JP6573568B2 - Operation prediction system, operation prediction method, and operation prediction program - Google Patents

Description

  The present invention relates to an operation prediction system, an operation prediction method, and an operation prediction program.

  In the railway system, there is a growing need for predicting operation conditions such as early arrival and delay of trains with high accuracy. For example, to achieve arrival timing adjustment, postponement instruction, and postponement control for each train so that the regenerative power generated when the arriving train decelerates can be efficiently used to accelerate other departure trains The train arrival time needs to be predicted with high accuracy. If the operation prediction accuracy of the entire train can be improved statistically, it can contribute to power saving statistically.

  Patent Document 1 describes a record of wind speeds measured at a certain point and past wind speeds for the purpose of reducing the risk of trains being exposed to danger from strong winds and minimizing the impact on train schedules. Based on this, a configuration has been proposed in which a predicted wind speed up to a predetermined time ahead at the point is calculated and output by sequential calculation using a time-series model expressing stochastic fluctuations.

  In addition, Patent Document 2 aims to provide travel time information on a transportation network that can be moved by rail or bus for the purpose of providing highly accurate travel time information. Probability distribution data that expresses time as a probability distribution is stored for each location where the boarding waiting time or travel time occurs, and the probability corresponding to the boarding waiting time and travel time occurring on the route connecting the departure point and destination Propose to obtain all distribution data, calculate the probability distribution of the total travel time by convolving the probability distribution data, determine the expected travel time based on the calculated probability distribution of the total travel time, and output it Yes.

  Further, Patent Document 3 is a system that predicts a moving process using a transportation facility that is repeatedly operated at a specific time. For each of the past moving objects, the passing date and time at each passing point passes through two passing points. Elapsed time, transit system specifying information, status information, boarding end date and time, and association data that associates the comparison results of each passing date and time with respect to the boarding end date and time are stored, and the passing date and time related to a specific passing point, The comparison result of the passage date and time relative to the boarding end date or time, or the elapsed time relating to two specific passage points, and the passage date and time, the comparison result or the passage time, the comparison result or A plurality of combinations with other information associated with the elapsed time are extracted, and based on the extracted combinations, regression analysis, or By statistical calculation to estimate the average or variance, the passage time when a future moving object passes a specific passing point, the comparison result of the passing date and time for a specific passing point with respect to the future boarding end date and time in a specific transportation, or the future It is proposed to calculate an estimate of the elapsed time that passes while a moving object passes through two specific passing points.

JP 2005-30988 A JP 2014-126500 A Japanese Patent No. 4839416

Patent Document 1 predicts the wind speed at a certain point, and does not directly predict train operation. Patent Document 2 creates travel time obtained as a result of route search for means of transportation as a probability distribution,
Although it is possible to obtain the risk of early arrival and delay on average for each route, it does not predict the individual operation status of the means of transportation. Further, Patent Document 3 performs operation prediction by combining a plurality of analysis models, but does not disclose adjustment of the combination balance of analysis models in order to improve prediction accuracy. There are several known methods for probabilistically expressing train arrival times using statistical information obtained from actual values of delays in train arrival times in a railway system. However, it is difficult to sufficiently increase the prediction accuracy in the prediction with a high risk of deviating (that is, the prediction at the place where the variance of the prediction is large in the past).

  One object of the present invention is to increase the operation prediction accuracy of a moving object as much as possible.

  In order to achieve the above and other objects, one aspect of the present invention is an operation prediction system for predicting an operation status of a mobile object, and an initial prediction of data related to the operation status of the mobile object With respect to a prediction error that is an error between the value and the actual value obtained through actual operation of the data, the average value of the prediction error with respect to the sample of the moving body determined by the specified operation location and operation time range, A statistical calculation unit that calculates a variance value and a covariance value, and an average value, a variance value, and a covariance value of the prediction error calculated by the statistical calculation unit for two or more different groups of samples are used as error variance values. It is a driving | operation prediction system provided with the prediction calculation part which calculates the predicted value of the data regarding the driving | running | working condition of the said mobile body by combining and using so that may become the minimum.

  According to the operation prediction system according to one aspect of the present invention, it is possible to increase the operation prediction accuracy of the moving body as much as possible.

FIG. 1 is a diagram showing an example of the overall system configuration of a business support system 1 according to an embodiment of the present invention. FIG. 2 is a diagram illustrating a hardware configuration example of the computer 10 used in the system of FIG. FIG. 3A is a diagram illustrating a configuration example of the plan information table 300. FIG. 3B is a diagram illustrating a configuration example of the plan information table 300. FIG. 3C is a diagram illustrating a configuration example of the plan information table 300. FIG. 3D is a diagram illustrating a configuration example of the plan information table 300. FIG. 4 is a diagram illustrating a configuration example of the train / station statistical record table 400. FIG. 5 is a diagram illustrating a configuration example of the station / time zone statistical recording table 500. FIG. 6 is a diagram illustrating an example of a train operation model. FIG. 7 is a diagram illustrating an example of an operation prediction process flow according to an embodiment of the present invention. FIG. 8 is a diagram illustrating an example of the update process flow of the statistical recording table. FIG. 9A is a diagram illustrating a configuration example of an input screen. FIG. 9B is a diagram illustrating a configuration example of the input screen. FIG. 10 is a diagram illustrating a configuration example of the output screen.

  Hereinafter, the present invention will be described in detail with reference to the drawings according to the embodiment. Note that components corresponding to each other are denoted by the same reference symbols throughout the drawings for describing the embodiment, and the repetitive description thereof will be omitted.

FIG. 1 shows a configuration example of a business support system 1 for supporting a business of a railway system, which includes an operation prediction system 100 according to the present embodiment. The business support system 1 is configured such that a plurality of computer systems are communicably connected, and the operation prediction system 100, the data server 200, the driving support system 1000, the operation management system 2000, the power management system 3000, and the service system 4000 communicate with each other. The network N is connected to each other. The operation prediction system 100, the data server 200, the driving support system 1000, the operation management system 2000, the power management system 3000, and the service system 4000, which are subsystems of the business support system 1, are each a computer having the configuration of FIG. Can be configured.

