 3 sets the operating information 6 as the model operating information 20 (see FIG. 5) and the environment information 7 as the model environment information 21 (see FIG. 5). The operating information 20 and the environment information 21 for the model are calculated to calculate the feature amount (corresponding to the first feature amount) for generating the learned model (step #2 in FIG. 3). Also, the learned model generation device 3 extracts failure information from the on-call data 8 for each exit ticket gate device 1 . In the following description, the extracted failure information is simply referred to as on-call data 8 (step #3 in FIG. 3).

Next, the trained model generation device 3 uses the calculated feature amount as input data for machine learning and the extracted failure information as teaching data, and generates the feature amount and failure information corresponding to the same ticket gate device 1. Machine learning is performed by AI (artificial intelligence) 9 in association, and by inputting operation information 6 for prediction (equivalent to operation information for prediction) and environment information 7 for prediction (equivalent to environment information for prediction), output A learned model 10 that outputs 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 . A specific configuration of the trained model generation device 3 will be described in detail later.

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

Next, the failure prediction execution unit 4 acquires the operating information 6 and the environment information 7 stored in the server 5 via the Internet line 2 . Then, the failure prediction execution unit 4 uses the operation information 6 continuously acquired during a predetermined period (second period) for each exit ticket gate device 1 as operation information for prediction, and Using the environment information 7 continuously acquired as the environment information for prediction, the operation information for prediction and the environment information for prediction are input to the trained model 10, and the failure prediction information 11 for each exit ticket gate device 1 is output (Fig. 3 step #6). A specific configuration of the failure prediction execution unit 4 will be described in detail later.

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

As described above, in this embodiment, the operation information 6 and the environment information 7 of each exit ticket gate device 1 are collected for a predetermined period, and these are calculated to calculate information suitable for prediction of failure as input data, Input data is labeled with actually generated on-call data 8, and failure prediction is performed using a trained model 10 that has undergone machine learning as teacher data. Therefore, failure prediction can be performed by comprehensively judging a large amount of information, and failure prediction of the exit ticket gate device 1 can be performed with high accuracy. In particular, there are many causes for the failure of the ticket gate device 1, and they interact in a complex manner, so it is difficult to determine the failure from the information artificially obtained from the ticket gate device 1. Since the learned model 10 is generated using machine learning, failure prediction can be performed by comprehensively judging more information. It is not impossible to use a program to artificially select the necessary information and predict a failure from the correlation between them, but it is possible to select the necessary information and derive the correlation between them. must be done artificially, and considering the complexity of the ticket gate device 1, it is not realistic. Therefore, by generating the learned model 10 and performing failure prediction as described above, failure prediction can be performed more easily and with higher accuracy than when such a program is used. Further, in the ticket gate device 1, the influence on the failure changes depending on the environment such as weather conditions such as air pressure and humidity. For example, depending on the environment, the susceptibility to ticket clogging changes, and the effect on the operational accuracy of electronic devices changes. Therefore, by including the environment information 7 in the input data and performing machine learning, it is possible to generate the learned model 10 that takes into consideration the complex interaction between the operation information 6 and the environment information 7 . Since the maintenance plan can be reviewed based on the highly accurate failure prediction, the possibility of failure of the ticket gate device 1 in actual operation can be reduced.

[Embodiment 1]
Next, as Embodiment 1, specific failure prediction of the ticket vending machine 1b and the fare adjustment machine 1c, which are the ticket gate devices 1, will be described with reference to FIGS. 1 to 8. FIG. In the following description, the ticket vending machine 1b will be mainly described as an example of the ticket vending machine 1b and the payment machine 1c.

[Trained model generation device]
First, a specific configuration example of the trained model generation device 3 and a specific example of a trained model generation flow will be described with reference to FIGS. 1 to 3 and FIGS. 4 to 8. FIG.

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 amount calculation unit 18 , and a feature amount extraction unit 19 .

The data communication unit 13 transmits and receives data to and from the ticket vending machine 1b, the AI 9, and the like via the Internet line 2. FIG. 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 in FIG. 6). #1).

Here, the ticket vending machine 1b and the fare adjustment machine 1c have more complicated internal structures than the ticket gate 1a. Further, the ticket vending machine 1b and the payment machine 1c are provided with the ticket issuing block 41, the bill block 42, the coin block 43, and the card block 44 as described above. being independent. Therefore, failure prediction may be performed for the ticket vending machine 1b and the payment machine 1c as a whole, but it is more appropriate to perform failure prediction independently for each of the ticket issuing block 41, bill block 42, coin block 43, and card block 44. is.

