A data-driven subway train operation performance simulation method and system

Abstract

The present invention discloses a method and system for simulating subway train operation performance based on data-driven, and relates to the technical field of urban rail transit. The method comprises: obtaining the planned operation data of each train from the actual planned operation diagram, and inputting each type of data in the planned operation data into the operation time prediction model, the train stop time prediction model and the train return time prediction model of the corresponding date type according to the time sequence according to the date type of the actual planned operation diagram for processing, and respectively obtaining the predicted operation time, the stop prediction time and the return prediction time; replacing the corresponding original data in the planned operation data with the predicted operation time, the stop prediction time and the return prediction time corresponding to each time step, and obtaining the target calculation operation diagram according to the new planned operation data; and accurately simulating the actual operation diagram corresponding to the planned operation diagram, so as to more comprehensively and accurately predict the actual operation status of the train.

Description

Technical Field

[0001] The present invention relates to the technical field of urban rail transit, and more specifically, to a data-driven subway train operation performance deduction method and system. Background Art

[0002] The train operation plan is the core technical document for the daily operation of the subway. It specifies in detail the time of entry and exit of all trains every day, the time of running in each section, the stop time at each station, and the turnaround time at each turnaround station. However, in actual operation, trains are inevitably disturbed by a variety of realistic factors, resulting in random deviations between the actual operation of each train and the planned operation plan. When this deviation exceeds a certain range, it will not only disrupt the subsequent driving plan, but may even cause the entire plan to fail to execute smoothly. Therefore, it is particularly important to grasp the deviation between the actual operation of the train and the plan in real time. Timely and accurate deduction of the actual operation plan of the subway train through train operation performance deduction can effectively help the operator adjust and optimize the planned operation plan, thereby ensuring the smooth operation of the entire subway system and improving operational efficiency and passenger satisfaction.

[0003] At present, although the specialized technology for the performance prediction of urban rail transit train operation is not yet mature, some scholars have tried to use the LSTM-KF model (i.e., a model combining long short-term memory network and Kalman filter) to predict the running trajectory of trains in the section. LSTM network is good at processing time series data. By analyzing the historical operation data of the train, it generates preliminary prediction results; then, the preliminary prediction results are further optimized in combination with the Kalman filter algorithm to achieve the prediction of train trajectories in a longer time domain. The above technical solutions mainly focus on predicting the operation of trains in the section, and cannot predict key nodes such as stop time and return time from the overall perspective of the line. Factors such as stop time and return time are affected by complex variables such as passenger flow and scheduling arrangements, and are discrete and sudden, and the LSTM-KF model is difficult to capture the characteristics of these data. In addition, although LSTM-KF has its advantages in section prediction, it lacks global prediction capabilities. To achieve more comprehensive and accurate performance prediction of urban rail transit trains, it is necessary to introduce a machine learning model that can handle complex multivariables.

[0004] Some scholars have also introduced algorithms such as random forest and XGBoost, which are used for railway train delay prediction, into the delay prediction of urban rail transit trains. This method is based on the train operation performance data of a subway company for one month on a line to analyze the spatial and temporal distribution of train delays. Through the feature selection module, the XGBoost algorithm is used to determine the important features of delay prediction, and a first-level prediction model based on LSTM is established. The second-level prediction model is combined with random forest and XGBoost to optimize the delay prediction results. The model parameters are adjusted through cross-validation and grid search, and the optimal model is finally selected for the prediction of the train's next station delay time. The above can accurately predict the train delay time at each station, but this technology processes the train operation data in the up and down directions separately, constructs a train arrival delay prediction model in a single direction, and predicts the delay situation of each station, while ignoring important factors that affect the integrity and continuity of train operation, such as turnaround scheduling and route arrangement, and cannot achieve a comprehensive deduction of the train operation performance as a whole. Summary of the invention

[0005] The purpose of the present invention is to provide a data-driven subway train operation performance simulation method and system, which accurately simulates the actual operation diagram corresponding to the planned operation diagram based on multi-source data such as line operation parameters, train historical operation data and train planned operation diagram, so as to more comprehensively and accurately predict the actual operation status of the train.

[0006] The above technical objectives of the present invention are achieved through the following technical solutions:

[0007] In a first aspect, the present application provides a data-driven subway train operation performance simulation method, comprising the following specific steps:

[0008] Obtain the train operation performance data for multiple days of each date type, and generate the historical operation data of each day of the corresponding date type. The historical operation data includes the operation path of each train, and the planned arrival time of each train at each station on the corresponding path, the planned departure time from each station, the actual arrival time at each station, and the actual departure time from each station;

[0009] Extract from the historical operation data the characteristic data of the running time of each train during the interval operation, the characteristic data of the turning time of each train when located on the turning line, and the characteristic data of the stop time of each train when stopping at each station, each characteristic data includes a characteristic value and a label value;

[0010] Outlier elimination processing is performed on each feature data respectively, and based on each feature data after outlier elimination processing, the corresponding preset initial models are trained respectively until each preset initial model reaches the training end condition, and each preset initial model that reaches the training end condition is determined as a running time prediction model, a train stop time prediction model, and a train turnaround time prediction model respectively;

[0011] The planned operation data of each train is obtained from the actual planned operation diagram, and according to the date type of the actual planned operation diagram, each type of data in the planned operation data is input into the operation time prediction model, train stop time prediction model, and train turnaround time prediction model of the corresponding date type in time sequence for processing, and the predicted operation time, stop prediction time, and turnaround prediction time are obtained respectively;

[0012] The corresponding original data in the planned operation data are replaced by the predicted operation duration, stop prediction time and return prediction time corresponding to each time step to obtain new planned operation data, and the target calculation operation diagram is obtained according to the new planned operation data.

