A subway section passenger flow prediction method based on LSTM network

Abstract

The present invention proposes a subway section passenger flow prediction method based on LSTM network, including: determining the section to be predicted of the designated subway line, obtaining the historical passenger flow data of the section to be predicted as of the operation day d, inputting the historical passenger flow data into the section passenger flow basic model based on LSTM network, and outputting the first prediction variable; real-time collection of real-time transaction data at the operation day d+1, analyzing the passenger's exit station and riding path through the real-time transaction data, and calculating the second prediction variable based on the analysis result and the preset update frequency; inputting the first prediction variable and the second prediction variable into the fully connected neural network together, and periodically predicting the passenger flow of the section to be predicted at the operation day d+1. The present invention uses the historical section passenger flow data to make a basic prediction of the future section passenger flow, and uses the real-time transaction data of the day to correct the result of the basic prediction to cope with the sudden short-term passenger flow, thereby improving the accuracy of the section passenger flow prediction.

Description

Technical Field

[0001] The present invention belongs to the field of urban rail transit, and in particular relates to a subway section passenger flow prediction method based on an LSTM network. Background Art

[0002] With the construction of urban rail transit, the scale and passenger flow of subway lines are growing, but the distribution of passenger flow in time and space is not uniform, and congestion occurs from time to time, usually in individual sections during peak hours. Subway operators usually hope to predict the occurrence of congestion in advance and adjust train scheduling accordingly. Existing subway passenger flow prediction technology usually targets inbound and outbound passenger flow, and is based on general technologies such as statistics, machine learning, and deep learning to predict passenger flow based on passenger flow on historical operating days. However, this prediction method can only predict inbound and outbound passenger flow. Although inbound and outbound passenger flow can assist station management, cross-section passenger flow is a more intuitive reference indicator for train scheduling. In addition, the passenger flow on historical operating days only represents the long-term trend of passenger flow. Predictions based only on passenger flow on historical operating days cannot cope with sudden high passenger flow caused by emergencies.

[0003] Therefore, the existing subway passenger flow forecasting technology has obvious room for improvement in terms of prediction range, accuracy, and real-time performance. Summary of the invention

[0004] In order to solve the shortcomings and deficiencies in the prior art, the present invention proposes a subway section passenger flow prediction method based on LSTM network, comprising:

[0005] Determine the section to be predicted of the designated subway line, obtain the historical passenger flow data of the section to be predicted as of the operation day d, input the historical passenger flow data into the section passenger flow basic model based on the LSTM network, and output the first prediction variable;

[0006] Real-time transaction data at d+1 on the operation day is collected in real time, the exit station and the boarding path of the passengers are analyzed through the real-time transaction data, and the second prediction variable is calculated based on the preset update frequency according to the analysis result;

[0007] The first prediction variable and the second prediction variable are input into the fully connected neural network to periodically predict the passenger flow of the section to be predicted on the operation day d+1.

[0008] Optionally, determining the section to be predicted of the designated subway line includes:

[0009] According to the historical departure density and historical vehicle rated passenger capacity of the designated subway line, the rated passenger capacity of each section (i, j) of the designated line is dynamically calculated, and the section whose historical passenger flow exceeds the rated passenger capacity is selected as the section to be predicted;

[0010] Among them, i represents the time period number, and the time period corresponding to i is [T i ,T i +Δt), T i is the starting time of the time period, Δt is the preset interval parameter, and the value range of i is an integer between [1, 24*60 / Δt];

[0011] j represents the section number, and the value range of j is a positive integer not exceeding the total number of sections.

[0012] Optionally, the subway section passenger flow prediction method further includes a training process of a section passenger flow basic model, and the training process includes:

[0013] Construct a basic model of cross-section passenger flow based on the architecture of LSTM network;

[0014] Obtain the historical passenger flow data of the section to be predicted from the operation day dm to the operation day d, and generate the section passenger flow value time series and date label time series according to the historical passenger flow data;

[0015] The numerical time series of the cross-sectional passenger flow and the date label time series are used together as training data, and the network parameters of the cross-sectional passenger flow basic model are trained according to the training data.

[0016] Optionally, the date label time series includes a regular date label time series and a special date label time series;

[0017] The conventional date label time series is used to mark whether the operating day is a working day, and the special date label time series is used to mark whether the operating day is a holiday.