FIG. 2 is a configuration example of the computer 10 that can be used as a subsystem. The computer 10 includes a processor 11, a memory 12, an auxiliary storage device 13, an input / output device 14, and a communication device 15, and these elements are connected to each other via a bus 16 so as to communicate with each other. The processor 11 is an arithmetic device such as a CPU (Central Processing Unit) that executes arithmetic processing. The memory 12 is a storage device such as a RAM, ROM, or NVRAM (Non-Volatile RAM) that provides a storage area for storing a program that provides a data processing function executed by the computer 10 and data used by the program. Have. The auxiliary storage device 13 includes a storage device such as a hard disk drive (HDD) or a semiconductor drive (SSD) that stores the program, data table, and the like. The input / output device 14 includes external data input devices such as a keyboard, mouse, and touch panel, and data output devices such as a monitor / display and a printer. The communication device 15 includes a communication device such as a network interface card (NIC) that enables communication with other devices via a communication network N that employs an appropriate communication protocol. In the computer 10, the processor 11 reads out the program stored in the auxiliary storage device 13 to the memory 12 and executes it, thereby realizing the function of each subsystem.

The business support system 1 is configured to realize safe and efficient train operation in the railway system by the function of each subsystem.
The driving support system 1000 includes, for example, an on-board monitor installed in a cab of a train, and has a function of supporting a driving operation of the driver by transmitting a driving prediction result by the driving prediction system 100 described later to the driver. . The driving support system 1000 can also be configured to exchange information with other devices such as a train information management device for comprehensively managing on-board information or an automatic driving device.

  The operation management system 2000 has a function of controlling signal display and route setting based on a train operation plan (diagram) set in advance, and also displays actual train operation status on the ground such as track circuits and proximity sensors. It acquires from a train detection system of this as an operation performance value, and has the function to provide to the data server 200 and the operation prediction system 100.

  The power management system 3000 has a function of managing operations of a power receiving / transforming facility, a power feeding facility, and the like for supplying power to various facilities including a train of the railway system. The service system 4000 has a function of providing passenger information services such as guidance display in the station premises and train approach display.

Next, the data server 200 will be described. The data server 200 stores, manages, and distributes data according to data input / output requests from subsystems such as the operation prediction system 100, the driving support system 1000, the operation management system 2000, the power management system 3000, and the service system. It has a function. The data server 200 in FIG. 1 stores initial plan data 210, actual plan data 220, constraint plan data 230, statistical record data 240, and prediction plan data 250. These data are stored in an appropriate data format such as a table. In addition to these data, the data server 200 is provided with programs for executing data input / output processing with respect to the subsystems, but these programs are not essential with regard to the configuration and functions of the present invention. Therefore, illustration and description are omitted.

  The initial plan data 210 is data created in advance by a railway system user and stored in the data server 200. Specifically, the initial plan data 210 is a train operation diagram, and the scheduled departure time and arrival time of each train at each station are stored for each line system of the railway system. Further, basic information for causing the operation prediction system 100 to function, such as various attribute information (station name, station kilometer, etc.) that is basic data of a train operation diagram, is also stored in the data server 200 as management data in advance.

  The actual plan data 220 stores data reflecting the actual train operation status, such as train arrival time and train departure time at each station. Specifically, actual values such as train arrival time and train departure time acquired from a signal security system connected to the operation management system 2000 are input from the operation management system 2000 and the actual plan data 220 is updated. The update of the actual plan data 220 may be executed at predetermined time intervals, or may be executed sequentially according to the operation status of each train.

  The constraint plan data 230 and the statistical record data 240 are data groups in which data used in the process of calculating the forecast plan data 250 from the initial plan data 210 and the actual plan 220 are accumulated. The constraint plan data 230 and the statistical record data 240 will be described later in relation to the data processing flow executed by the operation prediction system 100 of the present embodiment. The statistical record data 240 is recalculated and updated based on the update of the actual plan data 220.

  The prediction plan data 250 is data indicating predicted values of departure time and arrival time of a train at a prediction target station in a railway system to which the business support system 1 of the present embodiment is applied. The prediction plan data 250 is calculated by the operation prediction system 100 based on the initial plan data 210 and the actual plan data 220 stored in the data server 200. As described above, the constraint plan data 230 and the statistical record data 240 are generated and used in the process in which the operation prediction system 100 calculates the forecast plan data 250 from the initial plan data 210 and the actual plan data 220. The calculation process of the prediction plan data 250 will be described later.

  As described regarding the driving support system 1000, the operation management system 2000, the power management system 3000, and the service system 4000, which are the subsystems, the prediction plan data 250 is distributed to the driving support system 1000, and the operation prediction information is transmitted to the driver. provide. Further, the forecast plan data 250 is distributed to the operation management system 2000 and used for operations such as train arrangement and change of the operation plan. Further, the forecast plan data 250 is distributed to the power management system 3000 and is used for a power supply planning work in consideration of the power saving effect as the railway system. Moreover, the forecast plan data 250 is delivered to the service system 4000, and provides an information delivery service based on the latest operation forecast result to users and the like.

  Next, the operation prediction system 100 will be described. The operation prediction system 100 has the configuration illustrated in FIG. 2 and stores a program describing the operation prediction method according to the embodiment of the present invention in the auxiliary storage device 13. In the example of FIG. 1, the processor 11 reads out the prediction module 110 that is a program group stored in the auxiliary storage device 13 to the memory 12 and is stored in the memory 12. A prediction process is executed using the parameter 120.

The prediction module 110 includes a data reception unit 111, a constraint calculation unit 112, a statistical calculation unit 113, a correction calculation unit 114, and a data transmission unit 115, which are a plurality of submodules.
During operation, the operation prediction system 100 always waits for the actual plan data 220 distributed from the operation management system 2000 via the communication device 15, and every time the actual plan data 220 is distributed, that is, the actual plan data 220 is updated. The prediction plan data 250 is calculated by executing the prediction process every time.