[0013] Based on the above technical solution, the present invention can also be improved as follows.

[0014] Furthermore, the above-mentioned characteristic values ​​are planned values ​​in the historical operation data, and the label values ​​are actual values ​​corresponding to the characteristic values.

[0015] Furthermore, the above-mentioned running time prediction model is specifically as follows:

[0016] in:

[0017]

[0018] or

[0019] Where, T α represents the time period number, j represents the interval number, d represents the running direction number, d=1 represents the downlink direction, d=2 represents the uplink direction, ΔR e (j,n) represents the time deviation of train number n entering interval j, R p (j,n) represents the planned running time of train number n in section j, ΔR l (j,n) represents the time deviation of train number n leaving interval j, R a (j,n) represents the actual running time of train number n in section j. represents the predicted running time of train number n in interval j, M represents the total number of calculations, and α m Indicates the weight of each calculation result, h m (fj θ m ) represents the prediction result of the mth calculation, f j Represents the feature vector of all feature data in interval j, θ m represents f j A collection of parameters.

[0020] Furthermore, the above train stop time prediction model is specifically as follows:

[0021] in:

[0022]

[0023] or

[0024] Where, T α represents the time period number, i represents the station number, d represents the running direction number, d=1 represents the down direction, d=2 represents the up direction, S p (i,n) represents the planned stop time of train number n at station i, ΔS a (i,n) represents the stop time deviation of train number n at station i, S a (i,n) represents the actual stop time of train number n at station i. represents the predicted stop time of train No. n at station i, M represents the total number of calculations, α m Indicates the weight of each calculation result, h m (f i θ m ) represents the prediction result of the mth calculation, f i The feature vector representing all feature data of station i, θ m represents f i A collection of parameters.

[0025] Furthermore, the above train turnaround time prediction model is specifically as follows:

[0026] in:

[0027] Where, T α represents the time period number, i represents the station number, d represents the running direction number, d=1 represents the down direction, d=2 represents the up direction, B p (i,n) represents the planned return time of train number n at station i, B a (i,n) represents the actual return time of train number n at station i, ΔB e (i,n) represents the return time deviation of train number n at station i, ΔB f(i,n) represents the time deviation of train No. n completing the turnaround at station i. represents the predicted return time of train No. n at station i, M represents the total number of calculations, α m Indicates the weight of each calculation result, h m (g i θ m ) represents the prediction result of the mth calculation, g i The feature vector representing all feature data of station i, θ m Indicates g i A collection of parameters.

[0028] Furthermore, the earliest time when a train goes online in the historical operation data is determined as the target time when a train goes online in the planned operation data, and the latest time when a train goes offline in the historical operation data is determined as the target time when a train goes offline in the planned operation data after extending the preset time period.

[0029] In a second aspect, the present application provides a data-driven subway train operation performance deduction system, which is applied to a data-driven subway train operation performance deduction method according to any one of the first aspects, including:

[0030] The first module is used to obtain the train operation performance data of multiple days for each date type, and generate the historical operation data of each day for the corresponding date type. The historical operation data includes the operation path of each train, and the planned arrival time of each train at each station on the corresponding path, the planned departure time from each station, the actual arrival time at each station, and the actual departure time from each station;

[0031] The second module is used to extract characteristic data of the running time of each train during interval operation, characteristic data of the turning time of each train when it is located on the turning line, and characteristic data of the stop time of each train when it stops at each station from the historical operation data, and each characteristic data includes a characteristic value and a label value respectively;

[0032] The third module is used to perform outlier elimination processing on each feature data, and train the corresponding preset initial models based on each feature data after outlier elimination processing, until each preset initial model reaches the training end condition, and each preset initial model that reaches the training end condition is determined as a running time prediction model, a train stop time prediction model, and a train turnaround time prediction model;

[0033] The fourth module is used to obtain the planned operation data of each train from the actual planned operation diagram, and according to the date type of the actual planned operation diagram, input each type of data in the planned operation data into the operation time prediction model, train stop time prediction model, and train turnaround time prediction model of the corresponding date type in time sequence for processing, and obtain the predicted operation time, stop prediction time, and turnaround prediction time respectively;

[0034] The fifth module is used to replace the corresponding original data in the planned operation data with the predicted operating time, stop prediction time and return prediction time corresponding to each time step, obtain new planned operation data, and obtain the target calculation operation diagram based on the new planned operation data.

[0035] Furthermore, the fourth module further includes:

[0036] The first submodule is used to determine the earliest time when a train goes online in the historical operation data as the target time when a train goes online in the planned operation data, and to determine the latest time when a train goes offline in the historical operation data as the target time when a train goes offline after extending the preset time period.

