Urban rail transit passenger flow prediction method and system
By using ResCNN and nested LSTM neural networks to combine external features, the problem of ignoring the intrinsic relationship between station entry and exit volumes in existing technologies is solved, thereby improving the accuracy of urban rail transit passenger flow prediction.
Patent Information
- Application Number
- CN202510936737.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-08
- Publication Date
- 2025-11-18
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing methods for predicting travel demand at urban rail transit stations neglect the inherent relationship between arrival and departure volumes, lack modeling of the spatiotemporal characteristics and dynamic correlations of traffic data, and make insufficient use of external variables, resulting in inaccurate predictions.
A customized ResCNN module and a nested LSTM neural network are used to extract the spatiotemporal and temporal features of the stations. A multi-task learning method is used to characterize the correlation between passenger flow entering and exiting the station. The prediction is made by combining the rail transit timetable, the distribution of surrounding bus stations and regional land use characteristics.
The accuracy of urban rail transit passenger flow forecasting has been improved. By considering spatiotemporal characteristics and multi-task learning methods, the stability and accuracy of the forecasting model have been enhanced.
Smart Images
Figure CN120975283A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of passenger flow forecasting, and in particular to a method and system for forecasting passenger flow in urban rail transit. Background Technology
[0002] Travel demand forecasting for urban rail transit stations includes forecasting both arrival and departure volumes. Existing methods predict short-term arrival and departure volumes separately based on proposed models, treating travel demand forecasting as a single task and ignoring the inherent relationship between arrival and departure volume forecasts. Furthermore, they lack modeling of the spatiotemporal characteristics and dynamic correlations of traffic data, as well as the effective utilization of external variables. Summary of the Invention
[0003] The main objective of this invention is to overcome the aforementioned deficiencies in the prior art and propose a method for predicting passenger flow in urban rail transit. This method employs a customized neural network extraction module, considers the extraction of spatiotemporal travel characteristics, and uses a multi-task learning method to characterize the correlation between inbound and outbound passenger flow, thereby improving the accuracy of prediction.
[0004] The technical solution adopted in this invention,
[0005] A method for predicting urban rail transit passenger flow, including,
[0006] Get the inbound and outbound passenger flow of the station before the current time interval t;
[0007] The ResCNN module was used to extract the spatial features of passenger flow entering and exiting the station, thus obtaining the spatial features of passenger travel demand between stations.
[0008] Nested LSTM neural networks are used to extract the temporal features of passenger flow entering and exiting stations, thereby obtaining the temporal features of passenger travel demand between stations.
[0009] External features to be acquired include rail transit timetable features, surrounding bus station distribution features, and regional land use features;
[0010] A feature fusion layer is used to connect the extracted spatial features of passenger travel demand between stations, temporal features of passenger travel demand between stations, rail transit timetable features, surrounding bus station distribution features, and regional land use features. An attention-based nested LSTM neural network is used to decode the fused feature variables to predict the number of passengers entering and leaving each station in the rail transit grid.
[0011] Specifically, the ResCNN module is used to extract the spatial features of passenger flow entering and exiting the stations, thereby obtaining the spatial features of passenger travel demand between stations, as follows:
[0012] The ResCNN module:
[0013] X l+1 =R(X) l+1 θ l )+X l
[0014] Among them, X l+1 and X l Let θ represent the input and output of the residual unit, respectively, and R be the residual function; l It includes all the learnable parameters of the residual unit;
[0015] A batch normalization layer is used to standardize the input data, and a convolutional layer is added to the residual unit to extract the spatial features of passenger travel demand between stations.
[0016] Specifically, a nested LSTM neural network is used to extract the temporal features of passenger flow entering and exiting the stations, thereby obtaining the temporal features of passenger travel demand between stations, as follows:
[0017] In attention-based nested LSTM neural networks, the trained attention weights α are used for weighting, and the calculation method is as follows:
[0018] α=y(W A ⊙γ N +c A )
[0019]
[0020] Where y(.) is the learning function, which assigns different attention weights to different historical time steps of the input, W A and c A These are the trainable weights and biases, γ. N This is the output of a nested LSTM neural network, where ⊙ represents the matrix element-wise product. This is the weighted output of the nested LSTM neural network.