  The data reception unit 111 and the data transmission unit 115 use the input / output device 14 in FIG. 2 to perform processing for transmitting / receiving data such as the generated forecast plan data 250 to / from the data server 200, and in FIG. A function of providing a user interface via the output device 14 is provided. The constraint calculation unit 112 has a function of inputting the initial plan data 210 and the actual plan data 220 and outputting the constraint plan data 230 based on a train operation model for predicting train operations described later. The statistical calculation unit 113 has a function of receiving the actual plan data 220 and outputting a statistical record. A method of creating this statistical record will be described later. The correction calculation unit 114 has a function of receiving the constraint plan data 230 and the statistical recording data 240 and outputting the forecast plan data 250. A method of creating the forecast plan data 250 will be described later.

  Next, the data stored in the data server 200 will be described more specifically. 3A to 3D show configuration examples of a plan information table 300 that is a table for storing plan information. In the data server 200, the initial plan data 210 (FIG. 3A), the actual plan data 220 (FIG. 3B), the constraint plan data 230 (FIG. 3C), and the forecast plan data 250 (FIG. 3D) is managed.

  In the plan information table 300, fields of train ID 301, departure station ID 302, arrival station ID 303, departure time 304, arrival time 305, delay 306, and management 307 are provided in common. In the plan information table 300, in order to record and manage data for a plurality of days, the departure time 304 and the arrival time 305 may include the year, month, and date. In addition to the departure time 304 and the arrival time 305, a date field May be provided. The train ID 301 is an identification code for mutually identifying trains operating in the operation prediction target line section of the railway system, and is a train number, for example. The departure station ID 302 and the arrival station ID 303 are identification codes for mutually identifying stations in the operation prediction target line area of the railway system. The departure time 304 and the arrival time 305 indicate the time when the train specified by the train ID 301 departs from the station specified by the departure station ID 302 and the arrival station ID 303 or arrives at the next station. The delay 306 represents the delay of the actual value or the predicted value with respect to the planned values of the departure time 304 and arrival time 305 of the train in the initial plan data 210. The management 307 stores unique information corresponding to the type of the plan information table 300. In the present embodiment, “initial”, “actual”, “constraint”, and “prediction” are stored as information for identifying the type of plan information stored in the plan information table 300. The prediction module 110 uniquely refers to attribute information such as train names and station names stored on the data server 200 using the train ID 301, departure station ID 302, and arrival station ID 303 recorded in the plan information table 300. Can get.

  Here, the plan information table 300 illustrated in FIGS. 3A to 3D will be sequentially described. First, FIG. 3A shows an example of the initial plan data 210, and the train identified by the train ID = 001 departs from the station with the departure station ID = AA5 at 13:55 and changes to the station with the arrival station ID = AA4. It is planned to arrive at 14:00. Since the data as described in the operation plan diagram is recorded, the delay 306 is blank.

Next, the actual plan data 220 in FIG. 3B records the actual operation time acquired from the operation management system 2000. For example, the top record in FIG. 3B shows a case where the train identified by train ID = 001 departs from the station identified by departure station ID = AA5 with a delay of 2 minutes. There is a record of departure at 13:57. The arrival time of AA4, the next station, is also delayed by 2 minutes, reflecting the delay in departure from the previous station. Delay 30
6, 2 minutes is recorded as the difference between the departure time l3: 55 of the initial plan data 210 and the departure time of the actual plan data 220.

  3C and 3D are configured using the same data structure as the plan information table 300, and store constraint plan data 230 and prediction plan data 250, respectively. 3C and 3D illustrate the same numerical value as the initial plan data 210 as an example of the numerical value recorded in the item of each record, but actually, the constraint plan data 230 includes a train operation model (described later). The departure time 304 and the arrival time 305 calculated according to the constraint model) are recorded. Further, in the prediction plan data 250, a departure time 304 and an arrival time 305 reflecting the operation prediction result executed by the prediction module 110 are recorded. In the following, each item name is treated as a name reflecting the data structure of the plan information table 300. That is, for example, the delay 306 may be an item name of the actual plan data 220 or an item name of the prediction plan data 250 depending on the context.

  Next, the statistical recording table will be described. FIG. 4 shows a configuration example of a train / station-specific statistical record table 400 which is a table storing statistical records acquired and generated for each train and for each station. Here, the statistical record refers to data obtained by statistically processing a set of measured values (for example, departure time, etc.) for a target range focusing on individual targets such as a train, a station, etc. Yes. One record of the train / station statistical record table 400 of FIG. 4 includes train ID 401, departure station ID 402, arrival station ID 403, planned time 404, error average 405, error variance 406, covariance 407, number of samples 408, and An item of management 409 is set. The train ID 401, departure station ID 402, and arrival station ID 403 are the same as those in the plan information table 300 of FIGS. 3A to 3D. The planned time 405 indicates an initial planned value (numerical value on the plan diagram) at which the train with the corresponding train ID 401 departs from the station specified by the departure station ID 402. The error average 405 indicates the average value of the difference between the actual time when the train specified by the corresponding train ID 401 leaves the station specified by the departure station ID 402 and the planned time 404. The error variance 406 indicates the variance of the difference between the actual time and the planned time 404. The covariance 407 is a numerical value serving as an index indicating the degree of similarity between the error variance 406 and error variance in other statistical recordings (for example, error variance 505 in FIG. 5 described later). The number of samples 408 indicates the number of samples subjected to statistical processing with respect to the departure time of the station with the departure station ID 402 of the corresponding train ID 401. The management 409 stores information that identifies which other statistical record is the target of similarity determination in the calculation of the covariance 407. In the example of FIG. 4, since the error variance 406 and the error variance 505 of FIG. 5 are contrasted, the phrase “by station / time zone” is recorded in the management 409.

  4 shows an example in which data for one or a plurality of days up to the previous day is recorded as train / station-specific data 450 that is statistical data for each train / station. As will be described later, there is also a statistical recording table that records data for the current day only. In this way, in what time range sampling is performed for the statistical recording tables 400, 500, etc., is set according to user designation.