[0037] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any one of the methods in the first aspect when executing the computer program.

[0038] In a fourth aspect, the present application provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions enable a computer to execute any one of the methods in the first aspect.

[0039] Compared with the prior art, the present invention has at least the following beneficial effects:

[0040] In the present application, firstly, characteristic data of each type of data is obtained from historical operation data, and the corresponding preset initial model is trained using the characteristic data of each type of data, and in the training process, at least 3 consecutive associated data are used for training, so that the model can perform a more efficient learning process, thereby learning the difference between the planned value and the actual value and the law causing the difference at a deeper level; secondly, when each preset initial model reaches the maximum number of iterations or the corresponding loss function does not exceed the threshold, each preset initial model completes the training process, and in actual use, each type of data in the planned operation data is input into the corresponding and trained model for processing, so as to obtain the prediction results corresponding to each type of data; finally, the corresponding original data in the planned operation data is replaced by the prediction results corresponding to each type of data, so as to obtain new planned operation data, and finally the target calculation operation diagram is obtained according to the new planned operation data, and the target calculation operation diagram is determined as the final planned operation diagram of the day; based on multi-source data such as line operation parameters, historical train operation data and planned train operation diagram, the actual operation diagram corresponding to the planned operation diagram is accurately deduced, so as to more comprehensively and accurately predict the actual operation status of the train.

[0041] In this application, the method relies on the historical operation data of a certain subway line and uses machine learning technology to establish a data-driven model; when a new train plan operation diagram is input, the model can deduce the actual operation performance of the diagram after actual use, including the actual arrival time and actual departure time of the train at each station on the operation path, etc. When compiling a new plan operation diagram, the subway transportation planning department can use this technology to quickly and efficiently evaluate the feasibility, operation effect and fulfillment rate of the plan operation diagram, so as to compile a train plan operation diagram that better meets actual needs; during the implementation of the operation diagram, the subway dispatching department can use this technology to timely predict the subsequent operation status of each train, and then timely adjust the train dispatching strategy, thereby improving operational management efficiency and the ability to respond to emergencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0043] Figure 1 A method flow chart of the deduction method in an embodiment of the present invention;

[0044] Figure 2 It is an overall flow chart of the subway train operation performance simulation in an embodiment of the present invention;

[0045] Figure 3 A connection diagram of a deduction system in an embodiment of the present invention;

[0046] Figure 4 A schematic diagram of the connection of electronic equipment in an embodiment of the present invention;

[0047] Figure 5 It is a schematic diagram showing the comparison between the planned graph and the deduced actual graph within a certain time period in an embodiment of the present invention;

[0048] Figure 6 It is a schematic diagram for comparing the deviations of the arrival and departure times between the deduced actual graph and the planned graph, and between the real actual graph and the planned graph in an embodiment of the present invention;

[0049] Figure 7 It is a schematic diagram for comparing a real actual graph and a real-time deduced actual graph in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0051] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0053] In the description of the embodiments of the present invention, "plurality" means at least 2.

[0054] Embodiment 1: In order to accurately deduce the actual operation diagram corresponding to the planned operation diagram, so as to more comprehensively and accurately predict the actual operation status of the train, this embodiment provides a data-driven subway train operation performance deduction method, such as Figure 1 As shown, the following specific steps are included:

[0055] S1, obtain the train operation performance data of multiple days for each date type, and generate the historical operation data of each day for the corresponding date type. The historical operation data includes the operation route of each train, and the planned arrival time of each train at each station on the corresponding route, the planned departure time from each station, the actual arrival time at each station and the actual departure time from each station.

[0056] Among them, according to the date type (for example, weekdays or weekends) for the proposed train operation performance simulation, the date range for historical data selection is set (for example, all weekdays or all weekends in the past period of time), and the daily train operation performance data within the date range can be extracted from the ATS system to generate a basic data table of train historical operations (the data information includes: the operation route of each train, the planned arrival time, planned departure time, actual arrival time and actual departure time of each train at each station on its route).

[0057] S2, extracting from the historical operation data the characteristic data of the running time of each train during section operation, the characteristic data of the turning time of each train when located on the turning line, and the characteristic data of the stop time of each train when stopping at each station, each characteristic data includes a characteristic value and a label value.

[0058] Among them, the characteristic value is the planned value in the historical operation data, and the label value is the actual value corresponding to the characteristic value. For different types of data, the corresponding characteristic data is also different, which will be further explained below.

[0059] S3, perform outlier elimination processing on each feature data, and train the corresponding preset initial models based on each feature data after outlier elimination processing, until each preset initial model reaches the training end condition, and each preset initial model that reaches the training end condition is respectively determined as the running time prediction model, the train stop time prediction model, and the train turnaround time prediction model.

[0060] The meaning of each character can be expressed as: time period number T α (the hour in which the train is scheduled to arrive at the next station), section number j, running direction number d (when d=1, it indicates the downward direction, and the train runs in the direction from small to large section numbers; when d=2, it indicates the upward direction, and the train runs in the direction from large to small section numbers), train number n; the time deviation ΔR of train number n entering section j e (j,n) and the planned running time R of the j interval p (j,n).

