Stop time estimation device for traffic vehicle and computer program
The device classifies and analyzes train stop times to derive correlation rules, addressing the challenge of real-time train stop time estimation by enhancing accuracy and reliability in predicting future stop times.
Patent Information
- Application Number
- JP2023083261
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-19
- Publication Date
- 2024-11-29
AI Technical Summary
Conventional operation management technologies struggle to accurately estimate the stopping time of a train at a subsequent station, especially in real-time applications, as they do not account for the dynamic factors influencing passenger boarding and train congestion.
A device and computer program that classify stop times into classes based on elapsed time, analyze the combination patterns of these classes, and derive correlation rules to estimate future stop times by associating stop times of trains at previous stations, using a classification unit, analysis unit, and estimation unit to provide real-time predictions.
Enables accurate real-time estimation of train stop times at subsequent stations, improving the reliability of train operation planning and passenger information systems.
Smart Images

Figure 2024166870000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to a technique for estimating a stopping time of a traffic vehicle. [Background technology]
[0002] In traffic management of transportation vehicles such as trains, it is important to predict near-future station arrival and departure times taking delays into account. As a traffic management technology for such transportation vehicles, a train operation support system disclosed in Patent Document 1 calculates a station stop time estimate based on an estimated station stop time at a stop station of a preceding train, an estimated departure time of the preceding train, a target arrival time at a stop station of a succeeding train, an inter-station running pattern leading to a stop station where the succeeding train satisfies the target arrival time, an estimated number of passengers boarding and alighting, and a passenger occupancy rate. By performing traffic management using this estimated station stop time, it is expected that the accuracy of estimating the departure time of a train from a station will be improved. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] Patent Publication No. 2022-042138 Summary of the Invention [Problem to be solved by the invention]
[0004] Many conventional traffic management technologies, such as the train operation support system disclosed in Patent Document 1, take into account the demand from passengers using trains, and use various approaches to determine the amount of time required until departure, taking into account the congestion of each train and the number of passengers boarding and alighting, and evaluate the influence of temporal factors such as the time of day, day of the week, and month, as well as the distribution of passengers on the platform.
[0005] Therefore, while this is useful when planning operations or revising timetables over a long period of time, it is difficult to apply it to applications such as estimating the time a train currently carrying passengers will stop at subsequent stations.
[0006] An object of the present invention is to provide a technology for estimating the stopping time of a traffic vehicle that is suitable for applications requiring real-time performance by solving the above problems. [Means for solving the problem]
[0007] One aspect of the present invention for solving the above problem is a device for estimating a dwell time for a train that has arrived at one of a plurality of stop stations connecting a specified section until it departs for a subsequent stop station, the device comprising: a classification means for classifying each of the dwell times of a plurality of trains that stop at the plurality of stop stations into one of a plurality of stop time classes determined for each time lapse after arrival at the stop station; a storage means for storing the classified stop time classes in association with each of the plurality of trains for each stop station; an analysis means for searching for a correlation rule in which the stop time class of a subsequent stop station is derived due to the stop time class of one or more stop stations by analyzing a combination pattern of the stop time classes of each train at each stop station that has been stored; and an estimation means for estimating the stop time of a train arriving at the subsequent stop station based on the correlation rule.
[0008] Another aspect of the present invention is a computer program for causing a computer to operate as the above-described stopping time estimation device. Effect of the Invention
[0009] According to the above aspect, it is possible to provide a technology for estimating a stopping time of a traffic vehicle that is suitable for applications requiring real-time performance. [Brief description of the drawings]
[0010] [Figure 1] 1 is an overall configuration diagram of an operational service system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram illustrating an example of a functional configuration of a train stop time estimation device. [Diagram 3] FIG. 2 is a diagram illustrating an example of the configuration of a user terminal. [Figure 4](a) to (i) are distribution maps of stop times at each station during a given period. [Diagram 5] This is a diagram visualizing the changes in stop time classes at each station on the line. [Figure 6] FIG. 13 is a diagram illustrating the number of association rules extracted from a transaction. [Figure 7] FIG. 13 is a diagram showing the distribution of support and confidence for each transaction. [Figure 8] This is a chart verifying the error with the actual data for each station. [Figure 9] 13(a) to 13(i) are distribution diagrams showing the error between the stop time predicted by this embodiment and the actual data for each station. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0011] Hereinafter, an embodiment of the present invention will be described in which the present invention is applied to a train service system that provides information on the dwell times of one or more trains at each station in the near future. In this specification, information on dwell times, predicted values, and various other numerical values will be simply referred to as dwell times, predicted values, or numerical values. Furthermore, passengers will be referred to as users unless a distinction is particularly required.