FIG. 5 is a diagram illustrating a configuration example of a statistical record table 500 that stores station / time-specific data 550 that is a statistical record for each station and time zone. In each record of the statistical record table 500, items of departure station ID 501, arrival station ID 502, time zone 503, error average 504, error variance 505, covariance 506, number of samples 507, and management 508 are recorded. The departure station ID 501 and the arrival station ID 502 are the same as the departure station ID 402 and the arrival station ID 403 in the statistical recording table 400. The time zone 503 indicates the sampling range for the arrival time of the station specified by the arrival station ID 502 as a time range. For example, in the example of FIG. 5, it represents that the arrival time of the train which arrives at station AA4 for 20 minutes from 13:41 to 14:00 is made into the object of statistical processing. The error average 504, error variance 505, and covariance 506 are the same as the error average 405, error variance 406, and covariance 407 in the statistical recording table 400 of FIG. The number of samples 507 indicates the number of samples subjected to sampling during the time zone 503. The management 508 stores information for identifying which other statistical record is a target of similarity determination in the calculation of the covariance 506. In the example of FIG. 5, since the error variance 406 and the error variance 505 of FIG. 5 are contrasted, the phrase “by train / station” is recorded in the management 508.

  In addition, in FIG. 5, the data example for the day is given as a tendency of the prediction error of the arrival time of the preceding train for every 20-minute window (time zone) for each station. As described above, in what time range sampling is performed for the statistical recording tables 400, 500, etc., is set according to user designation.

  Next, the basic principle of the train operation prediction method in this embodiment will be described in detail. The train operation prediction of the present embodiment is performed based on the planned departure time of the train that leaves each station and the planned arrival time of the train that arrives at each station. Moreover, when acquiring the actual time value of the departure time of the train leaving each station and the actual value of the arrival time of the train arriving at each station from the operation management system 2000 as a subsystem, the train operation prediction is based on the actual value. Is executed.

FIG. 6 shows an example of a train operation model. In the model of FIG. 6, predicted values of arrival time and departure time of each station are calculated as the earliest time within a range satisfying a predetermined constraint condition for a train operation in time series. In the model of FIG. 6, time is plotted on the horizontal axis and distance (about kilometer) indicating the train position on the vertical axis, and the relationship in time and space of the variables used in the following equation (1) is shown. FIG. 6 shows that the train number is a positive integer i, the station number is a positive integer j, and for two trains operating in the same direction, the preceding train j-1 and the continuation train j are the station i-1, The situation of stopping at station i is shown. At this time, the time when the continuation train j arrives at the station i and the predicted time T ^(a) i, j which is the predicted value can be expressed by the following equation (1). Note that max (X, Y) is a function indicating an operation that takes the larger value of X or Y.

  In particular, in the train operation at the time of delay in the dense schedule, the continuation train may be kept waiting at the previous station (delayed) in order to avoid the risk that the continuation train stops between stations due to the influence of the preceding train. For this reason, the traveling time between stations during normal times and delays does not change so much, and the predicted arrival time based on the departure time of two stations is a value with a certain degree of accuracy. Hereinafter, the train operation prediction method based on the train operation model is referred to as a “prediction method based on a constraint model”.

Similarly, regarding the predicted departure time of the train, it is possible to predict whether the continuation train can depart from the preceding train as planned by setting a standard stop time and the like in advance. In the following, for simplification of explanation, only the case of predicting the train arrival time is taken up, but the train operation prediction method of the present invention is also applied to prediction of train departure time and other numerical predictions related to train operation. Can do.

There is a prediction error between the predicted arrival time of a train to a certain station and the train arrival time as the actual value. In train operation, there is an expectation that the continuation train after departure has arrived at or near station i at the time of the predicted time, and depending on the situation where the driver is on his / her own train, The arrival time at the station i varies depending on how the remaining distance is run, and the predicted value of the arrival time at the station i indicates the aspect of the stochastic process. This train operation can be expressed as a random movement according to the Brownian motion of fine particles. That is, the time when the continuation train j arrives at the station i is defined as a random variable T ^(b) i, j. For a time increment dTi, j, k between minute distances dXi, j, k (k is a natural number) obtained by subdividing the last unit distance Xi, j (= 1) reaching station i, a probability parameter (μ, Using σ), and further using the Wiener process increment dBi, j as the fluctuation component, the train operation is modeled by a differential equation of the following equation (2).

When all the minute distances dXi, j, k are added, a unit distance Xi, j is obtained. The initial value of the Wiener process Bi, j is zero by its definition. Then, T ^(a) i, j obtained by equation (1) of the constraint model is used as the initial value of the prediction time. At this time, the differential equation (2) can be uniquely solved to obtain the following equation (3). This is called a “prediction method using a Brownian motion model”.

  The probability parameters (μ, σ) can be estimated by measuring errors from the past prediction results and actual values, and using the error average value μ ′ and the standard deviation value σ ′, respectively. However, the value of the probability parameter to be used differs depending on which sample is selected and considered as an error. That is, even if the prediction method is limited to the Brownian motion model, there are various methods depending on the selection of statistical samples.

  For example, when a tendency of a prediction error is observed for a specific train and a specific station in a certain period up to the previous day, a probability distribution in which problems specific to the station such as connection at the station are reflected more deeply. Moreover, when the tendency of the prediction error of the day to a station immediately before is observed about a specific train, it becomes a probability distribution in which the habit of the driving operation of the driver on the train is reflected more deeply. In addition, regarding the preceding train, when the tendency of the prediction error on the current day at a specific station is observed, the probability distribution reflects the weather, the influence of alternative transportation, and the like more intensely. Furthermore, for the following train, when the tendency of the prediction error on that day at a specific station was observed, the case where the preceding train was made to wait for the interval adjustment due to the sudden illness at that station reflected more intensely. Probability distribution. As described above, the prediction result varies depending on which sample is selected and how an error from the predicted value is considered.

The error between the predicted value and the actual measured value is generally due to a movement inherent in the operation of a specific train up to the previous day (A), a movement sensitive to the entire operation status of the day (B), and an unpredictable event (C). Occurs based on. Therefore, by efficiently combining and using prediction errors based on different samples, it is possible to diminish the influence of unpredictable factors of C and suppress overall prediction errors. For example, a tendency that tends to be delayed due to a connecting connection at a specific station of a specific train appears in the actual error value as a movement up to the previous day. Alternatively, if it rains on that day, the time for getting on and off will increase, and there will be a tendency for the entire train operation to be delayed in addition to the train of interest. Thus, the motion A up to the previous day is fixed, and it can be considered that the final prediction error is determined according to the motion B to which the entire operation status of the day is sensitive.