[0061] Furthermore, the duration, deviation, and time of three consecutive intervals are used as feature data, which can be expressed as:

[0062] The planned running time of the rear section that has passed through three sections, when d = 1, is R p (j-1,n),R p (j-2,n),R p (j-3,n); when d=2, it is R p (j+1,n), Rp (j+2,n), R p (j+3,n);

[0063] The actual running time of the rear vehicle after passing through three sections, when d = 1, is R a (j-1,n),R a (j-2,n),R a (j-3,n); when d=2, it is R a (j+1,n), R a (j+2,n), R a (j+3,n);

[0064] The time deviation of entering the interval after having passed three intervals, when d = 1, is ΔR e (j-1,n),ΔR e (j-2,n),ΔR e (j-3,n); when d=2, it is ΔR e (j+1,n),ΔR e (j+2,n),ΔR e (j+3,n);

[0065] The departure time deviation of the three sections passed behind is ΔR when d = 1. l (j-1,n),ΔR l (j-2,n),ΔR l (j-3,n); when d=2, it is ΔR l (j+1,n),ΔR l (j+2,n),ΔR l (j+3,n).

[0066] Furthermore, the calculation formula for the time deviation of entering the interval is:

[0067] ΔR e (j,n)=S al (j,n)-S pl (j,n)(d=1); or ΔR e (j,n)=S al (j+1,n)-S pl (j+1,n)(d=2);

[0068] Among them, S pl (j,n)(or S pl (j+1,n)) is the planned departure time of the next station in section j, S al (j,n)(or S al (j+1,n)) is the actual departure time of the next station in section j.

[0069] Furthermore, the calculation formula for the time deviation of leaving the interval is:

[0070] ΔR l (j,n)=S aa (j+1,n)-S pa (j+1,n)(d=1); or ΔR l (j,n)=S aa (j,n)-S pa (j,n)(d=2);

[0071] Among them, S pa (j+1,n)(or S pa (j,n)) is the planned arrival time of the next station in section j, S aa (j+1,n)(or S aa (j,n)) is the actual arrival time of the next station in section j.

[0072] Further, calculate R a (j,n), that is, the actual running time of the current interval, is calculated as follows:

[0073] R a (j,n)=S aa (j+1,n)-S al (j,n)(d=1), or R a (j,n)=S aa (j,n)-S al (j+1,n)(d=2).

[0074] Specifically, the feature values ​​and label values ​​are screened through statistical analysis, their standard deviations are calculated, and it is defined that if the sample data exceeds the range of 3 times the standard deviation, it is regarded as an outlier and is eliminated.

[0075] Optionally, the above-mentioned running time prediction model is specifically:

[0076] in:

[0077]

[0078] or

[0079] Where, T α represents the time period number, j represents the interval number, d represents the running direction number, d=1 represents the downlink direction, d=2 represents the uplink direction, ΔR e (j,n) represents the time deviation of train number n entering interval j, R p (j,n) represents the planned running time of train number n in section j, ΔR l(j,n) represents the time deviation of train number n leaving interval j, R a (j,n) represents the actual running time of train number n in section j. represents the predicted running time of train number n in interval j, M represents the total number of calculations, and α m Indicates the weight of each calculation result, h m (f j θ m ) represents the prediction result of the mth calculation, f j Represents the feature vector of all feature data in interval j, θ m represents f j A collection of parameters.

[0080] Specifically, the corresponding model is further explained at each time point of three consecutive stations:

[0081] Each definition can be expressed as: Time period number T α , station number i, running direction number d, train number n; planned stop time S at station i p (i,n) and arrival time deviation ΔS a (i,n).

[0082] Among them, the planned stop times of the three trains before station i can be expressed as S p (i,n-1),S p (i,n-2),S p (i,n-3); the arrival time deviations of the three trains before station i can be expressed as ΔS a (i,n-1),ΔS a (i,n-2),ΔS a (i,n-3); the actual stop time of the three trains before station i is S a (i,n-1),S a (i,n-2),S a (i,n-3).

[0083] Furthermore, when the planned stop time of train No. n has passed three stations ahead, when d = 1, it is S p (i-1,n),S p (i-2,n),S p (i-3,n); when d=2, it is S p (i+1,n),S p (i+2,n),S p (i+3,n).

[0084] The arrival time deviation of train No. n, which has passed three stations before, is ΔS when d = 1. a(i-1,n),ΔS a (i-2,n),

[0085] ΔS a (i-3,n); when d=2, it is ΔS a (i+1,n),ΔS a (i2,n),ΔS a (i+3,n);

[0086] The actual stop time of train No. n after passing three stations ahead, when d = 1, is S a (i-1,n),S a (i-2,n),

[0087] S a (i-3,n); when d=2, it is S a (i+1,n),S a (i+2,n),S a (i+3,n).

[0088] The calculation formula for arrival time deviation is: ΔS a (i,n)=S aa (i,n)-S pa (i,n); where S aa (i,n) is the actual arrival time of station i, S pa (i,n) is the planned arrival time of station i.