[0021] Specifically, the external features obtained include rail transit timetable features, surrounding bus station distribution features, and regional land use features, among which,
[0022] The characteristic of rail transit timetable is N s,t ;
[0023]
[0024] N s,t This represents the number of trains arriving at station s during time period t, where This indicates the number of trains arriving at station s in the upstream direction during time period t. This represents the number of trains arriving at station s in the down direction during time period t;
[0025] Characteristics of the distribution of surrounding bus stops (B) s =[b s,1 ,b s,2 ,b s,3 ]
[0026] Among them, b s,1 ,b s,2 ,b s,3 This indicates the number of bus stops within 500, 1000, and 1500 meters of a rail transit station s.
[0027] Regional land use characteristics P s =[p s,tra ,p s,din ,p s,biz ,p s,int ,p s,res ,p s,gov ]
[0028] Among them, P s This represents the number of point of interest categories around site s, where transportation facilities / car services are represented by p. s,tra It indicates that the catering service category uses p s,din This indicates that commercial categories use p. s,biz This indicates that scenic spots and historical sites are represented by p. s,int This indicates that residential services are represented by p. s,res It indicates that government and financial institutions use p. s,gov express.
[0029] Specifically, during the training process, the Adam optimizer is used to optimize the proposed MLT-STNN model, and the loss function of the model is the mean absolute error.
[0030] Specifically, the prediction model for urban rail transit passenger flow forecasting is as follows:
[0031]
[0032] Where P represents the historical time step used for prediction, and T represents the prediction time step. Let m represent the set of passenger travel volumes at subway stations within time interval t, and let E represent the prediction task. t As an external feature, the station's passenger flow in and out of the station before the current time interval t. Passenger flow in and out of the station at the next or several future time intervals.
[0033] Another embodiment of the present invention provides an urban rail transit passenger flow prediction system, including,
[0034] Historical passenger flow acquisition unit: Acquires the inbound and outbound passenger flow of the station up to the current time interval t;
[0035] Spatial feature extraction unit: The ResCNN module is used to extract the spatial features of passenger flow entering and exiting the station, so as to obtain the spatial features of passenger travel demand between stations;
[0036] Temporal feature extraction unit: A nested LSTM neural network is used to extract the temporal features of passenger flow entering and exiting the station, so as to obtain the temporal features of passenger travel demand between stations;
[0037] External feature acquisition unit: Acquires external features including rail transit timetable features, surrounding bus station distribution features, and regional land use features;
[0038] Prediction Unit: The feature fusion layer connects the extracted spatial features of passenger travel demand between stations, temporal features of passenger travel demand between stations, rail transit timetable features, surrounding bus station distribution features, and regional land use features. An attention-based NLSTM neural network is used to decode the fused feature variables to predict the number of passengers entering and leaving each station in the rail transit grid.
[0039] In another aspect, the present invention 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 executes the computer program to implement the steps of the above-described urban rail transit passenger flow prediction method.
[0040] In another aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described urban rail transit passenger flow prediction method.
[0041] As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following beneficial effects:
[0042] This invention provides a method for predicting passenger flow in urban rail transit, specifically a method for predicting passenger flow in urban rail transit, comprising: obtaining the passenger flow entering and exiting stations before the current time interval t; using a ResCNN module to extract the spatial features of the passenger flow entering and exiting stations to obtain the spatial features of passenger travel demand between stations; using a nested LSTM neural network to extract the temporal features of the passenger flow entering and exiting stations to obtain the temporal features of passenger travel demand between stations; obtaining external features including rail transit timetable features, surrounding bus station distribution features, and regional land use features; and employing feature fusion. The fusion layer connects the extracted spatial features of passenger travel demand between stations, the temporal features of passenger travel demand between stations, the features of rail transit timetables, the features of surrounding bus station distribution, and the land use features of the region. An attention-based nested LSTM neural network is used to decode the fused feature variables to predict the inbound and outbound volumes of each station in the rail transit grid. This invention proposes an urban rail transit passenger flow prediction method that employs a customized neural network extraction module, considers the extraction of spatiotemporal travel features, and uses a multi-task learning method to characterize the correlation between inbound and outbound passenger flows, thereby improving prediction accuracy. Attached Figure Description
[0043] Figure 1 A flowchart of a method for predicting passenger flow in urban rail transit provided in an embodiment of the present invention;
[0044] Figure 2 A framework diagram of an urban rail transit passenger flow prediction system provided in this embodiment of the invention;
[0045] Figure 3 A schematic diagram of an electronic device provided in an embodiment of the present invention;
[0046] Figure 4 This is a schematic diagram of an embodiment of a computer-readable storage medium provided in this invention.