[0012] [Example of train stop time estimation device configuration] Fig. 1 is an overall configuration diagram of an operation service system according to this embodiment. The operation service system includes a train stop time estimation device 1. The train stop time estimation device 1 is a computer having a communication function and a storage function connected to a network NW such as the Internet, and has a computer unit having a CPU (Central Processing Unit), a ROM (Read only memory), a RAM (Random Access Memory), an SSD (Solid State Drive), and other large-capacity storage media and input / output interfaces therefor. A computer program for operating the computer unit as the train stop time estimation device 1 is installed in the large-capacity storage media.
[0013] The network NW is also connected to a plurality of user terminals 100 carried by users, a traffic control system 200 in which each transport operator manages the traffic of its trains, a location management system 300, and a plurality of trains 500 each equipped with a communication device that outputs the current position of the train, etc. The location management system 300 is a known Web system that provides real-time current position information of the user terminals 100, trains 500, etc., but if the current position information can be obtained from the user terminals 100 or each traffic control system 200, direct access to the location management system 300 is not necessary.
[0014] Although the configurations of each traffic control system 200 are various, what they have in common is a function of collecting information outputted from on-board devices mounted on trains 500 belonging to the relevant operator, multiple station interlocking devices installed in station offices, etc., in almost real time. The traffic control system 200 makes some of the collected information open, so the train stop time estimation device 1 acquires the information as appropriate. Then, the acquired information is converted into a data structure common to each operator, integrated in a chronological order, and stored. The information integrated and stored in a chronological order includes, for example, the identification information of the train 500, the service type, the course direction (upward ID / downward ID: ID is an abbreviation for identification information, the same applies below), position information, and identification information of stations at which the train 500 departs and arrives. In addition, the information is for analyzing which train 500 is currently moving in which direction and how, where (for example, at which stop station), in what state (for example, during vehicle inspection), for how long (for example, 3 minutes have passed) it is stopped, or whether the operation mode is different from the operation plan originally assigned to the train 500, including a return operation.
[0015] 2 is a diagram showing an example of a functional configuration of the train stop time estimation device 1. The computer program causes the computer unit to function as a communication control unit 11, an information acquisition unit 12, a classification unit 13, an analysis unit 14, an estimation unit 15, an editing unit 16, an information output unit 17, and a main control unit 18, and also builds, through the storage function of the computer, a transaction DB (DB is an abbreviation for database, the same applies below) 21, a rule map DB 22, a timetable master 23, a route master 24, a map DB 25, and an object DB 26 in a large-capacity storage medium 20 accessible by the functional blocks.
[0016] The transaction DB 21 stores transactions for each line, i.e., sets of data divided based on predetermined rules, and the classification and analysis results thereof. The rule map DB 22 stores rule maps for each line, which will be described later. The timetable master 23 stores train timetables published by transportation operators. The route master 24 stores route information including station names and their locations. The map DB 25 stores map images corresponding to the route information. The object DB 26 stores symbols, icons, etc. of stations and lines.
[0017] The communication control unit 11 controls communication with the user terminal 100 etc. via the network NW. The information acquisition unit 12 acquires various information necessary for estimating train stop times. The information output unit 17 outputs information to the communication partner. The main control unit 18 controls the operation of the communication control unit 11, the information acquisition unit 12, the classification unit 13, the analysis unit 14, the estimation unit 15, the editing unit 16 and the information output unit 17 in a comprehensive or complementary manner.