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

Although the predetermined period can be set arbitrarily, it can be, for example, one month. The model operation information 20 is information obtained from each of the four blocks of each ticket vending machine 1b and linked to each block of the ticket vending machine 1b. It consists of 2000 pieces of information (feature amounts). The model environment information 21 is information acquired from the ticket vending machine 1b and associated with each ticket vending machine 1b, and is composed of, for example, a total of 2000 pieces of information (feature amounts). Note that the model environment information 21 may also be acquired for each block in the same manner as the model operation information 20 .

In addition to model operation information 20, model environment information 21, and on-call data 8, the storage unit 15 stores input data 22 for learning, a first trained model 23, a second trained model 24, etc., which will be described later. Stores various data of The generation control unit 16 includes a processor such as a CPU or GPU, and controls operations of the trained model generation device 3 such as the model generation unit 17 and the feature amount calculation unit 18 .

Based on the control of the generation control unit 16, the feature amount calculation unit 18 calculates at least the model operating information 20 and the model environment information 21 from the acquired model operation information 20 and model environment information 21. A part of the operation information 20 is calculated to calculate the first feature amount 25 for each block (step #2 in FIG. 6).

For example, as shown in FIG. 7, the first feature value 25 is the utilization rate of the IC card per day calculated from the number of times each of the IC card, ordinary ticket, commuter ticket, and SF card in the operation information 6 is used. be done. Furthermore, as the first feature value 25, the total or average value of the usage rate of the IC card for 3 days or the total or average value for 7 days may be calculated. In addition, the differential value of the number of times the ordinary ticket is jammed in a predetermined period is calculated as the first feature amount 25 . Also, the first feature amount 25 may be the content of a failure in each block or a functional unit in each block.

Thus, the first feature quantity 25 of each block is configured by adding the calculated item and its value to the model operating information 20 and the model environment information 21 . As a result, the first feature quantity 25 consists of, for example, a total of 20000 pieces of information (feature quantity) in each block. The first feature value 25 of each block is information that may affect the failure of each block of the ticket vending machine 1b. It becomes possible to predict the failure of each block of the block with high accuracy.

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, the input data is the first feature amount 25 and the teacher data is the on-call data 8 (failure information) for each block. Some input data for learning 22 is input to the AI 9 for machine learning to generate a first trained model 23 (step #3 in FIG. 6). The on-call data 8 is information including on-call information performed when each block of the ticket vending machine 1b fails and repair results accompanying the on-call, and is accumulated for each on-call. Therefore, the on-call data 8 contains various information, and as shown in FIG. 8, the failure information extracted from the on-call data 8 includes at least an ID identifying the target ticket vending machine 1b and target block information. , the date and time of the on-call, the contents of the on-call, the failure part identified as a result of the repair, and the cause of the failure.

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 trained model 23 for each block, analyzes it, and determines the degree of influence of each feature quantity on failure prediction. Ask for Then, under the control of the generation control unit 16, the feature amount extraction unit 19 extracts, for example, 100 feature amounts from the 20000 feature amounts in descending order of influence on failure prediction for each block. The 100 feature values are set as the second feature value 26 (step #4 in FIG. 6).

Then, under the control of the generation control unit 16, the model generation unit 17 inputs learning input data 27, in which 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. machine learning to generate the second trained model 24 (step #5 in FIG. 6).

The first trained model 23 and the second trained model 24 are models used for predicting failure of the ticket vending machine 1b within a predetermined prediction period, for example, within seven 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 having such a decision tree structure evaluate a predetermined feature quantity (first feature quantity 25 or second feature quantity 26) at each branch point of the decision tree. , an evaluation value (failure probability) corresponding to the evaluation result is given to each branch point. Then, the failure prediction information 36 is obtained by summing the evaluation values along the branch of the decision tree. The first trained model 23 and the second trained model 24 may be ensemble models, such as XGBoost, Random Forest, LightGBM, CatBoost, and AdaBoost, in which a plurality of decision trees are provided in association.

In this way, from the operation information 6 and the environment information 7 acquired for each block, the feature values that may affect the failure of each block of the ticket vending machine 1b are selected, and the degree of influence on the failure is clearly identified. Then, the first feature amount 25 is generated by adding the feature amount. Then, the first trained model 23 is generated by using these first feature values 25 that 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 feature value is adequately reflected, and it is possible to accurately perform failure prediction of each block of the ticket vending machine 1b.