[0047] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Detailed Implementation
[0048] This invention proposes a method for predicting passenger flow in urban rail transit. The method employs a customized neural network extraction module, considers the extraction of spatiotemporal travel features, and uses a multi-task learning method to characterize the correlation between inbound and outbound passenger flow, thereby improving the accuracy of prediction.
[0049] like Figure 1 The following is a flowchart of a method for predicting passenger flow in urban rail transit. The specific method is as follows:
[0050] S101: Obtain the passenger flow entering and exiting the station before the current time interval t;
[0051] This invention employs a multi-task learning method to extract the correlation between short-term passenger flow prediction tasks for both inbound and outbound passenger flows at rail transit stations. Let represent the set of passenger flow trips at subway stations within time interval t. Therefore, the short-term passenger flow prediction problem for rail transit based on multi-task learning can be defined as, considering the passenger flow entering and exiting stations before the current time interval t, i.e. and external features E t Predicting the next or multiple future time intervals Passenger flow entering and exiting the station:
[0052]
[0053] Where P represents the historical time step used for prediction, and T represents the prediction time step.
[0054] S102: The ResCNN module is used to extract the spatial features of passenger flow entering and exiting the station to obtain the spatial features of passenger travel demand between stations;
[0055] To learn the spatiotemporal characteristics of rail transit passenger flow data, convolutional neural networks (CNNs) can be used to capture the potential spatial relationships between passenger flows at different stations. However, as the depth of CNNs increases, the network model exhibits a "degradation phenomenon." The error in the training set rebounds after decreasing to a certain level, and the extraction effect of spatial features becomes worse. This is due to the gradient explosion and gradient vanishing problems caused by the deepening of the network. To address this, a residual convolutional neural network (ResCNN) is constructed, embedding residual units into the network structure. Borrowing the idea of cross-layer connections, the specific method is to add a skip connection between the input and output of each layer, directly mapping some features from the lower layers to the higher layers, effectively alleviating the training degradation phenomenon that occurs in multi-layer neural networks.
[0056] X l+1 =R(X) l+1 θ l )+X l
[0057] Among them, X l+1 and X l Let θ represent the input and output of the residual unit, respectively, and R be the residual function; l It includes all the learnable parameters of the residual unit;
[0058] The application of residual convolutional neural networks effectively alleviates the problems of gradient vanishing and exploding in deep networks, enhances the stability during training, and thus improves the learning ability of the network.
[0059] S103: A nested LSTM neural network is used to extract the temporal features of passenger flow entering and exiting the station, so as to obtain the temporal features of passenger travel demand between stations.
[0060] Unlike structures that increase the depth of LSTM neural networks by stacking structures, nested LSTM neural networks improve the ability of neural network modeling, speed up training, and reduce the risk of overfitting by introducing multi-layer LSTM structures and shared parameters.
[0061] Specifically, a nested LSTM neural network is used to extract the temporal features of passenger flow entering and exiting the stations, thereby obtaining the temporal features of passenger travel demand between stations, as follows:
[0062] In attention-based nested LSTM neural networks, the trained attention weights α are used for weighting, and the calculation method is as follows:
[0063] α=y(W A ⊙γ N +c A )
[0064]
[0065] Where y(.) is the learning function, which assigns different attention weights to different historical time steps of the input, W A and c A These are the trainable weights and biases, γ. N This is the output of a nested LSTM neural network, where ⊙ represents the matrix element-wise product. This is the weighted output of the nested LSTM neural network.