[0018] Optionally, analyzing the passenger's exit station and riding path through real-time transaction data, and calculating the second prediction variable based on the analysis result and a preset update frequency, includes:

[0019] Extracting inbound transaction data from the real-time transaction data, the inbound transaction data being real-time transaction data of transactions that entered the station before the time period corresponding to the section to be predicted and did not leave the station during the time period;

[0020] Extract the station and card number from the entry transaction data, and determine whether the passenger generating the entry transaction data is a frequent passenger based on the card number;

[0021] If the passenger is a frequent passenger, obtain the probability distribution of the frequent passenger's exit station, combine the probability distribution of the travel path between the entry station and the exit station, and calculate the first probability of the frequent passenger passing the section to be predicted;

[0022] If the passenger is not a frequent passenger, obtain the probability distribution of the exit stations of all passengers during the period from operation day dm to operation day d, and calculate the second probability of other passengers passing the section to be predicted, in combination with the travel path between the entry station and the exit station;

[0023] All first probabilities and second probabilities are accumulated to obtain the second predictor variable.

[0024] Optionally, judging whether the passenger generating the station entry transaction data is a frequent passenger according to the card swiping number includes:

[0025] If in all historical transaction data from operation day dm to operation day d, the number of entries by the card swiping number in the time interval formed by one hour before and after the entry time exceeds m*p times, then the passenger is determined to be a frequent flyer, and p is a preset parameter.

[0026] Optionally, the obtaining of the probability distribution of the exit station of the frequent passenger, combining the probability distribution of the riding path between the entry station and the exit station, and calculating the first probability of the frequent passenger passing the section to be predicted includes:

[0027] Obtain the corresponding exit transaction data generated after the card number enters the station within the time interval formed by one hour before and after the current entry time during the period from operation day dm to operation day d, count the number of exit stations in the exit transaction data, and calculate the probability distribution P1 of the exit station of frequent passengers based on the statistical results ExitSt ;

[0028] Find all the ride paths between the entry station and the exit station and the selection probability P of each ride path through the passenger flow classification database ep ;

[0029] Combined with the train schedule, determine whether each travel route passes through the predicted section. If so, calculate the first probability as P1 ExitSt ×P ep .

[0030] Optionally, the step of obtaining the probability distribution of the exit stations of all passengers during the period from the operation day dm to the operation day d, and calculating the second probability of passengers other than regular passengers passing through the section to be predicted in combination with the riding path between the entry station and the exit station, comprises:

[0031] Obtain the corresponding exit transaction data generated after all passengers enter the station within the time interval formed by one hour before and after the current entry time during the period from operation day dm to operation day d, count the number of exit stations in the exit transaction data, and calculate the probability distribution P2 of the exit stations of frequent passengers based on the statistical results ExitSt ;

[0032] Find all the ride paths between the entry station and the exit station and the selection probability P of each ride path through the passenger flow classification database ep ;

[0033] Combined with the train schedule, determine whether each travel path passes through the predicted section. If so, calculate the second probability as P2 ExitSt ×P ep .

[0034] Optionally, the fully connected neural network includes a hidden layer, the activation function used by the fully connected neural network is Sigmoid, and the loss function used by the fully connected neural network is a mean square error function.

[0035] The beneficial effects brought by the technical solution provided by the present invention are:

[0036] The present invention uses historical cross-sectional passenger flow data to make basic predictions for future cross-sectional passenger flow, and uses real-time transaction data of the day to correct the results of the basic predictions to cope with sudden short-term passenger flow, thereby improving the accuracy of cross-sectional passenger flow predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solution of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0038] Figure 1 A schematic diagram of a flow chart of a subway section passenger flow prediction method based on an LSTM network proposed in an embodiment of the present invention;

[0039] Figure 2 This is a schematic diagram of the stage flow of the subway section passenger flow prediction method. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution 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 only part of the embodiments of the present invention, not all of the embodiments. 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.

[0041] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein.

[0042] It should be understood that in various embodiments of the present invention, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0043] It should be understood that in the present invention, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0044] It should be understood that in the present invention, "plurality" refers to two or more than two. "And / or" is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "Contains A, B and C", "Contains A, B, C" means that A, B, and C are all included, "Contains A, B or C" means that one of A, B, and C is included, and "Contains A, B and / or C" means that any one, any two, or any three of A, B, and C are included.