Furthermore, for each block, the first feature amount 25 is analyzed using the first trained model 23, the feature amount having a large degree of influence on the failure is extracted as the second feature amount 26, and these second feature amounts 26 as input data to generate a second trained model 24 . As a result, failure prediction can be performed using the second trained model 24 with higher failure prediction accuracy, and failure prediction of each block of the ticket vending machine 1b can be performed with higher accuracy.

[Failure prediction execution part]
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. 1 to 3 and FIGS.

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 .

The data communication unit 28 transmits and receives data to and from the ticket vending machine 1b, the trained model generation device 3, and the like via the Internet line 2. FIG. The model acquisition unit 29 acquires the first trained model 23 or the second trained model 24, which is a trained model used for failure prediction, for each block from the trained model generation device 3 via the data communication unit 28. and stored 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 operation information for prediction 34 and the environment information for prediction 35 are obtained by chronologically collecting the operation information 6 obtained from each block of the ticket vending machine 1b and the environment information 7 of the ticket vending machine 1b over a predetermined period (second period). It is the information Although the predetermined period can be set arbitrarily, it is, for example, the day before the failure prediction. The prediction control unit 32 includes a processor such as a CPU, and controls operations of the failure prediction unit 33 and the like. Note that the prediction environment information 35 may also be acquired for each block in the same manner as the prediction use information 34 .

Under 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 (FIG. 9). step #3), the failure prediction information 36 for each block is stored in the storage unit 31 and then output (step #4 in FIG. 9). The trained model is the second trained model 24 in this embodiment, but the first trained model 23 can also be used.

The failure prediction information 36 indicates the probability that each block of each ticket vending machine 1b will fail within a prediction period. If 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 of 0 or more and 1 or less for each block of the ticket vending machine 1b. The closer this value is to 1, the higher the probability of failure. When the second trained model 24 is composed of a decision tree, the failure prediction information 36 calculates the second feature amount 26 from the operation information for prediction 34 and the environment information for prediction 35, and calculates the second feature amount at each branch point. 26 evaluations are performed, and evaluation values are added each time a branch point is passed.

The failure prediction performed using the second trained model 24 was verified under the following conditions. A second trained model 24 was generated with a first period of 4 months and 19303 first feature quantities 25, 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 (percentage not to miss): 10%, and Recall (percentage not to miss): 70%. Therefore, according to the present embodiment, failures can be predicted with higher accuracy than when maintenance planning is performed based on conventional human judgment. It can be seen that the probability of failure during operation is dramatically reduced compared to the conventional technology.

Further, as the failure prediction information 36, a failure part that is predicted to fail in each block of each ticket vending machine 1b may be output together with its failure probability. For example, in each block of each ticket vending machine 1b, the top three parts with the highest failure probability are output.

In failure prediction, an evaluation value of each second feature quantity 26 is calculated. The failure probability for each part of each ticket vending machine 1b is obtained by summing up the evaluation values of the second feature quantity 26 predetermined to affect the failure of each part among the evaluation values of each second feature quantity 26. can be obtained by As the failure prediction information 36, for each block of the ticket vending machine 1b, a part with a high failure probability is output together with the failure probability.

In this way, in ticket issuing devices such as the ticket vending machine 1b and the payment machine 1c, by generating a learned model for each block of the ticket issuing block 41, bill block 42, coin block 43, and card block 44 and performing failure prediction, Even for a bidder with a complicated internal structure, it is possible to perform failure prediction specialized for each block and improve the prediction accuracy.

In addition, a trained model (first trained model 23, second trained model 24 ), it is possible to accurately predict the failure of each block of the ticket vending machine 1b, the payment machine 1c, and the like. In particular, failure prediction is performed using a second learned model 24 that is machine-learned using a second feature value 26 that is extracted in consideration of the magnitude of the effect on failure prediction. failure prediction can be performed with higher accuracy. As a result, based on the failure prediction information 36, the maintenance deadline for the block of the bid equipment that has a high probability of failure is clarified. becomes possible.

In addition, by outputting the parts with a high probability of failure for each block of the bid equipment, the parts to be maintained and inspected intensively are clarified, and maintenance and inspection can be performed efficiently.

[Embodiment 2]
Next, as a second embodiment, specific failure prediction of the ticket gate 1a, which is the exit ticket gate device 1, will be described with reference to FIGS. 1, 3, 5, 6, and 9-11.

[Trained model generation device]
First, a specific configuration example of the trained model generation device 3 and a specific example of a trained model generation flow will be described using FIGS. 5, 6, 10 and 11 while referring to FIGS. 1 and 3. .

