JP7232673B2 - TRAIN OPERATION CONTROL DEVICE AND TRAIN OPERATION CONTROL METHOD - Google Patents

Description

The present invention relates generally to controlling train operations.

Real-time wireless transmission of operating status data, such as train position, speed, and equipment status, from on-board (train) to ground (ground server) to help streamline operation management and maintenance. be able to. For example, with regard to grasping the position of trains in operation management, conventionally, except for wireless train control such as CBTC, train positions could only be grasped in units of track circuits. By using terrestrial distribution, the train position can be grasped continuously. As a result, it becomes possible to take more detailed measures when rescheduling trains or the like. With regard to maintenance, for example, the condition of train equipment can be grasped on the ground in real time while the train is in operation.

Patent Literature 1 discloses a technology related to operation re-planning when a train is delayed using the above-described operation status data. a train operation monitoring unit that calculates a delay time of a train; and a storage unit that stores a delay recovery rule for determining a recovery travel section for shortening the travel time in the operation plan in order to recover the delay time among a plurality of travel sections. and an arrival/departure time correction unit that, when the train is delayed, corrects the travel time in the recovery travel section according to the delay recovery rule, and re-plans the target arrival/departure time of the operation plan. device".

Japanese Patent Application Laid-Open No. 2018-7402

According to the technique described in Patent Document 1, when the presence or absence of a train delay and its delay time are detected from the driving situation data, the operation plan is rescheduled.

It is desirable that train delays do not occur. However, according to the technique described in Patent Document 1, re-planning is performed triggered by the occurrence of a train delay. Therefore, the technique described in Patent Document 1 cannot reduce the frequency of train delays.

The train operation control device stores past operation performance data (data indicating the past operation performance of each of a plurality of trains on the target route), delay presence/absence performance data (data indicating the relationship between time, train, station interval, and delay presence/absence), and Based on the above, learning data (learning data on whether or not a delay occurs) is generated. The device builds a prediction model (a model for detecting signs of delays) through machine learning using learning data, and collects the most recent operational performance data (the most recent operational performance of each of multiple trains on the target route). data shown) is input to the predictor model to detect the presence or absence of a predictor of delay for each of the plurality of trains.

According to the present invention, the presence or absence of a sign of train delay is detected, and the possibility of delay can be reduced by using the detection result.

1 shows an example of a system configuration of a train operation control device according to Embodiment 1 of the present invention; An example of a teaching method in the command teaching section is shown. Another example of the teaching method in the command teaching section is shown. An example of a plurality of operation patterns prepared by the operation pattern switching unit is shown. An example of a run curve is shown. An example of a comparison of planned run curves and actual run curves is shown. Another example of a comparison of planned run curves and actual run curves is shown. 4 shows a configuration example of a delay sign detection unit; The example of the system configuration|structure of the train operation control apparatus which concerns on Example 2 of this invention is shown. The example of the system configuration|structure of the train operation control apparatus which concerns on Example 3 of this invention is shown. An example of a system configuration of a train operation control device according to a fourth embodiment of the present invention is shown.

In the following description, an "interface device" may be one or more interface devices. The one or more interface devices may be one or more of the same type of communication interface device or two or more of different types of communication interface devices.

Also, in the following description, "memory" may be one or more memory devices, typically a main memory device. At least one memory device in the memory may be a volatile memory device or a non-volatile memory device.

Also, in the following description, a "persistent storage device" is one or more persistent storage devices. A permanent storage device is typically a non-volatile storage device (for example, an auxiliary storage device), and specifically, for example, an HDD (Hard Disk Drive) or an SSD (Solid State Drive).

Also, in the following description, "storage" may be at least memory of memory and persistent storage.

Also, in the following description, a "processor" is one or more processor devices. The at least one processor device is typically a microprocessor device such as a CPU (Central Processing Unit), but may be another type of processor device such as a GPU (Graphics Processing Unit). At least one processor device may be single-core or multi-core. At least one processor device may be a processor core. At least one processor device may be a broadly defined processor device such as a hardware circuit (for example, FPGA (Field-Programmable Gate Array) or ASIC (Application Specific Integrated Circuit)) that performs part or all of processing.