[0089] The actual stop time S of train No. n at station i a (i,n), the calculation formula is: S a (i,n)=S al (i,n)-S aa (i,n); where S al (i,n) is the actual departure time of station i, S aa (i,n) is the actual arrival time of station i.

[0090] Optionally, the train stop time prediction model is specifically:

[0091] in:

[0092]

[0093] or

[0094] Where, T α represents the time period number, i represents the station number, d represents the running direction number, d=1 represents the down direction, d=2 represents the up direction, S p(i,n) represents the planned stop time of train number n at station i, ΔS a (i,n) represents the stop time deviation of train number n at station i, S a (i,n) represents the actual stop time of train number n at station i. represents the predicted stop time of train No. n at station i, M represents the total number of calculations, α m Indicates the weight of each calculation result, h m (f i θ m ) represents the prediction result of the mth calculation, f i The feature vector representing all feature data of station i, θ m represents f i A collection of parameters.

[0095] Specifically, the corresponding model is further explained at each time point of three consecutive stations:

[0096] The parameters mean: time period number T α , station number i, running direction number d, train number n; the planned return time of the station is B p (i,n) and the entry return time deviation ΔB e (i,n).

[0097] Furthermore, the planned return time for the first three completed return trips is B p (i,n-1),B p (i,n-2), B p (i,n-3); the actual return time of the first three completed return trips can be B a (i,n-1),B a (i,n-2), B a (i,n-3); The entry and exit time deviation of the first three completed return trips can be ΔB e (i,n-1), ΔB e (i,n-2), ΔB e (i,n-3); The completion time deviation of the first three completed return trips can be ΔB f (i,n-1), ΔB f (i,n-2), ΔB f (i,n-3).

[0098] Therefore, the calculation formulas for the entry return time deviation and the completion return time deviation are:

[0099] ΔB e (i,n)=B ae (i,n)-B pe (i,n), ΔBf (i,n)=B af (i,n)-B pf (i,n);

[0100] Among them, B ae (i,n) is the actual entry and reversal time of reversal station i, B pe (i,n) is the planned entry time of the reversing station i, B af (i,n) is the actual completion time of the return station i, B pf (i,n) is the planned completion time of the reversal station i.

[0101] Furthermore, the actual return time of train No. n at the return station i is B a (i,n), the calculation formula is:

[0102] B a (i,n)=B af (i,n)-B ae (i,n).

[0103] Optionally, the above train turnaround time prediction model is specifically:

[0104] in:

[0105] Where, T α represents the time period number, i represents the station number, d represents the running direction number, d=1 represents the down direction, d=2 represents the up direction, B p (i,n) represents the planned return time of train number n at station i, B a (i,n) represents the actual return time of train number n at station i, ΔB e (i,n) represents the return time deviation of train number n at station i, ΔB f (i,n) represents the time deviation of train No. n completing the turnaround at station i. represents the predicted return time of train No. n at station i, M represents the total number of calculations, α m Indicates the weight of each calculation result, h m (g i θ m ) represents the prediction result of the mth calculation, g i The feature vector representing all feature data of station i, θ m Indicates g i A collection of parameters.

[0106] S4, obtaining the planned operation data of each train from the actual planned operation diagram, and according to the date type of the actual planned operation diagram, inputting each type of data in the planned operation data into the operation time prediction model, train stop time prediction model, and train turnaround time prediction model of the corresponding date type in time sequence for processing, and obtaining the predicted operation time, stop prediction time, and turnaround prediction time, respectively.

[0107] Optionally, the earliest time when a train comes online in the historical operation data is determined as the target time when a train comes online in the planned operation data, and the latest time when a train goes offline in the historical operation data is determined as the target time when a train goes offline in the planned operation data after extending the time by a preset period of time.

[0108] Among them, all the data of the planned operation diagram are input, assuming that the first trains in each direction are on-line on time according to the planned on-line time, and the earliest train on-line time is extracted from all the historical operation results, and this time is recorded as the starting time t (for example, if the earliest on-line time of the historical train is 04:55:00, then t=0 represents 04:55:00); extract the latest train off-line time and extend it by half an hour, and record this time as the end time T (for example, if the latest off-line time of the historical train is 01:20:00, then T is the difference between 01:20:00 and 04:55:00 plus 30 minutes, that is, T=75300 seconds), and set t=0 before the start of the deduction. In addition, let n represent the train number, N represent the total number of all trains running on-line at a certain time, and n=1 at the beginning of the deduction.

[0109] Furthermore, the stop time of all trains at the station is estimated; at time t, train n is at station i, including the following four situations:

[0110] (a): Train n has just left the section and arrived at the first station i. At this time, the actual arrival time S of train n is recorded. aa (i,n)=t, and move the train into the set γ (γ represents the set of all trains at the station), and let N=N+1, and calculate the stop time through the train stop time prediction model And record.

[0111] (b): Train n has just left the section and entered the intermediate station i. At this time, the actual arrival time S of train n is recorded. aa (i,n)=t, and move the train n from the set β to the set γ (β represents the set of all trains in the interval), and calculate the stop time through the train stop time prediction model And record.

