WO2021098619A1 - 一种短期地铁客流预测方法、系统及电子设备 - Google Patents

一种短期地铁客流预测方法、系统及电子设备 Download PDF

Info

Publication number
WO2021098619A1
WO2021098619A1 PCT/CN2020/128857 CN2020128857W WO2021098619A1 WO 2021098619 A1 WO2021098619 A1 WO 2021098619A1 CN 2020128857 W CN2020128857 W CN 2020128857W WO 2021098619 A1 WO2021098619 A1 WO 2021098619A1
Authority
WO
WIPO (PCT)
Prior art keywords
passenger flow
subway
matrix
layer
module
Prior art date
Application number
PCT/CN2020/128857
Other languages
English (en)
French (fr)
Inventor
叶洁瑕
赵娟娟
须成忠
刘延东
Original Assignee
中国科学院深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中国科学院深圳先进技术研究院 filed Critical 中国科学院深圳先进技术研究院
Publication of WO2021098619A1 publication Critical patent/WO2021098619A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/901—Indexing; Data structures therefor; Storage structures
    • G06F16/9024—Graphs; Linked lists
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/40—Business processes related to the transportation industry

Definitions

  • the invention relates to the field of spatiotemporal data mining or smart transportation technology, and more specifically, to a short-term subway passenger flow prediction method, system and electronic equipment.
  • passenger flow forecasting includes passenger flow forecasting at the granularity of time period, day, month, and quarter.
  • the research on short-term passenger flow forecasting began in the 20th century, and has achieved a wealth of research results.
  • the research methods are divided into the following categories: The first is the traditional linear forecasting model, including time series forecasting, Kalman filter model and linear regression model.
  • the second type is nonlinear prediction models, including traditional models based on wavelet theory, models based on chaos theory, models based on non-parametric regression, and models such as support vector machines and neural networks.
  • the third type is a prediction model based on simulation technology, including traffic simulation prediction method, dynamic traffic allocation prediction method, and cellular automata prediction method.
  • the purpose of the present invention is to provide a short-term subway passenger flow prediction method, system, and electronic equipment in view of the technical problems existing in the prior art, which can measure the passenger flow (including inflow and outflow) of each station in the subway network in a short period of time. Effective real-time forecasting.
  • a short-term subway passenger flow forecasting method includes:
  • Step a Collect subway source data
  • Step b The feature extraction module extracts the passenger flow feature matrix, constructs the directed power map of the subway, and builds the proximity matrix of the target station;
  • Step c According to the extracted passenger flow feature matrix and the proximity matrix of the target station, establish a graph convolutional neural network model to predict the passenger flow of the target station in the next time period;
  • Step d Output the passenger flow forecast value of the target station in the next time period.
  • Step S1 According to the collected subway source data, extract the number of people entering and leaving the gate in the set time period of all stations as the passenger flow feature matrix, and extract the number of people leaving the gate at the target station in the next time period as the passenger flow prediction target ;
  • Step S3 According to the constructed directed right graph, the proximity matrix of the target site is constructed, which indicates the shortest time it takes for all other sites to reach the target site.
  • extracting a passenger flow feature matrix and a passenger flow prediction target specifically includes the following:
  • Step S11 According to the obtained subway source data, count the number of passengers entering and leaving the gate at all stations;
  • Step S12 divide the time period with the time granularity t hours (t>0);
  • Step S13 Extract the number of people entering and leaving the gate of all stations in the divided time period as the passenger flow feature matrix
  • Step S14 Extract the number of people leaving the gate at the target station in the t+1 time period as the passenger flow prediction target.
  • step S3 building the proximity matrix of the target site specifically includes the following:
  • Step S31 According to the directed power map of the subway, for the target station that needs to predict its future passenger flow, by using the Dijkstra algorithm, calculate the shortest path predicted from the remaining stations to the target station;
  • Step S32 The obtained shortest path from each site to the target site is hierarchized according to the time granularity set in advance to indicate the spatial relationship between other sites and the target site;
  • Step S33 According to the obtained spatial relationship between each site and the target site, a proximity matrix is constructed for the target site, which indicates the shortest time it takes for all other sites to reach the target site.
  • the graph convolutional neural network model includes an input layer, a graph convolutional layer, and an output layer, wherein the graph convolutional layer includes a fusion layer and a fully connected layer; the graph convolutional layer uses two layers, the second of the model The layer and the third layer are both graph convolutional layers, and the first and fourth layers are the input layer and the output layer respectively; the method steps for predicting the passenger flow of the target site in the next time period are as follows:
  • Step Sa The input layer receives the passenger flow feature matrix and inputs it to the fusion layer of the second layer;
  • Step Sb The fusion layer also receives the proximity matrix of the target station, and multiplies the passenger flow feature matrix and the proximity matrix of the target station to obtain a fused new feature matrix, and outputs it to the corresponding fully connected layer;
  • Step Sc The fully connected layer is activated after receiving the input of the new feature matrix to obtain a higher-level passenger flow feature matrix, and extract abstract passenger flow features;