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 amount calculation unit 18 , and a feature amount extraction unit 19 .

The data communication unit 13 transmits and receives data to and from the ticket gate 1a, the AI 9, and the like via the Internet line 2. FIG. The model data acquisition unit 14 acquires the model operation information 20 and the model environment information 21 of the ticket gate 1a received by the data communication unit 13, and the on-call data 8, and stores them in the storage unit 15 (step in FIG. 6). #1). The model operation information 20 and the model environment information 21 are information obtained by chronologically collecting the operation information 6 and the environment information 7 acquired from the ticket gate 1a over a predetermined period (first period). Although the predetermined period can be set arbitrarily, it can be, for example, one month. The model operation information 20 and the model environment information 21 are information acquired from each ticket gate 1a and associated with the ticket gate 1a, and are composed of, for example, 2000 pieces of information (feature amounts) in total. In addition to model operation information 20, model environment information 21, and on-call data 8, the storage unit 15 stores input data 22 for learning, a first trained model 23, a second trained model 24, etc., which will be described later. Stores various data of The generation control unit 16 includes a processor such as a CPU or GPU, and controls operations of the trained model generation device 3 such as the model generation unit 17 and the feature amount calculation unit 18 .

Based on the control of the generation control unit 16, the feature amount calculation unit 18 calculates at least a part of the model operation information 20 out of the model operation information 20 and the model environment information 21 to obtain the first feature amount 25. Calculate (step #2 in FIG. 6). For example, as shown in FIG. 10, the first feature value 25 is calculated from the number of times each of the IC card, ordinary ticket, commuter ticket, and SF card in the operation information 6 is used to calculate the usage rate of the IC card per day. be done. Furthermore, as the first feature value 25, the total or average value of the usage rate of the IC card for 3 days or the total or average value for 7 days may be calculated. In addition, the differential value of the number of times the ordinary ticket is jammed in a predetermined period is calculated as the first feature amount 25 . Thus, the first feature amount 25 is configured by adding the calculated item and its value to the model operating information 20 and the model environment information 21 . As a result, the first feature quantity 25 consists of, for example, 20000 pieces of information (feature quantity) in total. The first feature value 25 is information that may affect the failure of the ticket gate 1a, and by generating a large amount of this information and making composite judgments, the failure of the ticket gate 1a can be predicted with high accuracy. become.

The model generating unit 17 includes a processor such as a CPU or a GPU, and learning input data in which the input data is the first feature value 25 and the teacher data is the on-call data 8 (failure information) based on the control of the generation control unit 16. 22 is input to the AI 9 for machine learning to generate a first trained model 23 (step #3 in FIG. 6). The on-call data 8 is information including on-call information performed when the ticket gate 1a breaks down and repair results accompanying the on-call, and is accumulated for each on-call. Therefore, the on-call data 8 includes various types of information, and as shown in FIG. 11, the failure information extracted from the on-call data 8 includes at least an ID that identifies the target ticket gate 1a and the call that was on-call. It includes the date and time, the contents of the on-call, the failure part identified as a result of the repair, and the cause of the failure.

The feature amount extraction unit 19 inputs the first feature amount 25 to the first trained model 23 and analyzes it under the control of the generation control unit 16 to obtain the degree of influence of each feature amount on failure prediction. Then, under the control of the generation control unit 16, the feature amount extraction unit 19 extracts, for example, 100 feature amounts from the 20,000 feature amounts in descending order of influence on failure prediction, and extracts the extracted 100 feature amounts. Let the feature amount be the second feature amount 26 (step #4 in FIG. 6).

Then, under the control of the generation control unit 16, the model generation unit 17 inputs learning input data 27, in which the input data is the second feature amount 26 and the teacher data is the on-call data 8, to the AI 9 for machine learning. , to generate the second trained model 24 (step #5 in FIG. 6).

The first trained model 23 and the second trained model 24 are models used for predicting failure of the ticket gate 1a within a predetermined prediction period, for example, within seven 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 having such a decision tree structure evaluate a predetermined feature quantity (first feature quantity 25 or second feature quantity 26) at each branch point of the decision tree. , an evaluation value (failure probability) corresponding to the evaluation result is given to each branch point. Then, the failure prediction information 36 is obtained by summing the evaluation values along the branch of the decision tree. The first trained model 23 and the second trained model 24 may be ensemble models, such as XGBoost, Random Forest, LightGBM, CatBoost, and AdaBoost, in which a plurality of decision trees are provided in association.