[0018] The classification unit 13 classifies each of the stop times of the trains 500 that stop at the stations on the route into one of a number of stop time classes that are determined for each time lapse after arriving at the stop station. In this example, the trains are classified in descending order into classes A when the stop time of the train is between 20 and 25 seconds, B when the stop time is between 25 and 30 seconds, ..., Class F, ..., and K when the stop time is 75 seconds or more, but the class times and classification methods are not limited to these. 20 seconds is the minimum stop time published by the operator in the timetable, etc. to prevent early departure. Each classified class is called a "stop time class." The classification unit 13 also stores the classified stop time classes for each stop station in association with each of the plurality of trains 500. The storage destination may be a temporary memory (not shown) of the computer unit, or a class storage area of the transaction DB 21.
[0019] The analysis unit 14 searches for an association rule in which the stop time class of a subsequent stop station is derived due to the stop time class of one or more stop stations by analyzing the stored stop time class of each train at each stop station or their combination pattern, and also analyzes the compatibility of a new transaction with the multiple stored association rules.
[0020] An association rule is a rule that expresses the relationship that when a transaction, i.e., a group of data (items) of the stop time class of trains that stop at each station, is divided into set D based on a predetermined rule, when a certain item set X is included, another item set Y is also included at the same time, and can be expressed by the following formula. "X⇒Y", "(X,Y) ⊂ D", "X∩Y=φ(empty set)"
[0021] Here, "X" is called the condition part (LHS: Left-Hand Side) and "Y" is called the conclusion part (RHS: Right-Hand Side). The significance of an association rule can be quantified using two parameters: support and confidence. Support is the rate at which itemset X appears in set D, and confidence is the rate at which itemset Y appears when itemset X appears. For example, the support of an association rule (X⇒Y) is expressed as the rate of transactions that contain both X and Y (X∪Y) in the entire set D. Confidence is also expressed as the rate of transactions that contain both X and Y among transactions that contain X in set D (support(X∪Y) / support(X).
[0022] In this embodiment, set D is, for example, a data set of the stop times of all trains at each of multiple stations connecting one line, and when the item set X for the stop time at the station n (n is a natural number of 3 or more)-1 before the station where a train departs from a certain station will next stop is known, the relationship with the item set Y for the stop time at the station m (m>n: m is a natural number of 2 or more) before that station is estimated.
[0023] The estimation may include a process of extracting an optimal association rule by searching. Searching for an optimal association rule corresponds to, for example, finding an association rule that satisfies parameters such as the number of consecutive stations Ns, an estimation threshold ths, and a reliability threshold, i.e., minimum support and minimum confidence.
[0024] The number of consecutive stations Ns is the number of stations that inherit properties (attributes) from the previous station. The estimation threshold ths is the upper limit for determining the significance of the estimated stop time. At 100%, the estimated stop time will necessarily be shorter than the actual stop time, and in that case, a correlation rule in which passengers miss the train will also be extracted. On the other hand, at 0%, the estimation itself is meaningless. Therefore, the estimation threshold ths is set to about 25%, which is a statistical value close to the actual operation status, so that a correlation rule can be extracted in which the estimated stop time of up to 25% of trains is likely to be shorter than the actual stop time. However, when a user who wants to be provided with a more accurate estimated stop time wants to know the stop time of the train several stations before the station where the user gets off, it is possible to dynamically change the estimation threshold ths to a large value and re-extract the subsequent correlation rules, since there will be no risk of missing the train.
[0025] The minimum support is the threshold for support (first threshold) and ranges from 0.0 to 1.0. For example, when the minimum support is 0.03, association rules that appear in 3% or more of all transactions are found, and association rules that appear less than that number of times are ignored or deleted. The minimum confidence is the threshold for confidence (second threshold) and ranges from 0.0 to 1.0. For example, when the minimum confidence is 0.1, only association rules that appear with a probability of 10% or more in the stop time class at the target station to be estimated are extracted from among association rules with support below the minimum support. In other words, the extracted association rules are those that are included in an item set exceeding the minimum support and that produce a combination pattern of stopping time classes that satisfies the minimum certainty.