In addition, in the following description, the function may be described using the expression “kkk unit”, but the function may be realized by executing one or more computer programs by a processor, or may be realized by executing one or more computer programs. It may be realized by the above hardware devices (for example, FPGA or ASIC), or may be realized by a combination thereof. The description of each function is an example, and multiple functions may be combined into one function, or one function may be divided into multiple functions.

Several embodiments are described below with reference to the drawings.

In the present embodiment, it is assumed that train operation status data is distributed to the ground server in real time by wireless communication. The train operation control device according to the present embodiment creates a model for detecting signs of delay using past operational performance data obtained from a ground server, and inputs the most recent operational performance data obtained from the ground server into the model. By doing so, the presence or absence of a sign of delay is detected. In this embodiment, model creation requires learning data including data with and without delay records. In the present embodiment, the train operation control device classifies at least a part of the past operation result data into those with delay results and those without delay results by judging whether or not there is a delay track record by comparing the planned run curve and the actual run curve, Generate training data containing classified data. The presence or absence of a sign of delay may be taught to both or one of a dispatcher and a train driver, or may be used to switch the operation pattern of an automatically operated train.

First, the structure of the train operation control apparatus 100 is demonstrated using FIG. Train operation control device 100 is assumed to be a device having an interface device (for example, an interface device for communication with a train), a storage device, and a processor connected to the interface device and storage device.

The train operation control device 100 acquires operation status data 151 from the train 101 and outputs to the train 101 data 157 on whether or not there is a sign of delay with respect to the train and an operation pattern command 156 . Here, in FIG. 1, trains on the route are represented by one train as a train 101 for the sake of convenience. An operation pattern command 156 is output.

The components of the train operation control device 100 are a ground server 102 , a learning data generation unit 103 , a delay sign detection unit 104 , a command teaching unit 107 , an operation pattern switching unit 105 and a sign transmission unit 106 . The ground server 102 may exist outside the train operation control device 100 . Command teaching unit 107 , driving pattern switching unit 105 , and portent transmission unit 106 are examples of functions that use delay portent presence/absence data 155 . Some of the functions of command teaching unit 107, driving pattern switching unit 105, and sign transmitting unit 106 may be omitted. In place of or in addition to the function of at least one of command teaching unit 107, driving pattern switching unit 105, and sign transmission unit 106, another function using delay sign presence/absence data 155 may be provided. .

The ground server 102 accumulates driving situation data 151 wirelessly distributed from the train 101 in real time. Here, "real-time delivery" means that data is sent at any time while the train is in operation. included in On the other hand, the method of acquiring and storing the operating status data 151 stored in the train 101 at a specific location such as a station or inspection area (location where the train stops for a certain period of time or more) is not included in the “real-time distribution”.

Information included in the operating status data 151 is at least the train number, time and train position of the train 101 . As the driving condition data 151, a method of using data of a driving condition recording device (a device for recording the driving condition) installed in the train by law (or for other reasons) can be considered. Alternatively, the train position may be an on-track position recognized by a GPS (Global Positioning System) device included in a device fixed on the train or in a portable terminal brought into the train by the crew member in charge.

Past driving performance data 152 is transmitted from the ground server 102 to the learning data generation unit 103 . Inside the learning data generation unit 103 , learning data 154 to be input to the delay sign detection unit 104 is generated. The learning data 154 is a data set obtained by dividing the past driving performance data 152 into two groups according to the presence or absence of delay performance. The internal configuration and processing of the learning data generation unit 103 will be described later.

The delay portent detection unit 104 receives the learning data 154 generated by the learning data generation unit 103 as input, and internally generates a portent model (model for delay portent detection). Then, the delay predictor detection unit 104 calculates the presence or absence of a delay predictor by inputting the most recent operation performance data 153 acquired from the ground server 102 into the predictor model, and calculates the presence or absence of a delay predictor, which is data indicating the calculated presence of a delay predictor. Data 155 is output. The delay predictor presence/absence data 155 is data in which the presence or absence of a delay predictor is assigned to each predetermined section in the route. The configuration and internal processing of the delay sign detection unit 104 will be described later. Delay predictor presence/absence data 155 is transmitted to command teaching unit 107 , driving pattern switching unit 105 , and predictor transmission unit 106 .