  Therefore, two types of prediction errors A and B based on two types of prediction methods A (trends up to the previous day) and a prediction method B (trends on the day) based on the Brownian motion model are considered. Here, a calculation example using the A and B2 variables will be described, but the corresponding multivariable optimization calculation may be performed using three or more variables so as to take into account three or more probability distributions. In addition to the Brownian motion model, there are various prediction methods, so other prediction methods may be combined.

Now, let us consider the final prediction error when combining the prediction methods A and B with the application ratios of q and r (q + r = 1), respectively. The final prediction error variance Dp is the prediction error variance σa by the prediction method A, the prediction error variance σb by the prediction method B, the error variance by the prediction method A, and the covariance σab of the error variance by the prediction method B. Can be expressed using the following equation (4). The variance Dp indicates the degree of risk of being out of prediction.

  Equation (4) is a quadratic equation viewed from the application ratio r of the prediction method B (trend of the day), and the distribution of r has an extreme value. That is, when Dp = 0 in equation (4) and the solution of r of the quadratic equation is obtained, q is also determined. Their application ratios q and r constitute an optimal solution that minimizes the risk of misprediction.

By using the error average μa in the case of the prediction method A and the error average μb in the case of the prediction method B, the train arrival time T ^(b) i, j as the prediction result can be obtained as shown in the equation (5). . T ^{(a) As} the initial value of i, j, a predicted value obtained by the constraint model prediction method is used. Bi. Showing the Wiener process. j may be regarded as an expected value of zero. Alternatively, the fluctuation can be calculated sequentially based on the equation (2), and can also be obtained by totaling.

In the prediction method that reflects the trend A up to the previous day, if all predicted values and actual values such as train arrival times are recorded, the average and variance values can be calculated as statistical processing. However, in the prediction method that reflects the trend B of the current day, the predicted value and the actual value are updated every time a train arrives at a specific station. Therefore, both the error variance of the trend A up to the previous day and the error variance of the trend B of the current day Since the variance is calculated based on all records, the calculation load on the system becomes excessive. Therefore, formula (6), which is a calculation formula for adding a new sample to the existing distribution, is used. Equation (6) indicates that a new actual value sample x has been added to the existing n samples.

Expression (6) is a calculation expression assuming that the prediction error as a sample has a normal distribution. The difference between the predicted value and the actually measured value is called a difference prediction error. It is customary to express the difference prediction error as ± 10 seconds, and it may be applied to a series of mathematical expressions as it is. Here, the formula (3) is expanded to use the formula (7). In other words, the ratio between the predicted value T ^(a) i, j based on the constraint model and the desired predicted value T ^(c) i, j is Er, and the logarithmic value of the ratio Er is the prediction error. For example, if the ratio Er is 1.1 times, the error is on the delay side, and if it is 0.9 times, the error is on the early arrival side.

  On the other hand, the train arrival time has a lognormal distribution. Further, the prediction error based on the constraint model or the Brownian motion model also takes a lognormal distribution. That is, the predicted value and the actual value have a lognormal distribution. Therefore, the ratio between the predicted value and the actual value also has a lognormal distribution, and the prediction error that is the logarithmic value of the ratio between the predicted value and the actual value has a normal distribution. For this reason, the logarithmic value of the ratio between the predicted value and the actual value is adapted to the normal distribution which is the premise of the calculation of the equation (6). When the expression (7) is used, since the prediction error does not take a negative value, it can be directly and accurately used at high speed without requiring correction of a calculation expression related to the probability distribution.

When the basic principle of train operation prediction according to the above-described embodiment of the present invention is arranged, the prediction method disclosed in this embodiment includes a method expressed by Equation (1), Equation (3), Equation (7), Equations (3) and (7) are used internally with Equation (1) as an initial value. Furthermore, in each mathematical formula, there are a wide variety of variations in how to select which sample from past performance values and consider it as a probability distribution. That is, the format of the plan information and the statistical record is slightly different depending on which actual value constituting the stochastic process is selected. However, changes in the format may be appropriately reflected. Further, for simplification of explanation, only the train arrival time has been described in the present embodiment, but it may be used for prediction of other events (train departure time, etc.) in train operation. Or the prediction method of this embodiment is applicable not only to train operation but to the whole mobile body.
Further, in the present embodiment, for the sake of simplicity of explanation, a configuration example of table data is shown based on the formula (3), but other mathematical formulas may be used. When using equation (7), equation (8) may be used instead of equation (5) to correct the predicted value.

  Next, data processing executed by the operation prediction system 100 configured by applying the train operation prediction method according to the present invention described above will be specifically described. FIG. 7 is a diagram for explaining an example of a prediction process flow for obtaining a predicted value reflecting an actual value for an initial plan value related to train operation based on a train operation diagram. This prediction process is realized by the program of the prediction module 110 stored in the operation prediction system 100, and in particular, the process for correcting the predicted value is executed by the correction calculation unit 114. The execution timing of the prediction process is generally set to the time when the actual plan data 220 stored in the data server 200 is changed, but can be periodically executed at an appropriate time interval, You may set so that it may perform based on the execution instruction from the outside.

  When the prediction module 110 starts processing in S701 (the symbol S represents “step”, the same applies in the present specification), first, the parameter 120 is set (S702). The prediction module 110 reads initial values of parameters (for example, a prediction target train, a target station, a probability distribution (prediction error) type, a target time range) created in advance and stored in the memory 12 or the like, which will be described later. If there is a user input, the process of replacing the parameter initial value with the user input value is also performed.

  User input can be performed using, for example, user input screens 900 and 910 shown in FIGS. 9A and 9B. FIG. 9A is an example of a prediction process setting screen 900 for setting a target for executing the prediction process by the operation prediction system 100. By setting a check box and text input, a line section and a time range for performing the prediction process can be set. I can do it. In addition, when starting a prediction process, you may set so that all the trains of all the line sections and all the stations which the continuous operation prediction system 100 manages may be made into object.