In this way, from the operation information 6 and the environment information 7, the feature values that may affect the failure of the ticket gate 1a are selected, and these feature values are calculated so as to clarify the degree of impact on the failure. is added to generate the first feature quantity 25 . Then, the first trained model 23 is generated by using these first feature values 25 that 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 feature amount is adequately reflected, and it is possible to accurately perform failure prediction of the ticket gate 1a.

Furthermore, the first feature amount 25 is analyzed using the first trained model 23, the feature amount having a large influence on the failure is extracted, and is used as the second feature amount 26. These second feature amounts 26 are input data to generate the second trained model 24 . As a result, failure prediction can be performed using the second trained model 24 with higher failure prediction accuracy, and failure prediction of the ticket gate 1a can be performed with higher accuracy.

[Failure prediction execution part]
Next, a specific configuration example of the failure prediction executing section 4 and a specific example of the failure prediction flow will be described with reference to FIGS. 1 and 3 and FIGS.

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 .

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

Under the control of the prediction control unit 32, the failure prediction unit 33 inputs the operation information for prediction 34 and the environment information for prediction 35 into the learned model to perform failure prediction (step #3 in FIG. 9). After the information 36 is stored in the storage unit 31, it is output (step #4 in FIG. 9). The trained model is the second trained model 24 in this embodiment, but the first trained model 23 can also be used.

The failure prediction information 36 indicates the probability that each of the ticket gates 1a will fail within the prediction period. When the prediction period is 7 days, the probability of failure of the ticket gate 1a within 7 days is indicated by a value of 0 or more and 1 or less for each ticket gate 1a. The closer this value is to 1, the higher the probability of failure. When the second trained model 24 is composed of a decision tree, the failure prediction information 36 calculates the second feature amount 26 from the operation information for prediction 34 and the environment information for prediction 35, and calculates the second feature amount at each branch point. 26 evaluations are performed, and evaluation values are added each time a branch point is passed.

The failure prediction performed using the second trained model 24 was verified under the following conditions. A second trained model 24 was generated with a first period of 4 months and 19303 first feature quantities 25, 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 (percentage not to miss): 10%, and Recall (percentage not to miss): 70%. Therefore, according to the present embodiment, failures can be predicted with higher accuracy than when maintenance planning is performed based on conventional human judgment, and the ticket gate 1a can be realized when the failure prediction of the present embodiment is performed. It can be seen that the probability of failure during operation is dramatically reduced compared to the conventional technology.

Furthermore, as the failure prediction information 36, a failure part that is predicted to fail in each ticket gate 1a may be output together with the failure probability. For example, in each ticket gate 1a, the top three parts with the highest failure probability are output.

In failure prediction, an evaluation value of each second feature quantity 26 is calculated. The failure probability for each part of each ticket gate 1a is obtained by summing the evaluation values of the second feature quantity 26 predetermined to affect the failure of each part among the evaluation values of each second feature quantity 26. can be obtained by As the failure prediction information 36, a part with a high failure probability is output together with the failure probability for each ticket gate 1a.

In this way, a trained model (first trained model 23, second trained model 23, second trained Since failure prediction is performed using the model 24), failure prediction of the ticket gate 1a can be performed with high accuracy. In particular, by performing failure prediction using the second learned model 24 machine-learned using the second feature value 26 extracted in consideration of the magnitude of the effect on failure prediction, the failure of the ticket gate 1a Prediction can be made more accurately. As a result, based on the failure prediction information 36, the deadline for maintenance of the ticket gate 1a with a high probability of failure is clarified, and the period and timing of the maintenance plan can be reviewed to perform maintenance and inspection more appropriately. It becomes possible.

In addition, by outputting the parts with a high probability of failure for each ticket gate 1a, the parts to be maintained and inspected intensively are clarified, and maintenance and inspection can be performed efficiently.

As shown in FIG. 12, it can be seen that the inspection cost and the number of on-calls can be improved by performing the failure prediction in the first and second embodiments as compared with the management based on the conventional maintenance plan.

[Maintenance plan]
The failure prediction execution unit 4 in Embodiments 1 and 2 may further include a maintenance planning unit 37 . The maintenance planning section 37 first determines a maintenance plan for performing periodic maintenance and inspection on each exit ticket gate device 1 (each ticket gate 1a, each ticket vending machine 1b, and each fare adjustment machine 1c). Then, based on the failure prediction information 36, the maintenance planning unit 37 performs maintenance/inspection on each ticket gate 1a, each ticket vending machine 1b, and each fare adjustment machine 1c within the expected period of failure. change the maintenance plan so that

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

In this way, by changing the maintenance plan based on the failure prediction information 36, it is possible to perform maintenance/inspection of the exit ticket gate device 1 before failure while performing maintenance/inspection efficiently. failure can be suppressed.