The command teaching unit 107 is an example of a predictor output unit, processes the received delay predictor presence/absence data 155, generates teaching data that is data to be taught to the operation commander, and outputs the teaching data. Specific examples include a method of displaying tabular data indicating the presence or absence of a predictor of delay for each predetermined section, and a method of displaying data indicating the presence or absence of a predictor of delay in different colors on a route map. In the following explanation, the default section is between stations, but one section does not need to correspond to one station, and it does not matter if it includes multiple stations or the section boundary exists at a position other than the station. . According to the example of FIG. 2, the teaching data is in tabular form and includes line name, station interval, direction (up, down), delay sign detection state, and delay sign detection time, and the teaching data is displayed. According to the example of FIG. 3, the teaching data is data in the form of a route map, and the presence or absence of delay sign detection between stations is displayed separately for uplink and downlink by color. A dispatcher who has confirmed information indicated by such teaching data can, for example, perform at least one of the following.
・For manually operated trains running between stations where there are signs of delay or in the vicinity, call attention not to get too close to preceding trains.
・For automated trains running between stations where there is a sign of delay or in the vicinity, command switching of operation patterns (target speed patterns between stations).

Based on the received delay predictor presence/absence data 155, the operation pattern switching unit 105 selects an operation pattern for an automatically operated train (a train other than an automatically operated train may be used) that runs between stations with a delay predictor or in the vicinity thereof. An operation pattern command 156 for commanding the switching of is transmitted. The operation pattern is switched by the train in response to the operation pattern command 156 . As an example of the operation pattern switching method, a method of preparing a plurality of operation patterns in advance and selectively using them can be considered. Examples of the plurality of operation patterns include energy-saving operation and delay suppression operation. According to the example of FIG. 4, there are a plurality of driving patterns even when traveling between the same stations in the same traveling time. Therefore, if there is a sign of delay, the driving pattern that is less likely to cause a delay should be selected. is desirable. According to the example of FIG. 4, the delay suppression operation pattern lowers the speed when entering the station 2 instead of increasing the maximum speed as compared with the energy-saving operation pattern. By operating in this way, it is possible to reduce the possibility of an outboard stop due to the influence of a preceding train located near station 2, and to suppress the spread of turbulence.

The predictor transmission unit 106 is an example of a predictor output unit. Based on the received information about the presence or absence of delay predictors 155, the predictor transmission unit 106 issues an anti-train delay predictor to a manually operated train traveling between stations with a predictor of delay or in the vicinity thereof. Presence/absence data (data indicating no sign of delay and data for trains) 157 is transmitted. The driver, who received the information indicated by the data 157 as a notification that there is a sign of delay, will try to avoid unnecessary external stops and external decelerations, such as not to approach the preceding train excessively. Delays can be suppressed.

The above is the description of the configuration of the train operation control device 100 . As described above, both the past operational performance data 152 and the latest operational performance data 153 are data obtained from the ground server 102 that obtains the operational status data indicating the operational status of the train 101 from the train 101 in real time by wireless communication. (for example, data acquired in real time). Both the past driving performance data 152 and the latest driving performance data 153 are based on the driving status data acquired by the ground server 102 for each of a plurality of trains (for example, all trains) on the target route.

Next, the internal configuration and processing of the learning data generation unit 103 will be described with reference to FIGS. 1, 5, 6 and 7. FIG.

The components of the learning data generation unit 103 are an actual run curve management unit 111 , a planned run curve management unit 112 , a delayed performance detection unit 113 , and a learning data management unit 114 .

The learning data generation unit 103 downloads the past driving performance data 152 from the ground server 102 at a predetermined timing, performs internal processing to be described later, and then transmits the learning data 154 to the delay sign detection unit 104 . Here, the "predetermined timing" is the timing at which the driving status data 151 is accumulated in the ground server 102 to the extent that it is possible to determine the actual delay of the operation. ).

The actual run curve management unit 111 downloads the driving performance past data 152 for the new period from the ground server 102 at a predetermined timing. Here, the "new period" indicates the period from the previous download timing until now. Then, the track record run curve management unit 111 divides the acquired past track record data 152 for each station, and outputs track record curve 161 as a curve represented on a two-dimensional plane of "position" and "time". FIG. 5 shows an example performance run curve 161 . FIG. 5 shows an example of a train departing from station A and arriving at station B, but the actual run curve 161 contains data in a similar format regarding both the inbound and outbound directions between each station on the target route. good.