[0045] It should be understood that in the present invention, "B corresponding to A", "B corresponding to A", "A corresponds to B" or "B corresponds to A" means that B is associated with A and B can be determined based on A. Determining B based on A does not mean determining B based only on A, but B can also be determined based on A and / or other information. A and B match when the similarity between A and B is greater than or equal to a preset threshold.

[0046] Depending on the context, "if" as used herein may be interpreted as "when" or "when" or "in response to determining" or "in response to detecting."

[0047] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0048] Embodiment 1

[0049] like Figure 1 As shown, this embodiment proposes a subway section passenger flow prediction method based on LSTM network, including:

[0050] S1: Determine the section to be predicted of the designated subway line, obtain the historical passenger flow data of the section to be predicted as of the operation day d, input the historical passenger flow data into the section passenger flow basic model based on the LSTM network, and output the first prediction variable;

[0051] S2: collect real-time transaction data at d+1 on the operation day, analyze the passengers' exit stations and travel routes through the real-time transaction data, and calculate the second prediction variable based on the preset update frequency according to the analysis results;

[0052] S3: Input the first prediction variable and the second prediction variable into a fully connected neural network to periodically predict the passenger flow of the section to be predicted on the operation day d+1.

[0053] The subway section passenger flow prediction method proposed in this embodiment is divided into a section analysis stage, a model construction stage and a passenger flow prediction stage. Figure 2 shown.

[0054] In the section analysis stage, this embodiment mainly performs a summary analysis of historical passenger flow to select sections with larger historical passenger flow for analysis. Therefore, this embodiment does not need to analyze all sections, which saves prediction and calculation resources, and can accurately select sections that are more necessary to predict based on historical passenger flow conditions, which is conducive to improving the effect of section passenger flow prediction.

[0055] After the operation day d, this embodiment enters the model construction stage. On the one hand, the LSTM model is constructed and trained to obtain the cross-sectional passenger flow basic model for outputting the first prediction variable, and on the other hand, the cross-sectional passenger flow correction model is constructed to output the second prediction variable. In addition, the training data is constructed based on the training results of the cross-sectional passenger flow basic model and the cross-sectional passenger flow correction model, so as to train the fully connected neural network to fit the final cross-sectional passenger flow prediction result through the first prediction variable and the second prediction variable.

[0056] This embodiment enters the passenger flow prediction stage at d+1 on the operation day. On the one hand, it obtains historical passenger flow data, predicts through the cross-sectional passenger flow basic model, and outputs the first prediction variable. On the other hand, it obtains the real-time transaction data of the day, calculates through the cross-sectional passenger flow correction model, and outputs the second prediction variable. The first prediction variable and the second prediction variable are then input into the fully connected neural network together, and the cross-sectional passenger flow is predicted through the fully connected neural network.

[0057] The implementation process of the above three stages is described in detail below.

[0058] In this embodiment, determining the section to be predicted of the designated subway line includes:

[0059] According to the historical departure density and historical vehicle rated passenger capacity of the designated subway line, the rated passenger capacity of each section (i, j) of the designated line is dynamically calculated, and the section whose passenger flow exceeds the rated passenger capacity is selected as the section to be predicted;

[0060] Among them, i represents the time period number, and the time period corresponding to i is [T i ,T i +Δt), T i is the starting time of the time period, Δt is the preset interval parameter, i is an integer between [1, 24*60 / Δt]; j is the section number, and j is a positive integer not exceeding the total number of sections. In this embodiment, Δt is 15 minutes, which is small enough to meet the train scheduling requirements of the line and large enough to ensure the stability of the section passenger flow time series.

[0061] After the section to be predicted is determined, after the end of the operation day d, the historical passenger flow data of the section to be predicted up to the operation day d is obtained, and the historical passenger flow data is input into the section passenger flow basic model based on the LSTM network , output the first predictor variable It indicates the cross-section passenger flow at operation day d+1 predicted based on the historical passenger flow data before operation day d+1. LSTM network is a special RNN network, which is mainly used to solve the gradient disappearance and gradient explosion problems in the long sequence training process. It is suitable for subway cross-section passenger flow prediction scenarios with time series characteristics.