  In this embodiment, parameter initial values are set in advance, and in S702, the user input screen of FIG. 9B is displayed to receive user input. The user input screen 910 in FIG. 9B designates which sample is selected and considered as an error, that is, a probability distribution to be calculated. Tables corresponding to the selected probability distribution (plan information table 300, statistics recording tables 400, 500) are prepared in the data server 200 in advance. The current time data used for the prediction process may be configured to be set via the input screen of FIG. 9B, or the current time used by the processor 11 of the computer 10 may be used.

In the user input screen 910 of FIG. 9B, specifically, the user designates the usage range of the actual value used for correcting the initial plan. Since the operation prediction system 100 of the present embodiment handles prediction errors obtained by the two methods of the prediction method A and the prediction method B, the user uses the input screen 910 to calculate each error (for example, the target train). , Target station, probability distribution (prediction error) type, target time range) are specified as parameters. As the prediction range, the types of trains include the train, the preceding train, and the succeeding train. The types of stations include the station, the previous station, and the next station. The time range includes the current day, the previous day, and multiple days up to the previous day. Combinations of these are conceivable for the method of using the actual values, and these designations are treated as parameters as described above. Specifically, for example, a user input screen 910 is displayed, and a parameter value can be set in S702 and / or S802 by selecting an actual value used by the user. When the check box of the train is checked, the train ID 401 is used for the prediction method A.
Statistical recording data 450 for each is created. Although omitted in FIG. 9B, the same designation item is prepared for the prediction method B by scrolling the display area in the lower part of the screen. In the present embodiment, the application ratio of the two prediction models calculated by the expression (4) may be designated by user input. Tables corresponding to the statistical recording tables 400 and 500 are prepared in advance for the number of combinations of actual values used by the user via the user input screen 910. The statistical calculation unit 113 and the correction calculation unit 114 prepare in advance a mathematical formula or the like that enables execution of statistical recording processing according to a combination of record values that can be set.

  In S703, the statistical calculation unit 113 calculates the prediction error A based on the set parameter, creates the statistical recording table 400, or updates the created statistical recording table 400. The detailed processing method will be described later with reference to FIG. In the parameter 120, when the prediction error A is specified to use the samples up to the previous day, the statistical calculation unit 113 may acquire the prediction error A that has been calculated and stored in the memory 12 in advance. Good.

  In step S <b> 704, the statistical calculation unit 113 calculates the prediction error B based on the set parameters, creates the statistical recording table 500, or updates the created statistical recording table 500. The detailed processing method will be described later with reference to FIG.

  Next, the prediction module 110 identifies the train ID 301 to be predicted in S705. If it is determined that there is a train ID 301 to be subjected to the prediction process, the prediction module 110 refers to the initial plan data 210 and selects a prediction target in order of trains that depart from the target station or arrive at the target station after the current time. . The prediction module 110 further calls a program stored in the constraint calculation unit 112, reads the initial plan data 210 and the actual plan data 220, creates the constraint plan data 230 using the equation (1), and creates the memory 12 Save to. If it determines with calculation having been completed about train ID301 used as all the prediction targets, the prediction module 110 will complete | finish a prediction process (S709).

  Next, in S <b> 706, the correction calculation unit 114 of the prediction module 110 calculates an optimum value for the ratio r that reflects the tendency of the prediction error A in the next prediction and the ratio q of the prediction error B using Equation (4). calculate.

  Next, in S707, the correction calculation unit 114 determines the prediction correction amount (corrected average value, correction variance) reflecting the prediction error based on the actual value using the formula (5), and the predicted arrival time Ti, j Then, a correction average value and a correction variance value are calculated. The corrected average value is obtained by multiplying the average values of the prediction errors A and B by the ratios q and r and adding them. Similarly, the correction variance is obtained by multiplying the variances of the prediction errors A and B by the ratios q and r and adding them.

Next, in S708, the prediction module 110 calculates a prediction value using equation (5). The ratios q and r of the prediction errors A and B have been calculated in S706. As the initial value T ⁽ⁱⁿⁱ⁾ i, j of the predicted arrival time, the corresponding predicted value of the constraint plan data 230 is used. The average value μ and the variance σ are read from the corresponding plan information, and Bi, j indicating the winner process is treated as zero expected value. Thus, equation (5) generates a unique numerical value.

  In the example of FIG. 7, when the prediction result is output in S708, the prediction module 110 returns to S705 to specify the prediction target, and when it is determined that there is no prediction target, the process is terminated. However, if it is determined in S705 that there is no prediction target, it is further determined whether the actual plan data 220 is updated according to the train operation status, and if it is determined that it is updated, the process returns to S702 and a new prediction is made. Processing may be executed.

  Next, the series of prediction processes illustrated in FIG. 7 executed by the operation prediction system 100 of the present embodiment will be described more specifically with reference to FIGS. 3A to 3D, 4, and 5. In the following, the time when the train identified by the train ID 401 “001” (hereinafter “the train”) arrives at the station identified by the station ID 403 “AA4” (hereinafter “the station”) is predicted. Will be explained.

  In S703 of FIG. 7, for the prediction error A, the statistical record for each train and station up to the previous day (the statistical recording table 400 of FIG. 4) is used. In S704, for the prediction error B, the statistical record for each station and time zone on the current day (statistical record table 500 in FIG. 5) is used. Consider a case where attention is paid to the arrival time at which the train arrives at the station in the course of sequentially processing all trains managed by the operation prediction system 100 in S705.

  In FIG. 4, the scheduled time 404 for arrival of the train at the station is 14:00. This matches the arrival time 305 planned for the train whose train ID 301 is “001” in the initial plan data 210 of FIG. 3A. In the statistical record data (train / station-specific data 450) up to the previous day in FIG. 4, the error average 405 of the predicted value and the actual value that predicted the arrival time of the train at the station is 20 seconds, and the error variance 406 is The sample number 408 is 33 for 5 seconds. In this statistical value example, the prediction error A tends to produce a prediction of an error average value with a certain degree of reliability because the error variance is small to some extent and the number of samples is large.

  On the other hand, in FIG. 5, the line including the 14:00, which is the planned arrival time 404 of the train (train ID 401 = 001) in FIG. 4, in the time zone 503 = 13: 41 to 14:00 In the prediction record B, the error average 504 is 35 seconds, the error variance 505 is 28 seconds, and the number of samples 507 is 5. In this statistical value example, the prediction error B tends to cause a larger delay in the prediction error in the preceding train as viewed from the error average 504, and the error variance 505 is larger than the prediction error A and the number of samples. Since 507 is small, it can be seen that the reliability of the error average value is low.