The planned run curve management unit 112 manages (for example, holds) a planned run curve for each station. The format of the planned run curve may be the same format as in FIG. The original data for the planned run curve includes the prescribed operating curve and planned timetable determined by each railway operator. In the operation curve, the relationship between "position" and "time" when traveling in a standard operation method for each station is represented by a curve on a two-dimensional plane. By combining this data with the departure/arrival times of the planned timetable, it is possible to prepare data in the same format as in FIG. 5 for the planned run curve. The planned run curve management unit 112 outputs a planned run curve between necessary stations as a planned run curve 162 in accordance with a request from the delay performance detection unit 113 .

After acquiring the actual run curve 161 from the actual run curve management unit 111, the actual delay detection unit 113 requests the planned run curve between the corresponding train and station from the planned run curve management unit 112 and acquires it. The actual delay detection unit 113 compares the actual run curve 161 and the planned run curve 162 regarding travel between stations on the target route. Then, the delay performance detection unit 113 generates delay presence/absence performance data 163 based on the comparison result, and transmits the delay presence/absence performance data 163 to the learning data management unit 114 . For example, the delay record detection unit 113 generates the delay presence/absence record data 163 when it can be determined that a delay has occurred by comparing the comparison result with a predetermined determination condition.

FIG. 6 shows an example comparison of actual run curve 161 and planned run curve 162 . The dashed line in FIG. 6 is the planned run curve, and the solid line is the actual run curve. In this example, the planned run curve and the actual run curve match up to train a, but the disturbance starts from train b. It can also be seen that the turbulence spreads to subsequent trains after train c. Here, as an example of a condition for judging the occurrence of a delay, it is assumed that "the difference in the train passing time between stations on the track exceeds the train interval in the planned timetable". In FIG. 6, it is determined that the train e is delayed at time T1. Also, although not shown, it is assumed that train f and train g are also determined to be delayed. At this time, the delay presence/absence actual data 163 is "between stations: station A→station B, train e", "between stations: station A→station B, train f", and "between stations: station A→station B, train g". set. The delay presence/absence performance data 163 may further include time. That is, the delay presence/absence record data 163 may be data indicating the relationship between time, train, station interval, and delay presence/absence.

In this embodiment, for each train, "occurrence of a delay" means that the difference between the time when the train is supposed to be at a certain position and the time when the train is actually at the certain position is exceeds a threshold determined based on the unit interval of (for example, if the diamond is every 15 seconds, the threshold is "15"). The unit interval of the timetable is a technical element unique to train operation control, and the occurrence of delays is defined according to such elements.

FIG. 7 shows another example of comparing actual run curves 161 and planned run curves 162 . In this example, as in FIG. 6, the planned run curve and the actual run curve match up to train a, but disturbance starts from train b. Although the following trains have been affected, train g has almost returned to running according to the planned run curve. In the example of FIG. 7, no delay is recognized based on the conditions for judging the occurrence of delay. At this time, no information is set in the delay presence/absence performance data 163 .

It goes without saying that the definition of the condition for judging the occurrence of a delay is not limited to the above example, and any condition based on the difference in position or shape between the planned run curve and the actual run curve may be used.

The learning data management unit 114 extracts a part of the past driving performance data 152 based on the delay presence/absence performance data 163, and generates and manages two data groups for each station and each direction. The first is a data group with a track record of delay, and the second is a group of data without a track record of delay. The learning data 154 includes these data groups.

The data included in the data group with delay record is the operation record for each of one or more trains that run differently than planned among one or more trains that run ahead of the train that started to experience delays. , and is composed of, for example, operation performance past data relating to a predetermined number of trains before the occurrence of a delay. In FIG. 6, since the delay presence/absence record data 163 includes "between stations: station A→station B, train e" (that is, to indicate that the train that started to be delayed is train e), there is a delay record. is composed of data of trains a to d, for example. That is, the data included in the data group with a track record of delays is not the data of trains that actually experienced delays, but the data of trains that have track records of delays.

The data included in the data group without delay record is the data with record of delay from the past operation record data 152, and the data of the train determined to have a delay (one or more trains that have been delayed). Data indicating the driving performance for each) is excluded (for example, data of trains running ahead of train a, and data illustrated in FIG. 7, among the data illustrated in FIG. 6) .

The learning data management unit 114 outputs the learning data 154 including the data group with the delay record and the data group without the delay record to the delay sign detection unit 104 .

The above is the description of the internal configuration and processing of the learning data generation unit 103 .