[0112] (c): Train n has stopped at station i. And record. When , if station i is not the terminal station, move it from set γ to set β; if station i is the terminal station, move it from set γ to set δ (δ represents the set of all trains on the return line). Record the actual departure time S of train n al (i,n)=t+1(or the actual time of entering the return point B ae (i,n)=t+1).

[0113] (d) Train n arrives at the first station after completing the turnaround. At this time, the actual arrival time S of train n is recorded first. aa (i,n) = t, and move the train from set δ to set γ, and calculate the stop time through the train stop time prediction model And record.

[0114] Furthermore, the interval running time of all trains running in the interval is estimated; at time t, train n is located in interval j, including the following two situations:

[0115] (a): Train n has just entered section j. At this time, the actual time of train n entering the section (i.e. the actual departure time of the current station) is recorded as S al (j,n)=t(the uplink direction is S al (j+1,n)=t), and move the train from set γ to set β, and calculate the interval running time through the running time prediction model And record.

[0116] (b): When train n runs in section j but does not reach the station, let And record. Because there needs to be a certain safety time interval T between adjacent trains. 运间 , so if there is a train at the station behind interval j, the interval running time like When , the train is moved from set β to set γ, and the actual departure time of the train (i.e. the actual arrival time at the rear station) S is recorded. aa (j+1,n)=t+1(or S aa (j,n)=t+1).

[0117] Furthermore, the turnaround time of all trains on the turnaround line is calculated. At time t, the train n is on the turnaround line in the following two cases:

[0118] (a): When train n just enters the turnaround, record the actual time B of train n entering the turnaround ae (i,n) = t, and move the train from set γ to set δ, and calculate the turnaround time through the train turnaround time prediction model And record.

[0119] (b): When train n has already reversed at the reversing station, let and record. Since adjacent trains need to satisfy a certain safety time interval T when reversing 折间 , so if there is a train reversing at the reversing station, the remaining reversing time When , then move this train from set δ to set γ, record the actual reversing completion time B of the train af T(i,n) = t and the actual arrival time S when the train arrives at the originating station aa T(i,n) = t + 1.

[0120] S5. Use the predicted running duration, predicted stop time, and predicted reversing time corresponding to each time step to replace the corresponding original data in the planned operation data, obtain new planned operation data, and obtain the target calculated operation diagram according to the new planned operation data.

[0121] Among them, update all train states and time data, such as stop time, section running time, reversing time, etc., through the predicted outputs of the above running time prediction model, train stop time prediction model, and train reversing time prediction model; if train n is to go offline and return to the depot, then when it arrives at the terminal station and the stop time , remove this train from the total number of car bodies currently in operation (i.e., N = N - 1); and the deduction system is updated in steps of 1 second, that is, let t = t + 1 until t < T is not established, at this time it means the deduction ends, and the deduced actual operation results are output in the form of an actual operation diagram, train timetable, etc.

[0122] The following further illustrates the deduction method in the present application through an example: A subway line in a certain city is used as an example. The line is 24.4 km long and the operation time is from 06:00 in the morning to 23:50 in the evening; a total of 22 stations are set up along the whole line and most of them are underground stations. The station numbers are 1 to 22 in sequence (it is stipulated that the down direction is from station 1 to station 22); there are 21 intervals on this line. The detailed information of each station and interval is shown in Table 1 and Table 2 respectively, and it is carried out according to the Figure 2 overall flow chart of the deduction shown.

[0123] Table 1 Station Information Table

[0124]

[0125] Table 2 Interval Information Table

[0126]

[0127] Among them, the planned operation diagram on January 14 of a certain year can be selected as the deduction target. According to the date type of that day, the train operation performance data of all weekends from November 4 of the previous year to January 13 of the current year are extracted from the ATS system to generate the basic data table of train historical operation (classified by date). The data format is shown in Table 3.

[0128] Table 3 Basic data table of historical train operation on a certain day

[0129]

[0130] By analyzing the specific time information in the basic data table of historical train operation, parameters such as stop time, section operation time, and return time can be calculated, thereby providing a basis for the establishment of subsequent prediction models.

[0131] Furthermore, the daily historical train operation basic data table is used to train the machine learning model for the running time of each section of the train, the stop time of each station, and the return time of the return station. According to different section numbers, station numbers, and train running directions, corresponding prediction models are established respectively; after the model training is completed, the best prediction model for each section, station, and return station is generated, and its optimal algorithm is determined. The specific contents are shown in Table 4, Table 5, and Table 6.

[0132] Table 4 Summary of mean absolute error in each interval

[0133]

[0134] Table 5 Summary of the average absolute error of each station

[0135]

[0136]

[0137] Table 6 Summary of the average absolute error of each turnaround station

[0138] Turnaround Station Name Return direction Mean absolute error / second Station 1 Uplink-Downlink 3.73 Station 22 Downlink-Uplink 2.03

[0139] The best model obtained through training is saved for each station, section and turnaround station, so that it can be called in the subsequent simulation steps. Specifically, the data of the planned operation diagram is input into the simulation system and the parameters such as t and T are assigned. The established stop time prediction model, section operation time prediction model and turnaround time prediction model are called to simulate the actual train operation performance, and finally generate the full-day train operation performance diagram. Figure 5 As shown (where the black dotted line is the planned diagram and the yellow solid line is the actual deduced diagram).