[0026] As a tool for searching for association rules, known tools such as "Apriori," "ECLAT," "Orange," and "RapidMiner" can be used.
[0027] The analysis unit 14 creates a rule map by applying the optimum association rules extracted as described above to all stations on the line, and stores this in the rule book DB 22 in association with the identification information of each station on the line.
[0028] Each time the current location and current time of the user terminal 100 carried by the user are newly acquired, the estimation unit 15 identifies the train the user is riding on, the line, and the next stop station, and searches for the optimal correlation rule for the subsequent station from the rule map for the line. Many correlation rules may be extracted, and in this case, it is necessary to narrow down the correlation rules to one of them. Any method can be used to narrow down the correlation rules, but here, as examples, a method of using 25% of the total occurrence frequency of each stop time class and a method of setting priorities in advance will be described.
[0029] The 25% value is based on the central limit theorem, a fundamental principle in statistics, which states that the sum of independent and identically distributed random variables approximates a normal distribution (e.g., election results). This central limit theorem says that if the sample size is 25 or more, the sample mean will approximate a normal distribution, making statistical inferences such as confidence intervals possible.
[0030] For example, suppose that there are 40 combination patterns in which the stop time classes at the last three stations of a train heading to a fourth station are class B, class C, and class B (all of which are assumed to satisfy the minimum support), and among these, the stop time classes at the next fourth station are class B 5 times (5 / 40), class C 10 times (10 / 40), class D 20 times (20 / 40), and class E 4 times (4 / 40) (all of which are assumed to satisfy the minimum support). The 25% value for the total of classes B, C, D, and E is (5+10+20+4)×0.25=9.75, which is closest to the occurrence frequency of class C, and therefore it is possible to identify one association rule that the stop time class at the next stop station after classes B, C, and B is class C.
[0031] In the method using priority, it is made to inherit the stop time class of multiple nearby stations as much as possible. For example, the correlation rule for estimating the stop time class of a train heading to the fourth station is the correlation rule that derives the stop time class of "Sta.4" following the following pattern. "Sta." is an abbreviation for station. Rank #1..."Sta.1" → "Sta.2" → "Sta.3" Rank #2: "Sta.2" → "Sta.3" Ranking #3: "Sta.1", "Sta.3" Rank #4: "Sta.1", "Sta.2" Rank #5: "Sta.3" Rank #6: "Sta.2" Rank #7: "Sta.1"
[0032] Depending on the current time and current location of the user terminal 100, the correlation rule for the next stop may not exist in the rule map DB 22. In this case, the time is estimated to be the arrival time at the next stop plus the minimum stop time, that is, class A.
[0033] The editing unit 16 edits the accumulated stop time classes and their changing modes for each station on the line so that they can be visualized for all the stop stations, and provides this to, for example, the user terminal 100 via the information output unit 17 and the communication control unit 11. For each stop station, map information stored in the map DB 25 or symbols of stop stations or trains stored in the object DB 26 may be mapped. Alternatively, editing may be performed using various visualization tools stored in the object DB 26, for example, Sankey Diagram.
[0034] [Example of user terminal configuration] 3 is a diagram showing an example of the configuration of the user terminal 100. The user terminal 100 is configured by installing an application program for analyzing operation conditions in a portable computer terminal such as a smartphone or a tablet terminal, or a notebook computer equipped with a communication function. This application program causes the portable computer terminal or the notebook computer to function as a communication unit 101, an information display unit 102 having a display screen, a first display control unit 103, a reception unit 104, a setting unit 105, a reception 106, and a second display control unit 107.
[0035] The communication unit 101 communicates with the train stop time estimation device 1 or the location management system 300. The information display unit 102 is, for example, a display using a touch panel, and displays information such as images through a client browser. The first display control unit 103 displays various setting screens on the touch panel. The reception unit 104 receives the setting screen from the user. The setting unit 105 sets the setting information in the train stop time estimation device 1. The reception unit 106 receives various information output from the train stop time estimation device 1. The second display control unit 107 displays the received various information on the information display unit 102.