Next, the configuration and internal processing of the delay sign detection unit 104 will be described with reference to FIG.

The components of the delay portent detection unit 104 are a machine learning logic unit 801 and a portent model 802 . The machine learning logic unit 801 receives the learning data 154 and outputs model definition parameters 811 for defining the predictor model 802 (that is, constructs (eg, generates or updates) the predictor model 802. The predictor model 802 is , the characteristics of which are determined by the model definition parameters 811, and upon receiving the most recent driving performance data 153, the delay prediction presence/absence data 155 is output.

The internal logic of the machine learning logic unit 801 and the predictor model 802 differs depending on the delay predictor detection method. Pattern matching based on image recognition will be described as an example of a technique for detecting a sign of delay. In this method, a run curve group (see FIG. 5) drawn on a plane with time and position as two axes is recognized as one image, and it is determined whether or not the image has a sign of delay. In this case, a neural network using deep learning is a representative technique and model that can be used as the machine learning logic unit 801 and the predictive model 802 in the same way as handwritten characters and illustrations are recognized.

Model defining parameters 811 are the weights and thresholds given to each neuron in the neural network. The machine learning logic unit 801 uses the learning data 154, that is, a data group with delay results and a data group without delay results, as teacher data, and defines a model so that these two types of data groups can be accurately classified. Determine the parameters 811, namely the weights and thresholds for each neuron. A neural network configured by applying the weights and thresholds is the predictor model 802 , which receives the most recent driving performance data 153 as input and outputs the presence or absence of a predictor of delay 155 .

Needless to say, the delay sign detection method is not limited to this example. In the case of any delay sign detection method, every time the learning data 154 is obtained from the learning data generation unit 103, the parameter tuning of the sign model 802 is performed through the machine learning logic unit 801 to improve the accuracy of the delay sign detection. I can let you. For example, the machine learning logic unit 801 may determine the model definition parameters 811 using part of the training data 154 and tune the model definition parameters 811 using the rest of the training data 154 .

In addition, when the learning data 154 is insufficient, such as when applying to a new route, a data group with a delay record and a data group without a delay record artificially created by simulation etc. are used as teacher data for machine learning. It can also be input to the logic unit 801 to generate the initial form of the prognostic model 802 .

The above is the description of the configuration and internal processing of the delay sign detection unit 104 .

According to the present embodiment, the determination of the presence or absence of delay results, which is necessary for generating the learning data 154, is performed by comparing the planned and actual run curves. In other words, since it is possible to evaluate the deviation between the plan and the actual performance, including the time when the train is running between stations other than the departure and arrival at the station, it is possible to determine the presence or absence of delays more flexibly than the method of determining the delay based on the deviation of the departure and arrival times at the station. I can handle it.

The above is the description of the first embodiment.

Example 2 will be described below. At that time, the points of difference from the first embodiment will be mainly described, and the explanations of the points in common with the first embodiment will be omitted or simplified.

The main difference between this embodiment and the first embodiment is that, in order to obtain the delay presence/absence actual data 163, instead of comparing the planned run curve and the actual run curve, the planned timetable and the actual timetable are compared. be.

First, the configuration of the train operation control device 900 will be explained using FIG.

The train operation control device 900 includes a learning data generation unit 903 instead of the learning data generation unit 103 .

Next, the internal configuration and processing of the learning data generation unit 903 will be described with reference to FIG.

The learning data generation unit 903 includes an actual timetable management unit 911 , a planned timetable management unit 912 , a delay performance detection unit 913 , and a learning data management unit 114 . The learning data generation unit 903 downloads the past driving performance data 152 from the ground server 102 at a predetermined timing, performs internal processing to be described later, and then transmits the learning data 154 to the delay sign detection unit 104 .

The track record management unit 911 downloads the driving record past data 152 for the new period from the ground server 102 at a predetermined timing. Then, the actual timetable management unit 911 extracts the actual values of the departure time and the arrival time for each train traveling between stations from the acquired operation history data 152 , and outputs them to the delay actual time detection unit 913 as the actual timetable 961 .

The planned timetable management unit 912 manages the planned values of departure time and arrival time for each train traveling between stations. The planned timetable management unit 912 outputs a planned timetable between necessary stations as a planned timetable 962 in accordance with a request from the actual delay detection unit 913 .