[0140] Furthermore, the deviations of arrival and departure time between the actual graph and the planned graph, and between the actual graph and the planned graph were statistically deduced. The results are as follows: Figure 6 As shown in the figure, it is found that there is an obvious difference between the actual diagram and the planned diagram. The actual diagram is usually tens of seconds later than the planned diagram in terms of train arrival and departure times. The deviation distribution between the deduced actual diagram and the planned diagram is highly consistent with the deviation distribution of the actual diagram, indicating that the method of the present invention can accurately deduce the actual operation diagram of the train.

[0141] Specifically, the method of the present invention can not only perform offline deduction of the actual operation diagram before the implementation of the planned operation diagram, but also perform real-time deduction of the future actual diagram based on part of the actual operation data that has been known during the implementation of the planned operation diagram; still taking this line as an example, the basic data of the line and the best model obtained by training are consistent with those in the above embodiment; the time period from 17:00 to 18:00 on January 14, 2024 can be selected as the target range for real-time deduction. First, input the planned operation diagram of the day and the actual operation data of the entire line before 17:00, and perform real-time deduction of subsequent train operation on this basis, such as Figure 7 shown.

[0142] from Figure 7 It can be seen that during the period from 17:00 to 18:00, the actual operation diagram of the real-time deduction is seamlessly connected with the actual operation diagram before 17:00; the method of the present invention is calculated on a 64-bit Windows 10 computer with an Intel i7-1165G7 CPU and 40GB memory, and the real-time deduction takes only 0.2 seconds, which proves the high efficiency of the method of the present invention in real-time deduction.

[0143] Embodiment 2: The embodiment of the present application provides a data-driven subway train operation performance deduction system, which is applied to a data-driven subway train operation performance deduction method in any one of Embodiment 1, such as Figure 3 As shown, including:

[0144] The first module is used to obtain the train operation performance data of multiple days for each date type, and generate the historical operation data of each day for the corresponding date type. The historical operation data includes the operation route of each train, and the planned arrival time of each train at each station on the corresponding route, the planned departure time from each station, the actual arrival time at each station and the actual departure time from each station.

[0145] The second module is used to extract characteristic data of the running time of each train during section operation, characteristic data of the turning time of each train when it is on the turning line, and characteristic data of the stop time of each train when it stops at each station from the historical operation data. Each characteristic data includes a characteristic value and a label value.

[0146] The third module is used to perform outlier elimination processing on each feature data, and train the corresponding preset initial models based on each feature data after the outlier elimination processing, until each preset initial model reaches the training end condition, and each preset initial model that reaches the training end condition is respectively determined as the running time prediction model, the train stop time prediction model, and the train return time prediction model.

[0147] The fourth module is used to obtain the planned operation data of each train from the actual planned operation diagram, and according to the date type of the actual planned operation diagram, input each type of data in the planned operation data into the operation time prediction model, train stop time prediction model, and train return time prediction model of the corresponding date type in time sequence for processing, and obtain the predicted operation time, stop prediction time, and return prediction time respectively.

[0148] The fifth module is used to replace the corresponding original data in the planned operation data with the predicted operating time, stop prediction time and return prediction time corresponding to each time step, obtain new planned operation data, and obtain the target calculation operation diagram based on the new planned operation data.

[0149] Furthermore, the fourth module further includes:

[0150] The first submodule is used to determine the earliest time when a train goes online in the historical operation data as the target time when a train goes online in the planned operation data, and to determine the latest time when a train goes offline in the historical operation data as the target time when a train goes offline after extending the preset time period.

[0151] Embodiment 3: This embodiment of the present application provides an electronic device, such as Figure 4 As shown, it includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, any method in Embodiment 1 is implemented.

[0152] Embodiment 4: The embodiment of the present application provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions enable a computer to execute any method in Embodiment 1.

[0153] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A data-driven method for predicting subway train operation performance, characterized in that: The specific steps include: Acquire the train operation performance data of multiple days for each date type, and generate the historical operation data of each day for the corresponding date type, wherein the historical operation data includes the operation path of each train, and the planned arrival time of each train at each station on the corresponding path, the planned departure time from each station, the actual arrival time at each station, and the actual departure time from each station; Extract from the historical operation data the characteristic data of the running time of each train during the interval operation, the characteristic data of the turning time of each train when located on the turning line, and the characteristic data of the stop time of each train when stopping at each station, each of the characteristic data respectively includes a characteristic value and a label value; Performing outlier elimination processing on each of the feature data, and training the corresponding preset initial models based on each of the feature data after outlier elimination processing, until each preset initial model reaches a training end condition, and determining each preset initial model that reaches the training end condition as a running time prediction model, a train stop time prediction model, and a train turnaround time prediction model; Obtaining the planned operation data of each train from the actual planned operation diagram, and inputting each type of data in the planned operation data into the operation time prediction model, train stop time prediction model, and train turnaround time prediction model of the corresponding date type in time sequence for processing according to the date type of the actual planned operation diagram, and obtaining the predicted operation time, stop time prediction time, and turnaround prediction time respectively; The corresponding original data in the planned operation data are replaced by the predicted operation duration, stop prediction time and return prediction time corresponding to each time step to obtain new planned operation data, and a target calculation operation diagram is obtained according to the new planned operation data.