[0066] S104: Obtain external features including rail transit timetable features, surrounding bus station distribution features, and regional land use features;
[0067] Rail transit operations have distinct characteristics. For example, compared to buses, trains operate according to precise timetables. Passengers typically arrive at the platform as the train approaches its destination, resulting in small peaks in passenger flow. Furthermore, to ensure a good match between capacity and passenger volume, train timetables are designed with passenger volume at specific times in mind; the more trains scheduled to arrive within a given time window, the greater the passenger flow will be within that window.
[0068] The characteristic of rail transit timetable is N s,t ;
[0069]
[0070] N s,t This represents the number of trains arriving at station s during time period t, where This indicates the number of trains arriving at station s in the upstream direction during time period t. This represents the number of trains arriving at station s in the down direction during time period t;
[0071] Rail transit, with its advantages of large capacity, high punctuality, and speed, meets the daily travel needs of urban residents to a certain extent. However, it has certain limitations in terms of station locations and routes, and commuters typically connect with buses, shared bicycles, and other modes of transportation. Therefore, the distribution and number of bus stops around rail transit stations have a certain impact on subway passenger flow.
[0072] Characteristics of the distribution of surrounding bus stops (B) s =[b s,1 ,b s,2 ,b s,3 ]
[0073] Among them, b s,1 ,b s,2 ,b s,3 This indicates the number of bus stops within 500, 1000, and 1500 meters of a rail transit station s.
[0074] Points of Interest (POIs) are a concept in Geographic Information Systems (GIS), referring to geographic entities closely related to people's lives, such as schools, banks, restaurants, hospitals, and supermarkets. The distribution of POIs in a region can reflect its land use characteristics to some extent. Passenger flow at rail transit stations is closely related to the facilities surrounding them. If a subway station is surrounded by many residential communities, its passenger flow will primarily consist of morning and evening commuters; if there are many restaurants nearby, the station will experience higher passenger volume during mealtimes. This study uses web crawling technology to obtain POI data from the Gaode Maps Open Platform to represent the land use around rail transit stations.
[0075] Regional land use characteristics P s =[p s,tra ,p s,din ,p s,biz ,p s,int ,p s,res ,p s,gov ]
[0076] Among them, P sThis represents the number of point of interest categories around site s, where transportation facilities / car services are represented by p. s,tra It indicates that the catering service category uses p s,din This indicates that commercial categories use p. s,biz This indicates that scenic spots and historical sites are represented by p. s,int This indicates that residential services are represented by p. s,res It indicates that government and financial institutions use p. s,gov express.
[0077] S105: A feature fusion layer is used to connect the extracted spatial features of passenger travel demand between stations, temporal features of passenger travel demand between stations, rail transit timetable features, surrounding bus station distribution features, and regional land use features. An attention-based nested LSTM neural network is used to decode the fused feature variables and predict the number of passengers entering and leaving each station in the rail transit grid.
[0078] Specifically, during the training process, the Adam optimizer is used to optimize the proposed MLT-STNN model, and the loss function of the model is the mean absolute error.
[0079] like Figure 2 This is a diagram illustrating the architecture of an urban rail transit passenger flow prediction system, specifically including:
[0080] Historical passenger flow acquisition unit 201: Acquires the passenger flow entering and exiting the station before the current time interval t;
[0081] This invention employs a multi-task learning method to extract the correlation between short-term passenger flow prediction tasks for both inbound and outbound passenger flows at rail transit stations. Let represent the set of passenger flow trips at subway stations within time interval t. Therefore, the short-term passenger flow prediction problem for rail transit based on multi-task learning can be defined as, considering the passenger flow entering and exiting stations before the current time interval t, i.e. and external features E t Predicting the next or multiple future time intervals Passenger flow entering and exiting the station:
[0082]
[0083] Where P represents the historical time step used for prediction, and T represents the prediction time step.