When the actual delay detection unit 913 acquires the actual timetable 961 from the actual timetable management unit 911 , it acquires the planned timetable between the corresponding train and station from the planned timetable management unit 912 . The actual delay detection unit 913 compares the actual timetable 961 and the planned timetable 962 regarding travel between stations on the target route. Then, the delay performance detection unit 913 generates delay presence/absence performance data 163 based on the comparison result, and transmits the delay presence/absence performance data 163 to the learning data management unit 114 . For example, the delay record detection unit 113 generates the delay presence/absence record data 163 when it can be determined that a delay has occurred by comparing the comparison result with a predetermined determination condition.

Determination of delay occurrence in the delay record detection unit 913 is made based on the difference between the planned timetable 962 and the actual timetable 961 regarding the departure time and/or arrival time between the corresponding train and station. The threshold value of the difference determined as the occurrence of delay can be set variously depending on the railway operator and the route. For example, for a route whose arrival and departure times are scheduled in increments of 15 seconds, 15 seconds may be set as the threshold. When the delay record detection unit 913 determines that there is a record of delay occurrence, a set of inter-station information and train number is output to the learning data management unit 114 as the delay presence/absence record data 163 . If it is not determined that there is a track record of delay occurrence, no information may be set in the delay presence/absence track record data 163 .

The above is the description of the internal configuration and processing of the learning data generation unit 903 .

According to this embodiment, the determination of the presence or absence of actual delays, which is necessary for generating the learning data 154, is performed by comparing the planned and actual departure/arrival times. With this method, it is possible to determine whether or not there is a delay based only on the station arrival/departure timing. Therefore, it is possible to reduce the calculation load of the delay record presence/absence determination process. In addition, even if the acquisition frequency of the driving situation data 151 is low outside the vicinity of the station due to the radio wave condition near the target line, and it is difficult to compare the run curves as in the first embodiment.

The above is the description of the second embodiment.

Example 3 will be described below. At that time, the points of difference from the first and second embodiments will be mainly described, and the explanations of the points in common with the first and second embodiments will be omitted or simplified.

The main difference between this embodiment and Embodiments 1 and 2 is that instead of comparing the planned run curve and the actual run curve and the comparison between the planned timetable and the actual timetable in order to obtain the delay presence/absence actual data 163, , the use of a database on the occurrence record of delay events in past operations.

First, the configuration of the train operation control device 1000 will be described with reference to FIG. 10 .

The train operation control device 1000 includes a learning data generation unit 1003 instead of the learning data generation unit 103 (and 903).

Next, the internal configuration and processing of the learning data generation unit 1003 will be described with reference to FIG.

The components of the learning data generation unit 1003 are the delay record database 1013 and the learning data management unit 114 . The learning data generation unit 1003 downloads the past driving performance data 152 from the ground server 102 at a predetermined timing, performs internal processing to be described later, and then transmits the learning data 154 to the delay sign detection unit 104 .

The delay record database 1013 is a database summarizing the record of occurrence of delay events in past operations, and stores whether or not there was a record of delay occurrence with respect to each station running of each train. The database is created externally, such as by an operation management device, and provided to the train operation control device 1000 . The delay record database 1013 makes a set of inter-station information and a train number for trains with a delay record, and outputs the set as delay presence/absence record data 163 to the learning data management unit 114 . The delay record database 1013 may be a database that indicates the relationship between times, trains, interstations, and the presence or absence of delays.

The above is the description of the internal configuration and processing of the learning data generation unit 1003 .

According to this embodiment, the determination of the presence or absence of a delay record required for generating the learning data 154 is performed by the delay record database 1013 provided from the outside. This method does not require comparison of run curves and departure/arrival times inside the learning data generation unit 1003, and therefore can reduce the computational load of the delay record presence/absence determination processing. In addition, since the contents of the delay record database 1013 can be edited artificially, it is possible to arbitrarily manipulate the cases that should be included in the delay record and the cases that should not be included, so that the learning data 154 can be created more flexibly. It is possible.

The above is the description of the third embodiment.

Example 4 will be described below. At that time, the points of difference from the first to third embodiments will be mainly described, and the explanations of the points in common with the first to third embodiments will be omitted or simplified.