2. According to claim 1, a data-driven subway train operation performance simulation method is characterized in that: The characteristic value is a planned value in historical operation data, and the label value is an actual value corresponding to the characteristic value.

3. The data-driven subway train operation performance simulation method according to claim 1 is characterized in that: The running time prediction model is specifically: in: or Where, T α represents the time period number, j represents the interval number, d represents the running direction number, d=1 represents the downlink direction, d=2 represents the uplink direction, ΔR e (j,n) represents the time deviation of train number n entering interval j, R p (j,n) represents the planned running time of train number n in section j, ΔR l (j,n) represents the time deviation of train number n leaving interval j, R a (j,n) represents the actual running time of train number n in section j. represents the predicted running time of train number n in interval j, M represents the total number of calculations, and α m Indicates the weight of each calculation result, h m (f j θ m ) represents the prediction result of the mth calculation, f j Represents the feature vector of all feature data in interval j, θ m represents f j A collection of parameters.

4. The data-driven subway train operation performance simulation method according to claim 1 is characterized in that: The train stop time prediction model is specifically: in: or Where, T α represents the time period number, i represents the station number, d represents the running direction number, d=1 represents the down direction, d=2 represents the up direction, S p (i,n) represents the planned stop time of train number n at station i, ΔS a (i,n) represents the stop time deviation of train number n at station i, S a (i,n) represents the actual stop time of train number n at station i. represents the predicted stop time of train No. n at station i, M represents the total number of calculations, α m Indicates the weight of each calculation result, h m (f i θ m ) represents the prediction result of the mth calculation, f i The feature vector representing all feature data of station i, θ m represents f i A collection of parameters.

5. The data-driven subway train operation performance simulation method according to claim 1 is characterized in that: The train turnaround time prediction model is specifically: in: Where, T α represents the time period number, i represents the station number, d represents the running direction number, d=1 represents the down direction, d=2 represents the up direction, B p (i,n) represents the planned return time of train number n at station i, B a (i,n) represents the actual return time of train number n at station i, ΔB e (i,n) represents the return time deviation of train number n at station i, ΔB f (i,n) represents the time deviation of train No. n completing the turnaround at station i. represents the predicted return time of train No. n at station i, M represents the total number of calculations, α m Indicates the weight of each calculation result, h m (g i θ m ) represents the prediction result of the mth calculation, g i The feature vector representing all feature data of station i, θ m Indicates g i A collection of parameters.

6. The data-driven subway train operation performance simulation method according to claim 1 is characterized in that: The earliest time when a train comes online in the historical operation data is determined as the target time when a train comes online in the planned operation data, and the latest time when a train goes offline in the historical operation data is determined as the target time when a train goes offline in the planned operation data after being extended by a preset time period.

7. A data-driven subway train operation performance deduction system, applied to a data-driven subway train operation performance deduction method according to any one of claims 1 to 6, characterized in that: include: The first module is used to obtain the train operation performance data of multiple days for each date type, and generate the historical operation data of each day of the corresponding date type, wherein the historical operation data includes the operation path of each train, and the planned arrival time of each train at each station on the corresponding path, the planned departure time from each station, the actual arrival time at each station, and the actual departure time from each station; The second module is used to extract characteristic data of the running time of each train during interval operation, characteristic data of the turning time of each train when located on the turning line, and characteristic data of the stop time of each train when stopping at each station from the historical operation data, each of which includes a characteristic value and a label value; The third module is used to perform outlier elimination processing on each of the feature data, and train the corresponding preset initial models based on each of the feature data after the outlier elimination processing, until each preset initial model reaches the training end condition, and determine each preset initial model that reaches the training end condition as a running time prediction model, a train stop time prediction model, and a train return time prediction model; The fourth module is used to obtain the planned operation data of each train from the actual planned operation diagram, and according to the date type of the actual planned operation diagram, input each type of data in the planned operation data into the operation time prediction model, train stop time prediction model, and train return time prediction model of the corresponding date type in time sequence for processing, and obtain the predicted operation time, stop time prediction time, and return time prediction time respectively; The fifth module is used to replace the corresponding original data in the planned operation data with the predicted operation time, stop prediction time and return prediction time corresponding to each time step, obtain new planned operation data, and obtain the target calculation operation diagram based on the new planned operation data.

8. The data-driven subway train operation performance simulation system according to claim 7 is characterized in that: The fourth module further includes: The first submodule is used to determine the earliest time when a train comes online in the historical operation data as the target time when a train comes online in the planned operation data, and to determine the latest time when a train goes offline in the historical operation data as the target time when a train goes offline in the planned operation data after extending the time by a preset period of time.

9. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the method according to any one of claims 1 to 6 is implemented when the processor executes the computer program.

10. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores computer instructions, which enable a computer to execute the method of any one of claims 1-6.

Images