[0036] [verification] Next, a description will be given of the estimation accuracy of the stop time estimation device 1 configured as above. Here, an example of the estimation result of the stop time at each station when nine trains (local trains) arrive and depart from the stations during the weekday morning rush hour on one line connecting ten stations operated by the applicant of the present application will be described. Several universities and high schools are located on this line, and the daily needs of students change. As a result, the stop time of trains at each station changes significantly from day to day. The changes in train stop time are caused by the degree of congestion, reasons for train delays, distances between stations, changes in assigned trains, the presence or absence of special events such as entrance exams, and the experience of crew members. These can be averaged out by statistically analyzing the changes in stop time classes at multiple stations for multiple trains over a given period, for example, six months.
[0037] Therefore, in this example, a dataset for the six months from April 2022 to September 2022 was accumulated with 1,598 rows (one row is an item at all stations except the starting station of one train) as training transactions. Figures 4(a) to (i) are distribution diagrams of stop times at each station during the above period. Figure 4(a) is a distribution diagram for the second station from the starting station, with the horizontal axis representing stop times (seconds) and the vertical axis representing the proportion (%) of trains that had such stop times relative to all trains. Figures 4(b) to 4(i) are distribution diagrams for successive stations, respectively.
[0038] In addition, in the timetable, there are one or more stations to which a margin stop time is added. "Margin stop time" refers to a static time that is added to the stop time in advance to allow for a margin of time when there is a possibility of delay from the scheduled time due to train delays or operating conditions. In this example, in order to dynamically and correctly estimate the correlation between items, this margin stop time is ignored (deleted) as much as possible when acquiring transactions.
[0039] The change in the stop time class of each station classified by the classification unit 13 is visualized as shown in FIG. 5. FIG. 5 is a Sankey Diagram, which is an example of a visualization tool. In the diagram, "Sta." is an abbreviation for station, "Sta.1" is the first station on the line, i.e., the starting station, "Sta.10" is the tenth station, i.e., the terminal station, and "Sta.2" to "Sta.9" represent intermediate stops. In addition, the square frames at each station and B to K written in their vicinity represent the stop time class at the station. The length of each square frame represents the number of trains belonging to the same stop time class. The belt-shaped links connecting stations represent the flow of change in the stop time class between stations, and the width of the belt represents the number of trains. In this way, by visualizing the stop time classes at each station and the links between stations and presenting them on the user terminal 100, the user can intuitively grasp, for example, the degree of variation in stop times at each station and the tendency of trains to be delayed at each station.
[0040] In searching and extracting association rules, the number of consecutive stations Ns was set to 3, the estimation threshold ths to 0.25, the minimum support to 0.003, and the minimum confidence to 0.1. The priority order was set so that the stopping time classes of multiple and nearby stations could be inherited as much as possible. Figure 6 shows the number of association rules extracted from transactions under the above parameters. In Figure 6, the horizontal axis is confidence and the vertical axis is the number of association rules. Additionally, station1, station2, and station3 are the numbers of stations to be taken over. A total of 2,830 association rules were extracted, and it can be seen that association rules with a confidence level of over 50% were obtained. Moreover, the distribution of support and confidence for each transaction is shown in Fig. 7. It can be seen from Fig. 7 that by inheriting the properties of multiple stations, it is possible to extract high-quality association rules even when the number of consecutive stations Ns is small and the support is low.
[0041] In addition, a data set for one month in October 2022 for the same line, the same station, and the same train as the above training transaction was obtained as a test transaction, and the results of estimation using exactly the same parameters as this example are shown in Figure 8. Here, two reference methods 1 and 2 were used to make it easier to verify the accuracy of this example (Proposed algorithm). Reference method 1 classifies the stop time at each station into classes in the same manner as this example, and then calculates the difference between the actual stop time at each station and the 25% value of the past stop time data (a number indicating the number counting from the smallest, with the whole being 100, and the 50th percentile being the median). Reference method 2 is the difference between the 25% value estimated from the stop time pattern of the station one station before the target station and the actual stop time.