[0084] Spatial feature extraction unit 202: The ResCNN module is used to extract the spatial features of passenger flow entering and exiting the station to obtain the spatial features of passenger travel demand between stations;
[0085] To learn the spatiotemporal characteristics of rail transit passenger flow data, convolutional neural networks (CNNs) can be used to capture the potential spatial relationships between passenger flows at different stations. However, as the depth of CNNs increases, the network model exhibits a "degradation phenomenon." The error in the training set rebounds after decreasing to a certain level, and the extraction effect of spatial features becomes worse. This is due to the gradient explosion and gradient vanishing problems caused by the deepening of the network. To address this, a residual convolutional neural network (ResCNN) is constructed, embedding residual units into the network structure. Borrowing the idea of cross-layer connections, the specific method is to add a skip connection between the input and output of each layer, directly mapping some features from the lower layers to the higher layers, effectively alleviating the training degradation phenomenon that occurs in multi-layer neural networks.
[0086] X l+1 =R(X) l+1 θ l )+X l
[0087] Among them, X l+1 and X l Let θ represent the input and output of the residual unit, respectively, and R be the residual function; l It includes all the learnable parameters of the residual unit;
[0088] The application of residual convolutional neural networks effectively alleviates the problems of gradient vanishing and exploding in deep networks, enhances the stability during training, and thus improves the learning ability of the network.
[0089] Time feature extraction unit 203: Uses nested LSTM neural network to extract the time features of passenger flow entering and leaving the station, and obtains the time features of passenger travel demand between stations;
[0090] Unlike structures that increase the depth of LSTM neural networks by stacking structures, nested LSTM neural networks improve the ability of neural network modeling, speed up training, and reduce the risk of overfitting by introducing multi-layer LSTM structures and shared parameters.
[0091] Specifically, a nested LSTM neural network is used to extract the temporal features of passenger flow entering and exiting the stations, thereby obtaining the temporal features of passenger travel demand between stations, as follows:
[0092] In attention-based nested LSTM neural networks, the trained attention weights α are used for weighting, and the calculation method is as follows:
[0093] α=y(W A ⊙γ N +cA )
[0094]
[0095] Where y(.) is the learning function, which assigns different attention weights to different historical time steps of the input, W A and c A These are the trainable weights and biases, γ. N This is the output of a nested LSTM neural network, where ⊙ represents the matrix element-wise product. This is the weighted output of the nested LSTM neural network.
[0096] External feature acquisition unit 204: Acquires external features including rail transit timetable features, surrounding bus station distribution features, and regional land use features;
[0097] Rail transit operations have distinct characteristics. For example, compared to buses, trains operate according to precise timetables. Passengers typically arrive at the platform as the train approaches its destination, resulting in small peaks in passenger flow. Furthermore, to ensure a good match between capacity and passenger volume, train timetables are designed with passenger volume at specific times in mind; the more trains scheduled to arrive within a given time window, the greater the passenger flow will be within that window.
[0098] The characteristic of rail transit timetable is N s,t ;
[0099]
[0100] N s,t This represents the number of trains arriving at station s during time period t, where This indicates the number of trains arriving at station s in the upstream direction during time period t. This represents the number of trains arriving at station s in the down direction during time period t;
[0101] Rail transit, with its advantages of large capacity, high punctuality, and speed, meets the daily travel needs of urban residents to a certain extent. However, it has certain limitations in terms of station locations and routes, and commuters typically connect with buses, shared bicycles, and other modes of transportation. Therefore, the distribution and number of bus stops around rail transit stations have a certain impact on subway passenger flow.
[0102] Characteristics of the distribution of surrounding bus stops (B) s =[b s,1 ,b s,2 ,b s,3 ]
[0103] Among them, bs,1 ,b s,2 ,b s,3 This indicates the number of bus stops within 500, 1000, and 1500 meters of a rail transit station s.
[0104] Points of Interest (POIs) are a concept in Geographic Information Systems (GIS), referring to geographic entities closely related to people's lives, such as schools, banks, restaurants, hospitals, and supermarkets. The distribution of POIs in a region can reflect its land use characteristics to some extent. Passenger flow at rail transit stations is closely related to the facilities surrounding them. If a subway station is surrounded by many residential communities, its passenger flow will primarily consist of morning and evening commuters; if there are many restaurants nearby, the station will experience higher passenger volume during mealtimes. This study uses web crawling technology to obtain POI data from the Gaode Maps Open Platform to represent the land use around rail transit stations.