The main difference between this embodiment and Embodiments 1 to 3 is that the operation status data of the train is stored at a specific location such as a station or inspection area is stopped), the amount accumulated in the train until reaching the location is uploaded to the ground server. Data on the most recent driving conditions are obtained via the ground equipment of the signal protection equipment. In this example, any of the methods exemplified in Examples 1 to 3 can be used to create the predictive model.

First, the configuration of the train operation control device 1100 will be described with reference to FIG. 11 .

The train operation control device 1100 acquires the latest operation performance data 153 from the signal protection device wayside facility 1108 . Signal security equipment ground equipment 1108 is an existing equipment that constantly receives train position 1158 from train 101 for signal control. Here, in FIG. 11, the trains on the route are represented by one train as the train 101 for the sake of convenience. Acquisition is performed, and the most recent operating performance data 153 indicates the most recent performance for all trains concerned.

The ground server 102 accumulates driving situation data 151 acquired from the train 101 . Here, the driving situation data 151 is uploaded to the ground server 102 at a specific place such as a station or an inspection area, which has been accumulated in the train until it reaches that place. Compared to the real-time data sharing described in Examples 1 to 3, this embodiment has a narrower range for communication between ground vehicles, and can be done with a narrow-area communication facility such as Wifi (registered trademark). Advantageous.

In the train operation control device 1100, the internal configuration and processing of the learning data generation unit 1103 are the learning data generation unit 103 of the first embodiment, the learning data generation unit 903 of the second embodiment, and the learning data generation of the third embodiment. Any of the parts 1003 can be used.

According to this embodiment, the operation status data 151 is collectively downloaded at stations and maintenance areas, and the most recent operation performance data 153 is obtained from the existing signal protection device ground equipment 1108 . Therefore, it is not necessary to exclusively deliver the driving status data 151 for generating the past driving performance data 152 and the latest driving performance data 153 in real time, and it is possible to reduce communication equipment and communication costs.

The above is the description of the fourth embodiment.

Although several embodiments have been described above, these are examples for explaining the present invention, and are not meant to limit the scope of the present invention only to these embodiments. The present invention can also be implemented in various other forms. For example, two or more of Examples 1 to 4 may be combined. Specifically, for example, the train operation control device may further include a data selector. The learning data generation unit uses, as the delay presence/absence actual data 163, the first type data obtained by comparing the planned run curve and the actual run curve, the second type data obtained by comparing the planned timetable and the actual timetable, It may be possible to acquire two or more types of data among the third type data in the database 1013 prepared in advance and related to the occurrence record of delay events in past operations, and the data selection unit selects from the two or more types of data. You may obtain the type of data specified. Based on at least one of the radio wave condition of the target route and the calculation load of the learning data generation unit, the data selection unit selects one or more types of delay presence/absence result data 163 among the above two or more types of data. data may be selected. For example, the data selector may do at least one of the following.
- If the radio wave condition of the target route is equal to or higher than the first threshold, the first type data is selected.
- If the radio wave condition of the target route is less than the second threshold (the second threshold ≤ the first threshold), at least one of the second and third data is selected.
- If the calculation load of the learning data generation unit is equal to or greater than the threshold value A, at least one type of data is selected from the second type data and the third type data.
If the calculation load of the learning data generation unit is less than the threshold B (threshold B≦threshold A), the first type data is selected.

100, 900, 1000, 1100... train operation control device

Claims (12)