[0042] This chart shows the average, median, and standard error of the difference between the actual stop time and the estimated stop time. Figure 8 shows that the average and median values for each station estimated by this example are closer to the actual stop time at almost all stations than the values estimated by reference methods 1 and 2. Moreover, the variance in the distribution of stop times is smaller.
[0043] Figures 9(a) to (i) are distribution diagrams showing the error between the stop time predicted by the train stop time estimation device 1 and the actual data for each station. Figure 9(a) is a distribution diagram for the second station from the first station, with the horizontal axis showing the stop time (seconds) and the vertical axis showing the proportion (%) of trains with such stop times to all trains. Figures 9(b) to 9(i) are distribution diagrams for successive stations, respectively. As can be seen from these diagrams, sufficient prediction accuracy is obtained for most stations.
[0044] As described above, in this embodiment, it is possible to estimate the stop times at each station in the near future by using correlation rules derived from the analysis results of past transactions, for example, by using correlation rules in a rule map stored in rule map DB22, so that it becomes easy to predict train operation in real time for each station or for multiple stations. In this embodiment, the stopping time can be estimated using only the current stopping times at some stations, which is particularly useful for realizing real-time estimation processing. In addition, by repeatedly extracting optimal correlation rules using a rule map created in advance, it becomes possible to more accurately estimate the arrival and departure times of the train at the following stations each time the train departs from a station. For example, it becomes possible to easily estimate the stopping times of a train that has just departed at the nearest station and future stations at that time.
[0045] The proposed method based on the pattern of stop times between stations can be effectively used even when timetables are revised or train diagrams are temporarily changed, as long as the pattern of train stops does not change due to the opening of new stations, etc. In addition, the estimation threshold and minimum confidence can be set arbitrarily and dynamically changed, making it applicable to various use cases.
[0046] [Variations] In this embodiment, an example is given in which a Sankey Diagram is used as a visualization tool for changes in stop time classes, but other diagrams may be used. Also, an image in which the train's operating route and the stop time at each station are mapped on a route map may be used. In this case, the detailed operating route of the train may be highlighted, and the station or train may be replaced with a symbol read from the object DB. In this embodiment, an example has been described in which the transportation vehicle in which the user boards is a train, but the present invention can also be applied to transportation vehicles other than trains, such as buses, passenger ships, and airplanes. [Explanation of symbols]
[0047] 1 train stop time estimation device, 11 communication control unit, 12 information acquisition unit, 13 classification unit, 14 analysis unit, 15 estimation unit, 16 editing unit, 17 information output unit, 18 main control unit, 20 large-capacity storage medium, 21 transaction DB, 22 rule map DB, 100: User terminal, 200: Traffic control system, 500: Trains.
Claims
1. A device for estimating a stop time of a train that arrives at one of a plurality of stop stations connecting a predetermined section until it departs for a subsequent stop station, A classification means for classifying each of the stop times of the trains that stop at the stop stations into one of a plurality of stop time classes that are determined based on the time lapse after the train arrives at the stop station; a storage means for storing the classified stop time classes in association with each of the plurality of trains for each stop station; an analysis means for searching for an association rule in which the stop time class of a subsequent stop station is derived due to the stop time class of one or more stop stations by analyzing a combination pattern of the stop time class of each train at each stop station that is stored; an estimation means for estimating the stop time of a train arriving at the subsequent stop station based on the correlation rule; A traffic vehicle stopping time estimation device comprising:
2. 2. The stopping time estimation device according to claim 1, wherein said analysis means creates a rule map including a plurality of said correlation rules, and extracts from said rule map a correlation rule corresponding to a current time and a current position of any one of said trains.
3. The condition parameters used for extracting the association rules are dynamically changeable. The vehicle stopping time estimating device according to claim 2.
4. A priority order is defined for preferentially using the extracted association rules. The vehicle stopping time estimating device according to claim 3.
5. The method further includes a visualization means for visualizing the stop time class and its change state stored in the storage means for each stop station. The vehicle stopping time estimation device according to claim 1 .
6. A computer program causing a computer to operate as the stopping time estimation device according to any one of claims 1 to 5.
Citation Information
Patent Citations
Train operation support system and method for the same
JP2022042138A