[0062] This embodiment also includes a training process of the cross-sectional passenger flow basic model, and the training process includes:

[0063] Construct a basic model of cross-section passenger flow based on the architecture of LSTM network;

[0064] Obtain the historical passenger flow data of the section to be predicted from the operating day dm to the operating day d, that is, the historical passenger flow data of the section to be predicted (i, j) within the operating day range of (dm, d]. In this embodiment, the method for determining m is: select the date span from the large-scale expansion of the line network to the current date, and it should be not less than 90. Generate a section passenger flow numerical time series and a date label time series based on the historical passenger flow data. The date label time series includes a regular date label time series and a special date label time series. The regular date label time series is used to mark whether the operating day is a working day, and the special date label time series is used to mark whether the operating day is a holiday. Preferably, this embodiment constructs three time series with a length of m, which are the section passenger flow numerical time series, Regular date label time series and special date label time series The labels include working days, rest days with a length of 1-2 days, rest days with a length of 3-4 days, and rest days with a length of more than 5 days. The labels include New Year's Eve, the day before New Year's Day, Christmas Eve, Women's Day, and surrounding major event days. It is used to introduce the influencing factors of different dates on passenger flow in the LSTM network prediction process, which is conducive to improving the prediction accuracy of the cross-sectional passenger flow basic model.

[0065] The numerical time series of the cross-sectional passenger flow and the time series of the date label are used as training data, and the network parameters of the cross-sectional passenger flow basic model are trained based on the training data. The initial value of the LSTM hidden state dimension is set to 64, but it is manually adjusted during the training process.

[0066] This embodiment also constructs a cross-section passenger flow correction model for the section to be predicted (i, j) after the operation day d ends. Used to analyze passengers' exit stations and travel routes through real-time transaction data, and calculate the second prediction variable based on the preset update frequency according to the analysis results, and the cross-sectional passenger flow correction model The calculation process includes:

[0067] Extract the entry transaction data from the real-time transaction data, the entry data is the real-time transaction data of the passengers who entered the station before the time period corresponding to the section to be predicted and did not leave the station during the time period. Extract the entry station EnterSt and the card number CardId from the entry transaction data, and determine whether the passenger who generated the entry transaction data is a frequent passenger according to the card number CardId.

[0068] Preferably, the judgment method of a frequent passenger is as follows: if in all historical transaction data from operation day dm to operation day d, the number of times the card number enters the station in the time interval formed by one hour before and after the entry time exceeds m*p times, then the passenger is judged to be a frequent passenger, p is a preset parameter, and the value of p in this embodiment is 0.6. For example, if the entry time of CardId on operation day d+1 is 9 o'clock, then the number of occurrences of CardId between 8 o'clock and 10 o'clock every day in the m days before the operation day is counted, and if it exceeds m*p times, then the passenger is judged to be a frequent passenger.

[0069] It can be seen that frequent passengers are people who often enter the station to take the subway at a fixed time period, and their subway riding routes are usually relatively fixed. Therefore, this embodiment analyzes the probability of frequent passengers passing the section to be predicted separately, that is, obtains the probability distribution of the exit station of frequent passengers, and combines the probability distribution of the riding route between the entry station and the exit station to calculate the first probability of frequent passengers passing the section to be predicted, including:

[0070] Obtain the corresponding exit transaction data generated after the card number enters the station within the time interval formed by one hour before and after the current entry time during the period from operation day dm to operation day d, count the number of exit stations in the exit transaction data, and calculate the probability distribution P1 of the exit station of frequent passengers based on the statistical results ExitSt In this embodiment, P1 ExitSt That is, the ratio of the number of appearances at each station to the total number of exits.

[0071] Find all the ride paths between the entry station and the exit station and the selection probability P of each ride path through the passenger flow classification database ep In this embodiment, the riding path and the corresponding selection probability are determined by a conventional subway sorting method, which will not be described in detail here.

[0072] Combined with the train schedule, determine whether each travel route passes through the section to be predicted. Specifically, for any EP, if the EP contains section j, find the first train that the passenger can take in the train schedule according to the entry time, and simulate the passenger's boarding, transfer, and exit behaviors. If the passenger happens to pass section j in time period i, the first probability is calculated as P1 ExitS ×P ep .