[0105] Regional land use characteristics P s =[p s,tra ,p s,din ,p s,biz ,p s,int ,p s,res ,p s,gov ]
[0106] Among them, P s This represents the number of point of interest categories around site s, where transportation facilities / car services are represented by p. s,ra It indicates that the catering service category uses p s,din This indicates that commercial categories use p. s,biz This indicates that scenic spots and historical sites are represented by p. s,int This indicates that residential services are represented by p. s,res It indicates that government and financial institutions use p. s,gov express.
[0107] Prediction Unit 205: The feature fusion layer connects the extracted spatial features of passenger travel demand between stations, temporal features of passenger travel demand between stations, rail transit timetable features, surrounding bus station distribution features, and regional land use features. An attention-based NLSTM neural network is used to decode the fused feature variables to predict the number of passengers entering and leaving each station in the rail transit grid.
[0108] Specifically, during the training process, the Adam optimizer is used to optimize the proposed MLT-STNN model, and the loss function of the model is the mean absolute error.
[0109] Figure 3As shown, an embodiment of the present invention provides an electronic device 300, including a memory 310, a processor 320, and computer code 311 stored in the memory 320 and executable on the processor 320. When the processor 320 executes the computer code 311, it implements an urban rail transit passenger flow prediction method provided by the embodiment of the present invention.
[0110] Since the electronic device described in this embodiment is the device used to implement the embodiments of the present invention, those skilled in the art can understand the specific implementation methods and various variations of the electronic device in this embodiment based on the methods described in the embodiments of the present invention. Therefore, how the electronic device implements the methods in the embodiments of the present invention will not be described in detail here. Any device used by those skilled in the art to implement the methods in the embodiments of the present invention is within the scope of protection of the present invention.
[0111] Please see Figure 4 , Figure 4 This is a schematic diagram of an embodiment of a computer-readable storage medium provided in this invention.
[0112] like Figure 4 As shown, this embodiment provides a computer-readable storage medium 400, on which a computer program 411 is stored. When the computer program 411 is executed by a processor, it implements a method for predicting urban rail transit passenger flow provided in this embodiment of the invention.
[0113] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.
[0114] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0115] This invention provides a method for predicting passenger flow in urban rail transit. Specifically, the method includes: acquiring the passenger flow entering and exiting stations up to the current time interval t; using a ResCNN module to extract the spatial features of the passenger flow entering and exiting stations to obtain the spatial features of passenger travel demand between stations; using a nested LSTM neural network to extract the temporal features of the passenger flow entering and exiting stations to obtain the temporal features of passenger travel demand between stations; and acquiring external features including rail transit timetable features, surrounding bus station distribution features, and regional land features. This invention proposes an urban rail transit passenger flow prediction method. It utilizes a feature fusion layer to connect the extracted spatial features of passenger travel demand between stations, the temporal features of passenger travel demand between stations, rail transit timetable features, surrounding bus station distribution features, and regional land use features. An attention-based nested LSTM neural network is then used to decode the fused feature variables to predict the inbound and outbound passenger flows at each station in the rail transit grid. This method employs a customized neural network extraction module, considers the extraction of spatiotemporal travel features, and uses a multi-task learning method to characterize the correlation between inbound and outbound passenger flows, thereby improving prediction accuracy.
[0116] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. The above descriptions are merely specific embodiments of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.
[0117] The above are merely specific embodiments of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention using this concept shall be deemed as infringing upon the protection scope of the present invention.
Claims
1. A method for predicting passenger flow in urban rail transit, characterized in that, include, Get the inbound and outbound passenger flow of the station before the current time interval t; The ResCNN module was used to extract the spatial features of passenger flow entering and exiting the station, thus obtaining the spatial features of passenger travel demand between stations. Nested LSTM neural networks are used to extract the temporal features of passenger flow entering and exiting stations, thereby obtaining the temporal features of passenger travel demand between stations. External features to be acquired include rail transit timetable features, surrounding bus station distribution features, and regional land use features; A feature fusion layer is used to connect the extracted spatial features of passenger travel demand between stations, temporal features of passenger travel demand between stations, rail transit timetable features, surrounding bus station distribution features, and regional land use features. An attention-based nested LSTM neural network is used to decode the fused feature variables to predict the number of passengers entering and leaving each station in the rail transit grid.