Note that the maintenance planning unit 37 is not limited to the configuration provided in the failure prediction executing unit 4, and may be provided at an arbitrary location. For example, the maintenance planning section 37 may be configured to be connected to the Internet line 2 like the maintenance planning section 12 shown in FIG.

[Another embodiment]
(1) In each of the above-described embodiments, the configurations of the trained model generation device 3 and the failure prediction execution unit 4 are not limited to the case where they are divided into functional blocks as shown in FIG. As long as the functions of the execution unit 4 can be realized, the configuration of the functional blocks is arbitrary. For example, the functional blocks of the generation control unit 16, the model generation unit 17, the feature amount calculation unit 18, and the feature amount extraction unit 19 are not limited to such configuration, It may be composed of functional blocks obtained by subdividing each functional block. Further, in each of the above embodiments, part or all of the functions of the trained model generation device 3 and the failure prediction execution unit 4 may be integrated.

(2) In each of the above-described embodiments, the trained model generating device 3 may be configured to generate a trained model by itself and complete the process without transmitting the trained model to the failure prediction execution unit 4 . Conversely, the failure prediction execution unit 4 may perform failure prediction using a separately generated trained model instead of using the trained model generated by the trained model generation device 3 .

(3) In each of the above-described embodiments, part or all of the learned model generation method realized by the trained 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 are arbitrary. can be realized with the configuration of Part or all of the learned model generation method realized by the trained model generation device 3 and the failure prediction method realized by the failure prediction execution unit 4 may be implemented 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 a CPU or GPU included in the generation control unit 16 executes the program.

(4) In each of the above embodiments, the on-call data 8 may include only on-call information and may not include repair results. In this case, the on-call data 8 (failure information) consists of an ID for identifying the target exit ticket gate device 1, the date and time of the on-call, and the content of the on-call. Note that the on-call content may be only information indicating whether or not the on-call has been performed. It consists of the date and time of the call and the presence or absence of on-call.

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

(6) In each of the above-described embodiments, the configuration is not limited to performing failure prediction only when the prediction period is seven days. . For example, the prediction period is set to 1 month (30 days), 14 days, 7 days, and 3 days. Output the probability of failure within 7 days, the probability of failure within 3 days, and the probability of failure within 3 days. In this case, the first trained model 23 and the second trained model 24 are a trained model for finding the probability of failure within one month and a probability of failure within 14 days for each ticket gate device 1 or its block. , a trained model for obtaining the probability of failure within seven days, and a trained model for obtaining the probability of failure within three days. Then, the failure prediction execution unit 4 performs failure prediction using the four types of the first trained models 23 or the four types of the second trained models 24 for each exit ticket gate device 1 or each block thereof. Output prediction information 36 . For example, failure prediction is performed 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 with reference 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. do.

As a result, the probability of failure during a plurality of periods is shown, and the timing at which maintenance/inspection should be performed can be grasped more accurately. Along with this, the maintenance plan can be changed more accurately.

(7) In each of the above embodiments, the first trained model 23 and the second trained model 24 may be updated as failure prediction is performed. By inputting the prediction operation information 34 and the prediction environment information 35 used in the failure prediction performed in a predetermined period to the first trained model 23 or the second trained model 24 for machine learning , the first trained model 23 or the second trained model 24 is updated. At this time, the result of failure prediction using the prediction operation information 34 and the prediction environment information 35 and the result of maintenance/repair may be added to the teacher data.

In this way, by updating the learned model along with failure prediction, the accuracy of the learned model is improved, making it possible to perform more accurate failure prediction.

(8) In each of the above embodiments, the exit ticket gate device 1 may be configured without an environment 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 to the learned model 10 during failure prediction. Furthermore, other information may be used to generate the trained model 10, and any one of the operating information 6, the environmental information 7, and other information may be combined and used to generate the trained model 10. can be

The environmental information 7 may not have a large effect on failure prediction. In such a case, the exit ticket gate device 1 can have a simpler configuration by adopting a configuration that does not include an environment sensor.

(9) The exit ticket gate device 1, the trained model generation device 3, the failure prediction execution unit 4, and the maintenance planning unit 12 are not limited to being connected to the Internet line 2, and can exchange data with each other in any configuration. Any configuration is fine. For example, at least 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 allow transmission and reception via a dedicated line such as a LAN. It may be configured to exchange data.