  In S706 of FIG. 7, the error variance 406 of the prediction error A is σa, 28 that is the error variance 505 of the prediction error B is σb, and −0.11 of the covariance 407 and the covariance 506 is σab. With the risk Dp = 0, the solution of the quadratic equation (4), that is, the optimum value of the predicted application ratio q (or r) is calculated. Note that the covariance between the prediction error A and the prediction error B is −0.11 at this point, which is close to zero and shows a tendency that the two probability distributions do not link much.

  In S707, a predicted value of arrival time at which the train arrives at the station is obtained. Although it is possible to predict the arrival time of the train at the station by using only one of the prediction error A and the prediction error B, the optimal value of the prediction application ratio q (or r) is a plurality of prediction errors. Therefore, it is possible to stably generate highly accurate prediction values by minimizing the risk of unforeseen predictions.

  In S708, a predicted value is output as the arrival time 305 of the predicted plan data 250 (FIG. 3D). Note that the departure time may be similarly predicted by the above prediction processing procedure.

  Next, statistical information update processing according to the present embodiment will be described. FIG. 8 is a diagram illustrating a processing procedure for updating statistical information based on an error between the actual value and the predicted value in the operation prediction. The statistical calculation method used here is stored as a program in the statistical calculation unit 113 of the prediction module 110.

This statistical information update process can be executed when, for example, the actual plan data 220 stored in the data server 200 is changed (a sample is added). When the process starts in S801 of FIG. 8, first, the statistical calculation unit 113 sets the parameter 120 in S802, as in the process of S702 in the prediction process of FIG. Next, the statistical calculation unit 113 reads the initial plan data 210, the actual plan data 220, the constraint plan data 230, the statistical record data 240, and the prediction plan data 250 in S803.

  Next, in S804, the statistical calculation unit 113 calculates the ratio between the actual value in the actual plan data 220 and the predicted value in the predicted plan data 250, and the logarithmic value of the ratio, using Equation (7). . A predicted value is calculated at time T1, and the actual value for the predicted value is input after time T2 (> T1). Since the prediction error cannot be calculated for an item for which the actual value is not input in the actual plan data 220, the processing is skipped. In the items of management 409 and management 508 of the statistical recording tables 400 and 500, a record indicating which prediction value of the prediction error has been aggregated is entered. In S805, the statistical calculation unit 113 creates a predicted value of the departure time 304 or the arrival time 305 in the prediction plan data 250, and if the prediction errors have not been aggregated in the statistical recording tables 400 and 500, (6 ) Is used to create or update statistical recording data (for example, error average, error variance, covariance, number of samples) (step S806). By the above statistical information update process, the error of the actual value with respect to the predicted value regarding the train operation obtained by the operation prediction system 100 is reflected in the prediction process in a timely manner, and the operation prediction accuracy can be improved.

  Next, a prediction result output example in the operation prediction system 100 of the present embodiment will be described. FIG. 10 is a diagram illustrating an example of a prediction result output screen. The prediction result output screen is generated for the prediction result calculated by the prediction module 110 of the operation prediction system 100, and is output to the monitor screen of the input / output device 14, the on-train monitor screen of the driving support system 1000, and the like.

  A screen 80 illustrates plan information and shows a two-dimensional space of a position 81 and a time 82. For example, a vector of (departure station, departure time) − (arrival station, arrival time) represents the movement of the train by a line segment, as in a normal train operation diagram. As already described, the predicted value of the train arrival time has a probability distribution, and can be expressed using a probability density function determined by an average value and variance. In FIG. 10, the area 84 (triangular area) showing the distribution of the predictive medium probability in proportion to the output value of the probability density function is expressed by shading. In the example of FIG. 10, as schematically shown in the lower stage, the region 84 has a position where the train starts to decelerate on the line segment representing the train, and the initial predicted value of the arrival time. It is defined as a triangle formed by two intersections of a line segment drawn from the apex to the predicted arrival time corresponding to the prediction error in the case of the latest and the earliest and the horizontal line indicating the station position. The probability distribution indicating the predicted value of the arrival time takes a maximum value in the initial predicted value and decreases to the arrival side and the early arrival side. Therefore, the region 84 is changed into the minute region 84A within an appropriate numerical range of the probability distribution. A display density corresponding to the probability density is assigned to each area 84A. Such a display method is merely an example, and any display form may be employed as long as the form can visually represent a change in the probability distribution with respect to the initial predicted value.

Here, since the error variance 406 and the like stored in the statistical recording table 400 as statistical information is a numerical value assuming a normal distribution, if the output value of the probability density function is used as it is, the distribution becomes equal to the left and right with respect to the average value. . Actually, the lognormal distribution shown in the equation (8) is the original shape of the prediction error distribution, and the probability density distribution is originally biased to the right side (extension side) of the line segment representing the train ID = 001 in FIG. It is. Therefore, by using a predetermined constant and increasing the probability density (probability density on the arrival delay side) of taking an output larger than the average value, a method of drawing is provided in S708 (prediction result output step) in FIG. Accordingly, an appropriate gray level expression can be obtained approximately. Or you may use (8) Formula itself as a probability density function. FIG.
According to the prediction result output screen example, the departure train makes effective use of the regenerative power generated by the arrival train by determining departure time of the departure train 805 in synchronization with the arrival prediction time of the arrival train 806. This makes it possible to increase the energy saving effect of the railway system.

  As described above in detail, according to the operation prediction system 100 of the present embodiment, the prediction error related to the final operation prediction is made as much as possible by appropriately combining the prediction errors related to the operation prediction based on different sample groups. The prediction accuracy can be increased by reducing the size. As a result, particularly in the railway system, the regenerative power generated when the train arriving at the station decelerates can be effectively consumed as powering power for other trains leaving the station. It can contribute to efficiency improvement.

  The embodiment described above is an example, and the present invention is not limited to these. That is, the present invention can be applied in various ways, and all embodiments are included in the scope of the present invention. For example, the present invention is not limited to a railway system, and can be applied to operation status prediction processing in various systems including mobile bodies (aircraft, route buses, etc.) that are operated on a predetermined schedule.