Based on past operation performance data, which is data indicating the past operation performance of each of the multiple trains on the target route, and delay presence/absence performance data, which is data indicating the relationship between time, train, station interval, and presence or absence of delay a learning data generation unit that generates learning data that is data for learning whether or not an occurrence occurs;
A prediction model, which is a model for detecting signs of delay occurrence, is constructed by machine learning using the learning data, and operation results, which are data indicating the latest operation results of each of the plurality of trains on the target route. a delay sign detection unit that detects the presence or absence of a delay sign for each of the plurality of trains by inputting the most recent data into the sign model;
with
The learning data includes a data group with a track record of delays and a data group without a track record of delays for each station interval and each direction,
The group of data with a track record of delays is one or more trains that ran different from the plan among one or more trains that ran ahead of the train that started to be delayed, among the past operation record data. It is data that shows the operation results for each, and is composed of data of trains that have a track record of causing delays, not data of trains that actually caused delays.
The group of data without a track record of delays is composed of data obtained by excluding the group of data with a track record of delays and the data indicating the track records of each of the one or more trains in which delays occurred, among the past data of track records of operation. cage,
The past operation record data is data obtained from a ground server that obtains operation status data, which is data indicating the operation status of the train, from the train in real time by wireless communication,
The latest operation performance data is data obtained from the ground server for each of the plurality of trains, or data obtained from ground equipment of a signal security device and containing train position information.
A train operation control device characterized by: The train operation control device according to claim 1,
For each train, the occurrence of a delay is defined as the difference between the time when the train will be at a certain position and the time when the train is actually at that position, based on the unit interval of the timetable. is above the specified threshold,
A train operation control device characterized by: The train operation control device according to claim 1 or 2 ,
The latest operation performance data is data acquired from the ground equipment,
The driving status data acquired from the train by the ground server is data acquired from the train when the train is stopped at a place where it stops for a certain period of time or longer.
A train operation control device characterized by: The train operation control device according to any one of claims 1 to 3 ,
For each train, the operating status data includes at least data indicating the location of the train at each time,
A train operation control device characterized by: The train operation control device according to claim 4 ,
For each train, the position on the train is the position on the train recognized by a GPS device included in a device fixed on the train or a portable terminal brought into the train by the crew in charge.
A train operation control device characterized by: The train operation control device according to any one of claims 1 to 5 ,
The delay presence/absence performance data is data obtained by comparing a planned run curve and an actual run curve according to the operation performance past data,
A train operation control device characterized by: The train operation control device according to any one of claims 1 to 6 ,
The delay presence/absence performance data is data obtained by comparing a planned timetable with an actual timetable according to the operation performance past data,
A train operation control device characterized by: The train operation control device according to any one of claims 1 to 7 ,
The delay presence/absence track record data is data obtained from a database prepared in advance regarding the record of occurrence of delay events in past operations,
A train operation control device characterized by: The train operation control device according to any one of claims 1 to 8 ,
an omen output unit that outputs information indicating the presence or absence of a omen of delay detected by the omen of delay detection unit;
A train operation control device, further comprising: The train operation control device according to any one of claims 1 to 9 ,
An operation pattern switching unit that switches an operation pattern of a train that is automatically operated based on the presence or absence of a delay sign detected by the delay sign detection unit;
A train operation control device, further comprising: The train operation control device according to any one of claims 1 to 10,
further comprising a data selection unit,
The learning data generation unit acquires the type of data selected by the data selection unit from among the following two or more types of data as the delay presence/absence result data,
- Type 1 data obtained by comparing the planned run curve with the actual run curve according to the past operational performance data,
- The second type of data obtained by comparing the planned timetable with the actual timetable according to the past operational data,
・ The third type of data in the database prepared in advance regarding the occurrence record of delay events in past operations,
The data selection unit selects one of the two or more types of data as the delay presence/absence result data based on at least one of the radio wave condition of the target route and the calculation load of the learning data generation unit. select more than one type of data,
A train operation control device characterized by: The train operation control device provides past operation performance data, which is data indicating the past operation performance of each of a plurality of trains on the target route, and delay presence/absence performance data, which is data indicating the relationship between time, train, station interval, and whether or not there is a delay. Generate learning data, which is data for learning whether or not a delay occurs, based on
The train operation control device builds a sign model that is a model for detecting signs of delay occurrence by machine learning using the learning data,
The train operation control device inputs the latest operation performance data, which is data indicating the latest operation performance of each of the plurality of trains on the target route, into the prediction model, thereby determining whether there is a sign of delay for each of the plurality of trains. to detect
The learning data includes a data group with a track record of delays and a data group without a track record of delays for each station interval and each direction,
The group of data with a track record of delays is one or more trains that ran different from the plan among one or more trains that ran ahead of the train that started to be delayed, among the past operation record data. It is data that shows the operation results for each, and is composed of data of trains that have a track record of causing delays, not data of trains that actually caused delays.
The group of data without a track record of delays is composed of data obtained by excluding the group of data with a track record of delays and the data indicating the track records of each of the one or more trains in which delays occurred, among the past data of track records of operation. cage,
The past operation record data is data obtained from a ground server that obtains operation status data, which is data indicating the operation status of the train, from the train in real time by wireless communication,
The latest operation performance data is data obtained from the ground server for each of the plurality of trains, or data obtained from ground equipment of a signal security device and containing train position information.
A train operation control method characterized by:

Images