(10) The failure information used to generate the trained 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 exit ticket gate device 1 may determine and create failure information and input it to the learned model generation device 3 via the Internet line 2 or the like, or the exit ticket gate device 1 may determine the failure by itself. Alternatively, fault information may be generated and input to the trained model generation device 3 via the Internet line 2 or the like.

(11) In the second embodiment, the ticket gate 1a is divided into a plurality of blocks, and failure prediction is performed for each block, as in the first embodiment. It is good also as a structure to be used.

Conversely, in the ticket vending machine 1b and the fare adjustment machine 1c, failure prediction may be performed for each ticket vending machine 1b or fare adjustment machine 1c as in the second embodiment, instead of for each block as in the first embodiment.

Furthermore, in the exit ticket gate device 1, the blocks are not limited to the configuration in which the blocks are divided into the ticket issuing block 41, the banknote block 42, the coin block 43, and the card block 44, but are divided into arbitrary blocks, and failure prediction is performed for each block. Also good.

The present invention is not limited to ticket gate devices installed at railway stations and other transportation stations, but also applies to ticket gate devices installed in various facilities such as theme parks. It can be applied to failure prediction systems, failure prediction programs, and trained models.

3 Trained model generation device 4 Failure prediction execution unit 6 Operation information 7 Environmental information 8 On-call data 9 AI
10 trained model 11 failure prediction information 12 maintenance planning unit 14 model data acquisition unit 15 storage unit 16 generation control unit 17 model generation unit 18 feature amount calculation unit 19 feature amount extraction unit 20 model operation information 21 model environment information 22 Input data for learning 23 First learned model (learned model)
24 Second trained model (learned model)
25 first feature quantity (feature quantity)
26 second feature quantity (feature quantity)
27 learning input data 29 model acquisition unit 30 prediction data acquisition unit 31 storage unit 32 prediction control unit 33 failure prediction unit 34 prediction operation information 35 prediction environment information 36 failure prediction information 37 maintenance planning unit

Claims (12)