[0073] This embodiment analyzes the probability of other passengers, except frequent passengers, passing the section to be predicted separately, that is, obtains the probability distribution of the exit stations of all passengers during the period from operation day dm to operation day d, and calculates the second probability of other passengers, except frequent passengers, passing the section to be predicted in combination with the riding path between the entry station and the exit station, including:

[0074] Obtain the corresponding exit transaction data generated after all passengers enter the station within the time interval formed by one hour before and after the current entry time during the period from operation day dm to operation day d, count the number of exit stations in the exit transaction data, and calculate the probability distribution P2 of the exit stations of frequent passengers based on the statistical results ExitSt In this embodiment, P1 ExitSt That is, the ratio of the number of appearances at each station to the total number of exits.

[0075] Find all the ride paths between the entry station and the exit station and the selection probability P of each ride path through the passenger flow classification database ep In this embodiment, the riding path and the corresponding selection probability are determined by a conventional subway sorting method, which will not be described in detail here.

[0076] Combined with the train schedule, determine whether each travel path passes through the section to be predicted. Specifically, for any EP, if the EP contains section j, find the first train that the passenger can take in the train schedule according to the entry time, and simulate the passenger's boarding, transfer, and exit behaviors. If the passenger happens to pass section j in time period i, the second probability is calculated as P2 ExitSt ×P ep .

[0077] Finally, all the first and second probabilities are accumulated, that is, the probabilities of all passengers who generate real-time transaction data passing through the predicted section (i, j) are added together to obtain the second prediction variable. The second prediction variable represents the predicted section passenger flow C on the operation day d+1 based on the real-time station entry situation. i d j +1 .

[0078] This embodiment classifies whether a passenger is a frequent passenger based on the frequency of his / her entry within a certain period of time, and analyzes the probability of frequent passengers passing through a section separately. Compared with a method of performing a unified probability analysis on all passengers, this method is more targeted at frequent passengers, thereby achieving the effect of improving the accuracy of prediction results.

[0079] Finally, the first predictor and the second predictor are fed into the fully connected neural model This embodiment also proposes a fully connected neural model The training process includes:

[0080] Select the date in the range of (d-[m / 2], d] for the operation date, and construct the sample set X, where the element type in X is a vector Where k is the date serial number, the range is (d-[m / 2], d], To use the model The predicted passenger flow value, To use the model The calculated value, is the historical statistical value of the section passenger flow. During the training process, according to and Analyze the passenger flow prediction value of section (i, j) and analyze the passenger flow prediction value based on the loss function. The error is adjusted according to the error Parameters in .

[0081] In this embodiment, The number of neurons in the hidden layer was initially set to 5 and was manually adjusted during training.

[0082] In the passenger flow prediction stage, we first use the cross-sectional passenger flow basic model according to The first predictor variable for forecasting operation day d+1 On the operation day d+1, the inbound and outbound passenger flow of the entire line network on d+1 is collected in real time, and the model is corrected using the cross-sectional passenger flow Periodically calculate and update the second predictor variable The preset update frequency in this embodiment is set to 5 minutes.

[0083] Preferably, the entry and exit records generated by the subway gate are transmitted to the line network control center in real time via TCP and saved in the middleware Kafka to facilitate the cross-section passenger flow correction model. Call.

[0084] After updating, use Periodically calculate and predict the cross-sectional passenger flow in time period i and section j, and publish the predicted cross-sectional passenger flow to the outside world via HTTP.

[0085] The serial numbers in the above embodiments are for description only and do not represent the sequence of the components during assembly or use.