1 business support system 10 computer 11 processor
12 memory 13 auxiliary storage device 14 input / output device 15 communication device 100 operation prediction system 110 prediction module 111 data reception unit 112 constraint calculation unit 113 statistical calculation unit 114 correction calculation unit 115 data transmission unit 120 parameter storage unit 200 data server 210 initial plan Data 220 Actual plan data 230 Constraint plan data 240 Statistical record data 250 Predictive plan data 300 Plan information tables 400 and 500 Statistical record table 1000 Operation support system 2000 Operation management system 3000 Power management system 4000 Service system

Claims (10)

An operation prediction system for predicting the operation status of a moving object,
The prediction error, which is an error between the initial prediction value for the data related to the operation status of the moving object and the actual value obtained through the actual operation for the data, is determined by the specified operation location and the operation time range. A statistical calculation unit that calculates an average value, a variance value, and a covariance value of the prediction error for a sample of a moving object;
By using the average value, variance value, and covariance value of the prediction error calculated by the statistical calculation unit for two or more different groups of samples in combination so as to minimize the error variance value, the moving object A prediction calculator that calculates a predicted value of data related to the operation status of
Operation prediction system equipped with. The operation prediction system according to claim 1,
For the two moving objects that pass through two specific remote points, the prediction calculation unit determines the moving time between the two points of the moving object, and the preceding moving object starts from the one point. The initial predicted value is calculated with the minimum time interval that should be secured before the mobile body to continue from the point of arrival as a constraint,
The different groups of samples are a first group including the actual values collected by the day before the prediction process execution, and a second group including the actual values collected on the day of the prediction process execution,
The statistical calculation unit calculates an average value, a variance value, and a covariance value of the prediction error for the first group and the second group, respectively.
The prediction calculation unit relates to the first group and the second group, the first group that minimizes the variance of the prediction error from the average value, the variance value, and the covariance value of the prediction error. Calculating a ratio of the average value, the variance value, and the covariance value of the prediction error for the second group, and correcting the initial prediction value according to the ratio;
Operation prediction system. The operation prediction system according to claim 1,
The statistical calculation unit is an operation prediction system that calculates a logarithmic value of a ratio between a predicted value and an actual value of data related to the operation status of the moving object as a prediction error. The operation prediction system according to claim 2,
The sample included in the first group is an arrival time for each of the plurality of moving objects that arrive at the specific point in a predetermined period until the day before the prediction process is performed,
The sample included in the second group is an arrival time for each of the plurality of moving objects that arrive at the specific point for each specific time range on the prediction process execution day.
Operation prediction system. The operation prediction system according to claim 1,
While enabling the prediction calculation unit to designate at least one of the moving object, the point, and the operating time range of the moving object that are subject to prediction processing,
For each different group of samples, an input instructing unit that makes it possible to specify the time range in which the mobile object, the point, and the actual value are collected as the application range of the actual value,
Equipped with an operation prediction system. The operation prediction system according to claim 1,
In a plane in which a first axis representing the position of the moving body and the point by a distance from a predetermined reference point and a second axis representing the movement of the moving body by time are two axes, The moving trajectory of the mobile object is displayed as a line segment connecting the position and departure time of the specific point and the position and arrival time of the next point between two adjacent points,
For the arrival time as data relating to the operation status of the mobile object, the occurrence probability for each predicted arrival time is calculated based on the error variance value, and around the end point reaching the next point of the line segment According to the occurrence probability, it is visually identifiable, and the visually identifiable range is set to be wider on the downstream side than the line segment on the upstream side. An operation prediction system equipped with an output unit. The operation prediction system according to claim 1,
The moving body is a train operated by a railway system, and the point is a station installed in the railway system ,
For the two moving objects that pass through two specific remote points, the prediction calculation unit determines the moving time between the two points of the moving object, and the preceding moving object starts from the one point. The initial predicted value is calculated with the minimum time interval that should be secured before the mobile body to continue from the point of arrival as a constraint,
The different groups of samples are a first group including the actual values collected by the day before the prediction process execution, and a second group including the actual values collected on the day of the prediction process execution,
The statistical calculation unit calculates an average value, a variance value, and a covariance value of the prediction error for the first group and the second group, respectively.
The prediction calculation unit relates to the first group and the second group, the first group that minimizes the variance of the prediction error from the average value, the variance value, and the covariance value of the prediction error. For the second group, the ratios q, r (q + r = 1) of the variances σ _a and σ _{b of} the prediction error and the covariance σ _ab are expressed by equation (1).
And calculate so that Dp = 0.
According to the ratios q and r, the initial predicted value T ^(a) i, j is expressed by equation (2).
(Bi, j are Wiener processes)
To correct by
Operation prediction system. The operation prediction system according to claim 7,
The statistical calculation unit is a logarithmic value of the ratio between the predicted value and the actual value for the data related to the operation status of the moving object.
(Bi, j are Wiener processes)
An operation forecasting system that calculates as a prediction error. An operation prediction method for predicting operation status of a moving object,
A computer with a processor and memory
The prediction error, which is an error between the initial prediction value for the data related to the operation status of the moving object and the actual value obtained through the actual operation for the data, is determined by the specified operation location and the operation time range. For the mobile sample, calculate the mean value, variance value, and covariance value of the prediction error,
By using the average value, the variance value, and the covariance value of the prediction error calculated by the statistical calculation unit for two or more different groups of samples in combination so that the variance value of the error is minimized, Calculate the predicted value of data related to operation status,
Operation prediction method. An operation prediction program used to predict the operation status of a moving object, which is a computer having a processor and a memory,
The prediction error, which is an error between the initial prediction value for the data related to the operation status of the moving object and the actual value obtained through the actual operation for the data, is determined by the specified operation location and the operation time range. Regarding the sample of the moving object, the average value, the variance value, and the covariance value of the prediction error are calculated,
By using the average value, the variance value, and the covariance value of the prediction error calculated by the statistical calculation unit for two or more different groups of samples in combination so that the variance value of the error is minimized, Calculate the predicted value of data related to operation status,
Operation prediction program.