A trained model generation device for generating a trained model used for predicting failures of bid equipment,
Receiving model operation information as operation information of the tendering device and model environment information as the environment information of the tendering device, which are acquired over time in each of the plurality of tendering devices in a predetermined first period; , a model data acquisition unit that receives failure information when a failure occurs in each of the bid devices during the first period;
a feature quantity calculation unit that generates a first feature quantity by calculating the model operating information and the model environment information;
First learning for outputting 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 value as input data and the failure information as teacher data. A trained model generation device, comprising: a model generation unit that generates a trained model. a feature quantity extraction unit that analyzes the first trained model and extracts a second feature quantity from the first feature quantity under a predetermined condition considering the degree of influence for outputting the failure prediction information; ,
When the operation information and the environment information are input, the model generation unit generates the failure prediction information of each of the bid devices by machine learning using the second feature as input data and the failure information as teacher data. 2. A trained model generating apparatus according to claim 1, wherein a second trained model to be output is generated, and failure prediction of said bid device is performed using said second trained model. the bidding device includes a plurality of blocks;
the operation information for the model is acquired for each of the blocks;
A first feature is generated for each block,
the first trained model is generated for each of the blocks;
a second feature is generated for each block;
3. The trained model generating apparatus according to claim 2, wherein said second trained model is generated for each block. A failure prediction device that uses a trained model to predict failures of bidding equipment,
Calculating model operation information, which is operation information of the tendering devices, and model environment information, which is environmental information of the tendering devices, which are acquired over time in each of the plurality of tendering devices in a predetermined first period. The operation information and the environment information are input by machine learning using the first feature amount generated by the above as input data and failure information when each of the above-mentioned bidding devices fails during the above-mentioned first period as teacher data. Then, a model acquisition unit for acquiring the learned model generated to output failure prediction information of each of the bid devices;
a prediction data acquisition unit that receives the prediction operation information as the operation information and the prediction environment information as the environment information, which are acquired over time in each of the plurality of bidder devices in a predetermined second period;
a failure prediction unit that inputs the operation information for prediction and the environment information for prediction to the learned model acquired through the model acquisition unit and outputs the failure prediction information for each of the bid devices. Failure prediction device. The trained model generation device according to claim 1;
a prediction data acquisition unit that receives the prediction operation information as the operation information and the prediction environment information as the environment information, which are acquired over time in each of the plurality of bidder devices in a predetermined second period;
a failure prediction unit that inputs the operation information for prediction and the environment information for prediction to the first trained model generated by the trained model generation device and outputs the failure prediction information for each of the devices to be bid; failure prediction system. The trained model generation device according to claim 2;
a prediction data acquisition unit that receives the prediction operation information as the operation information and the prediction environment information as the environment information, which are acquired over time in each of the plurality of bidder devices in a predetermined second period;
a failure prediction unit that inputs the operation information for prediction and the environment information for prediction to the second trained model generated by the trained model generation device, and outputs the failure prediction information for each of the devices to be bid; failure prediction system. The trained model generation device according to claim 3;
Predictive operation information, which is the operation information obtained over time in a predetermined second period in each of the blocks of the plurality of the selling devices; a prediction data acquisition unit that receives the prediction environment information, which is the environment information that is acquired in an objective manner;
Failure prediction for inputting the operation information for prediction and the environment information for prediction into the second trained model generated by the trained model generation device and outputting the failure prediction information for each block of the device to be bid A failure prediction system comprising: A failure prediction program that uses a trained model to predict failures of bid equipment,
Calculating model operation information, which is operation information of the tendering devices, and model environment information, which is environmental information of the tendering devices, which are acquired over time in each of the plurality of tendering devices in a predetermined first period. The operation information and the environment information are input by machine learning using the first feature amount generated by the above as input data and failure information when each of the above-mentioned bidding devices fails during the above-mentioned first period as teacher data. then obtaining the trained model generated to output failure prediction information for each of the bid equipment;
a step of receiving prediction-use operating information as the operating information and prediction-use environment information as the environment information, which are acquired over time in each of the plurality of bidder devices in a predetermined second period;
A failure prediction program for causing a computer to execute a step of inputting the operation information for prediction and the environment information for prediction to the acquired learned model, and outputting the failure prediction information for each of the bid devices. A failure prediction program that uses a trained model to predict failures of bid equipment,
Receiving model operation information as operation information of the tendering device and model environment information as the environment information of the tendering device, which are acquired over time in each of the plurality of tendering devices in a predetermined first period; , a step of receiving failure information when a failure occurs in each of said bidding devices during said first period;
generating a first feature amount by calculating the model operating information and the model environment information;
First learning for outputting 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 value as input data and the failure information as teacher data. generating a finished model;
a step of receiving prediction-use operating information as the operating information and prediction-use environment information as the environment information, which are acquired over time in each of the plurality of bidder devices in a predetermined second period;
A failure prediction program for causing a computer to execute a step of inputting the operation information for prediction and the environment information for prediction into the first trained model and outputting the failure prediction information for each of the bid devices. A failure prediction program that uses a trained model to predict failures of bid equipment,
Receiving model operation information as operation information of the tendering device and model environment information as the environment information of the tendering device, which are acquired over time in each of the plurality of tendering devices in a predetermined first period; , a step of receiving failure information when a failure occurs in each of said bidding devices during said first period;
generating a first feature amount by calculating the model operating information and the model environment information;
First learning for outputting 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 value as input data and the failure information as teacher data. generating a finished model;
a step of analyzing the first trained model and extracting a second feature quantity from the first feature quantity under a predetermined condition considering the degree of influence for outputting the failure prediction information;
Second learned for outputting the failure prediction information of each of the bid devices when the operation information and the environment information are input by machine learning using the second feature as input data and the failure information as teacher data. generating a model;
a step of receiving prediction-use operating information as the operating information and prediction-use environment information as the environment information, which are acquired over time in each of the plurality of bidder devices in a predetermined second period;
A failure prediction program for causing a computer to execute a step of inputting the operation information for prediction and the environment information for prediction into the second trained model and outputting the failure prediction information for each of the bid devices. A trained model for causing a computer to output failure prediction information for the ticketing device based on predictive operation information and predictive environment information acquired from the ticketing device,
It consists of a decision tree consisting of multiple branch points arranged in a tree structure,
The predictive operating information is operating information acquired over time by the bidding device over a predetermined period, and the predictive environment information is environmental information acquired over time by the bidding device over a predetermined period,
The operating information for prediction and the environment information for prediction are input, a plurality of feature amounts are calculated from the operating information for prediction and the environment information for prediction, and the feature amount corresponding to each of the branch points is calculated. An evaluation value corresponding to the feature amount is calculated, a traveling direction is determined, and the progression to the next branch point is repeated, and the evaluation values calculated at the advanced branch points are added up. A trained model that allows a computer to function to calculate and output failure prediction information. 12. The trained model according to claim 11, wherein a plurality of said decision trees are combined.

Images