[0086] The above descriptions are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A subway section passenger flow prediction method based on LSTM network, characterized in that: include: Determine the section to be predicted of the designated subway line, obtain the historical passenger flow data of the section to be predicted as of the operation day d, input the historical passenger flow data into the section passenger flow basic model based on the LSTM network, and output the first prediction variable; Real-time transaction data at d+1 on the operation day is collected in real time, the exit station and the boarding path of the passengers are analyzed through the real-time transaction data, and the second prediction variable is calculated based on the preset update frequency according to the analysis result; The first prediction variable and the second prediction variable are input into a fully connected neural network to periodically predict the passenger flow of the section to be predicted on the operation day d+1; Determining the section to be predicted of the designated subway line includes: According to the historical departure density and historical vehicle rated passenger capacity of the designated subway line, the rated passenger capacity of each section (i, j) of the designated line is dynamically calculated, and the section whose historical passenger flow exceeds the rated passenger capacity is selected as the section to be predicted; Among them, i represents the time period number, and the time period corresponding to i is [T i ,T i +Δt), T i is the starting time of the time period, Δt is the preset interval parameter, and the value range of i is an integer between [1, 24*60 / Δt]; j represents the section number, and the value range of j is a positive integer not exceeding the total number of sections; The subway section passenger flow prediction method also includes a training process of a section passenger flow basic model, and the training process includes: Construct a basic model of cross-section passenger flow based on the architecture of LSTM network; Obtain the historical passenger flow data of the section to be predicted from the operation day dm to the operation day d, and generate the section passenger flow numerical time series and date label time series according to the historical passenger flow data; The numerical time series of the cross-sectional passenger flow and the date label time series are used as training data, and the network parameters of the cross-sectional passenger flow basic model are trained according to the training data; The step of analyzing the passenger's exit station and riding path through real-time transaction data, and calculating the second prediction variable based on the analysis result and a preset update frequency includes: Extracting inbound transaction data from the real-time transaction data, the inbound transaction data being real-time transaction data of transactions that entered the station before the time period corresponding to the section to be predicted and did not leave the station during the time period; Extract the station and card number from the entry transaction data, and determine whether the passenger generating the entry transaction data is a frequent passenger based on the card number; If the passenger is a frequent passenger, obtain the probability distribution of the frequent passenger's exit station, combine the probability distribution of the riding path between the entry station and the exit station, and calculate the first probability of the frequent passenger passing the section to be predicted; If the passenger is not a frequent passenger, obtain the probability distribution of the exit stations of all passengers during the period from operation day dm to operation day d, and calculate the second probability of other passengers passing the section to be predicted, in combination with the travel path between the entry station and the exit station; All first and second probabilities are accumulated to obtain the second predictor variable; The method of obtaining the probability distribution of the exit station of the frequent passengers and calculating the first probability of the frequent passengers passing through the section to be predicted by combining the probability distribution of the riding path between the entry station and the exit station includes: Obtain the corresponding exit transaction data generated after the card number enters the station within the time interval formed by one hour before and after the current entry time during the period from operation day dm to operation day d, count the number of exit stations in the exit transaction data, and calculate the probability distribution P1 of the exit station of frequent passengers based on the statistical results ExitSt ; Find all the ride paths between the entry station and the exit station and the selection probability P of each ride path through the passenger flow classification database ep ; Combined with the train schedule, determine whether each travel route passes through the predicted section. If so, calculate the first probability as P1 ExitSt ×P ep ; The method of obtaining the probability distribution of the exit stations of all passengers during the period from the operation day dm to the operation day d, and calculating the second probability of other passengers except the regular passengers passing the section to be predicted in combination with the riding path between the entry station and the exit station, comprises: Obtain the corresponding exit transaction data generated after all passengers enter the station within the time interval formed by one hour before and after the current entry time during the period from operation day dm to operation day d, count the number of exit stations in the exit transaction data, and calculate the probability distribution P2 of the exit stations of frequent passengers based on the statistical results ExitSt ; Find all the ride paths between the entry station and the exit station and the selection probability P of each ride path through the passenger flow classification database ep ; Combined with the train schedule, determine whether each travel route passes through the predicted section. If so, calculate the second probability as P2 ExitSt ×P ep .

2. According to a method for predicting subway section passenger flow based on LSTM network according to claim 1, it is characterized in that: The date label time series includes a regular date label time series and a special date label time series; The conventional date label time series is used to mark whether the operating day is a working day, and the special date label time series is used to mark whether the operating day is a holiday.

3. The method for predicting subway section passenger flow based on LSTM network according to claim 1, characterized in that: The step of judging whether the passenger generating the station entry transaction data is a frequent passenger according to the card swiping number includes: If in all historical transaction data from operation day dm to operation day d, the number of entries by the card swiping number in the time interval formed by one hour before and after the entry time exceeds m*p times, then the passenger is determined to be a frequent flyer, and p is a preset parameter.

4. The method for predicting subway section passenger flow based on LSTM network according to claim 1, characterized in that: The fully connected neural network includes a hidden layer, the activation function used by the fully connected neural network is Sigmoid, and the loss function used by the fully connected neural network is a mean square error function.

Images