2. The urban rail transit passenger flow prediction method according to claim 1, characterized in that, The ResCNN module is used to extract the spatial features of passenger flow entering and exiting the stations, thereby obtaining the spatial features of passenger travel demand between stations, specifically: The ResCNN module: X l+1 =R(X l+1 ,the l )+X l Among them, X l+1 and X l Let θ represent the input and output of the residual unit, respectively, and R be the residual function; l It includes all the learnable parameters of the residual unit; A batch normalization layer is used to standardize the input data, and a convolutional layer is added to the residual unit to extract the spatial features of passenger travel demand between stations.
3. The urban rail transit passenger flow prediction method according to claim 1, characterized in that, A nested LSTM neural network is used to extract the temporal features of passenger flow entering and exiting the stations, thereby obtaining the temporal features of passenger travel demand between stations, specifically: In attention-based nested LSTM neural networks, the trained attention weights α are used for weighting, and the calculation method is as follows: α=y(W A ⊙γ N +c A ) Where y(.) is the learning function, which assigns different attention weights to different historical time steps of the input, W A and c A These are the trainable weights and biases, γ. N This is the output of a nested LSTM neural network, where ⊙ represents the matrix element-wise product. This is the weighted output of the nested LSTM neural network.
4. The urban rail transit passenger flow prediction method according to claim 1, characterized in that, External features obtained include rail transit timetable features, surrounding bus station distribution features, and regional land use features, among which, The characteristic of rail transit timetable is N s,t ; N s,t This represents the number of trains arriving at station s during time period t, where This indicates the number of trains arriving at station s in the upstream direction during time period t. This represents the number of trains arriving at station s in the down direction during time period t; Characteristics of the distribution of surrounding bus stops (B) s =[b s,1 ,b s,2 ,b s,3 ] Among them, b s,1 ,b s,2 ,b s,3 This indicates the number of bus stops within 500, 1000, and 1500 meters of a rail transit station (s). Regional land use characteristics P s =[p s,tra ,p s,din ,p s,biz ,p s,int ,p s,res ,p s,gov ] Among them, P s This represents the number of point of interest categories around site s, where transportation facilities / car services are represented by p. s,tra It indicates that the catering service category uses p s,din This indicates that commercial categories use p. s,biz This indicates that scenic spots and historical sites are represented by p. s,int This indicates that residential services are represented by p. s,res It indicates that government and financial institutions use p. s,gov express.
5. The urban rail transit passenger flow prediction method according to claim 1, characterized in that, During the training process, the Adam optimizer is used to optimize the proposed MLT-STNN model, and the loss function of the model is the mean absolute error.
6. The urban rail transit passenger flow prediction method according to claim 1, characterized in that, The specific prediction model for urban rail transit passenger flow forecasting is as follows: Where P represents the historical time step used for prediction, and T represents the prediction time step. Let m represent the set of passenger travel volumes at subway stations within time interval t, and let E represent the prediction task. t As an external feature, the station's passenger flow in and out of the station before the current time interval t. Passenger flow in and out of the station at the next or several future time intervals.
7. A passenger flow prediction system for urban rail transit, characterized in that, include, Historical passenger flow acquisition unit: Acquires the inbound and outbound passenger flow of the station up to the current time interval t; Spatial feature extraction unit: The ResCNN module is used to extract the spatial features of passenger flow entering and exiting the station, so as to obtain the spatial features of passenger travel demand between stations; Temporal feature extraction unit: A nested LSTM neural network is used to extract the temporal features of passenger flow entering and exiting the station, so as to obtain the temporal features of passenger travel demand between stations; External feature acquisition unit: Acquires external features including rail transit timetable features, surrounding bus station distribution features, and regional land use features; Prediction Unit: The feature fusion layer connects the extracted spatial features of passenger travel demand between stations, temporal features of passenger travel demand between stations, rail transit timetable features, surrounding bus station distribution features, and regional land use features. An attention-based NLSTM neural network is used to decode the fused feature variables to predict the number of passengers entering and leaving each station in the rail transit grid.
8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the method according to any one of claims 1-6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method described in any one of claims 1-6.