  • Step Sd The fusion layer of the third layer receives the higher-level passenger flow characteristic matrix, and repeats Sb-Sc;
  • Step Se The output layer obtains the passenger flow prediction value of the next time period of the target station according to the passenger flow feature matrix output by the fully connected layer in the third layer, and compares it with the passenger flow prediction target obtained in step S14 to obtain a reliable Output after the passenger flow forecast value.
  • the model formula of the new feature matrix is as follows:
  • H (l) represents the lth layer
  • H (l+1) represents the l+1 layer
  • A represents the matrix containing spatial structure information
  • W (l) is the weight matrix that needs to be trained for the l layer
  • ⁇ represents the activation function.
  • a short-term subway passenger flow prediction system which includes a collection module, a feature extraction module, a graph convolutional neural network module, and an output module;
  • the collection module is used to collect subway source data, including subway smart transaction card data sources and subway route maps;
  • the feature extraction module extracts a passenger flow feature matrix, constructs a directed right graph of the subway, and constructs a neighboring matrix of the target station;
  • the graph convolutional neural network module predicts the passenger flow of the target station in the next time period according to the extracted passenger flow feature matrix and the proximity matrix of the target station;
  • the output module is used to output the passenger flow forecast value of the target station in the next time period.
  • the feature extraction module includes an extraction module, a building module, and a building module;
  • the extraction module used to extract the passenger flow feature matrix to obtain the number of people entering the gate and the number of people leaving the gate within a set time period of all stations; also used to extract the passenger flow prediction target to obtain the number of people leaving the gate at the target station in the next time period ;
  • the building module used to build the proximity matrix of the target site, representing the shortest time it takes for all other sites to reach the target site.
  • the graph convolutional neural network module includes an input module, a graph convolution module, and a passenger flow prediction module, and the graph convolution module includes a fusion module and a fully connected module;
  • the input module used to receive the passenger flow feature matrix and input it to the fusion module;
  • the fusion module used to receive the passenger flow feature matrix and the proximity matrix of the target station, and multiply the two to obtain a new fusion feature matrix;
  • the fully connected module is used to activate after receiving a new feature matrix input to obtain a more advanced passenger flow feature matrix, and extract abstract passenger flow features;
  • the passenger flow prediction module used to compare the more advanced passenger flow feature matrix with the passenger flow prediction target, predict the passenger flow of the target station in the next time period, and obtain the passenger flow prediction value;
  • the model formula of the new feature matrix is as follows:
  • H (l) represents the lth layer
  • H (l+1) represents the l+1 layer
  • A represents the matrix containing spatial structure information
  • W (l) is the weight matrix that needs to be trained for the l layer
  • ⁇ represents the activation function.
  • An electronic device including:
  • At least one processor At least one processor
  • a memory communicatively connected with the at least one processor; wherein,
  • the memory stores instructions that can be executed by the one processor, and the instructions are executed by the at least one processor, so that the at least one processor can execute the short-term subway system described in any one of 1 to 6 above.
  • Step a Collect subway source data
  • Step b The feature extraction module extracts the passenger flow feature matrix, constructs the directed power map of the subway, and builds the proximity matrix of the target station;
  • Step c According to the extracted passenger flow feature matrix and the proximity matrix of the target station, establish a graph convolutional neural network model to predict the passenger flow of the target station in the next time period;
  • Step d Output the passenger flow forecast value of the target station in the next time period.
  • the present invention has the following advantages:
  • the invention builds a graph convolutional neural network model by collecting subway source data, extracting the passenger flow characteristic matrix and constructing the neighboring matrix of the target station, integrating the spatial structure and time factors of the subway network, and the passenger flow information of each station in the entire network, not only taking into account the time factor Taking into account the space factor of the subway, it can effectively predict the short-term subway passenger flow and improve the accuracy of short-term subway passenger flow prediction.
  • Figure 1 is a schematic diagram of the short-term subway passenger flow prediction method of the present invention.
  • Figure 2 is a flow chart of the short-term subway passenger flow prediction method of the present invention.
  • Figure 3 is a flow chart of the principle of the feature extraction module in the present invention.
  • Fig. 4 is a schematic diagram of the graph convolutional neural network model of the present invention.
  • Fig. 5 is a flowchart of the graph convolutional neural network model of the present invention.
  • Fig. 6 is a schematic diagram of the short-term subway passenger flow prediction system of the present invention.
  • FIG. 7 is a schematic diagram of the hardware device structure of the short-term subway passenger flow prediction method provided by the present invention.
  • the prediction method includes:
  • Step a Collect subway source data, including subway smart transaction card data source and subway route map.
  • Step b The feature extraction module extracts the passenger flow feature matrix, constructs the directed power map of the subway, and builds the proximity matrix of the target station.
  • Step c According to the extracted passenger flow feature matrix and the proximity matrix of the target station, a graph convolutional neural network model is established to predict the passenger flow of the target station in the next time period.
  • Step d Output the passenger flow forecast value of the target station in the next time period.
  • the feature extraction module performs the following operations according to the obtained subway source data:
  • Step S1 Extract the passenger flow feature matrix and the passenger flow prediction target, which specifically includes (continue to refer to Figure 3):
  • Step S11 According to the collected data source of the subway smart transaction card, count the number of passengers entering and leaving the gate of all stations.
  • Step S13 In the divided time period, that is, t-m+1, t-m+2, ..., t, in total m time periods, the number of people entering and leaving the gate of all stations is extracted as a passenger flow feature matrix.
  • Step S14 Extract the number of people leaving the gate at the target station in the t+1 time period as the passenger flow prediction target, and the passenger flow prediction target is the true value of the passenger flow in the t+1 time period.
  • step S1 it is assumed that the predicted time granularity t is half an hour, that is, the outbound passenger flow of the target station in the future half an hour is predicted.
  • the subway line graph is regarded as a directed right graph, and edges are established between two adjacent stations on the same subway line, and no edges are established between non-adjacent stations.
  • the edges are bidirectional, because adjacent stations can communicate with each other, and the weight of an edge is the time spent between stations.
  • Step S3 Build the proximity matrix of the target station, which represents the shortest time it takes for all the remaining stations to reach the target station, and is used to characterize the spatial structure of the subway.
  • the details include the following (continue to refer to Figure 3):
  • Step S31 According to the directed power map of the subway, use the Dijkstra algorithm (other algorithms can also be used for calculation) for the target station that needs to predict its future passenger flow, and calculate the predictions from the remaining stations to the target station The shortest path.
  • Dijkstra algorithm other algorithms can also be used for calculation
  • the shortest path between the subway stations refers to the departure from station A to station B, and there may be multiple lines between the two.
  • the line with the shortest time is the shortest path from station A to station B.
  • the time taken is the length of the shortest path.
  • Step S32 The obtained shortest path from each station to the target station is stratified according to the set time granularity V, which can be divided into n levels, namely 0-V, V-2V, ... (n-1)V- nV, which represents the spatial relationship between other sites and the target site.
  • V the set time granularity
  • n levels namely 0-V, V-2V, ... (n-1)V- nV, which represents the spatial relationship between other sites and the target site.
  • 0-V belongs to the first-tier neighbors of the target site
  • V-2V belongs to the second-tier neighbors of the target site, and so on.
  • the shortest paths from the remaining sites to the target site are stratified according to the time granularity of 5 minutes, that is, the first-tier neighbors belonging to the target site within 0-5 minutes, and the first neighbors belonging to the target site within 5-10 minutes. Second-tier neighbors, and so on.
  • Step S33 According to the obtained spatial relationship between each site and the target site, build a proximity matrix for the target site to indicate the shortest time it takes for the remaining sites to reach the target site.
  • the first row of the proximity matrix represents the first-tier neighbors of the target site, and the i-th row represents the i-th neighbors of the target site. Since the proximity matrix of the target station contains the time taken by all other stations to the target station, it can represent the spatial structure of the subway.
  • the graph convolutional neural network model includes an input layer, a graph convolution layer, and an output layer, where the graph convolution layer includes a fusion layer and a fully connected layer.
  • the graph convolutional layer is preferably two layers, that is, the graph convolutional neural network model includes four layers, the first layer is the input layer, the second and third layers are both graph convolutional layers, The fourth layer is the output layer.
  • the neural network model is based on the fully connected neural network, inserting a fusion layer before each fully connected layer, and the last layer is the output layer.
  • the graph convolutional layer may include one layer or more than one layer, and the number of layers can be increased or decreased according to actual needs. In this embodiment, the more layers are obtained through multiple experiments, the effect does not get better, but training The model time has increased, so the two most effective layers are used.
  • Step Sa The input layer receives the passenger flow feature matrix and inputs it to the fusion layer of the second layer.
  • Step Sb The fusion layer also receives the proximity matrix of the target station, and multiplies the passenger flow characteristic matrix and the proximity matrix of the target station to obtain a fused new characteristic matrix, and outputs it to the fully connected layer of the corresponding layer. Since the passenger flow feature matrix contains time information, and the neighbor matrix contains space information, the new feature matrix integrates time and space information.
  • H (l) represents the lth layer
  • H (l+1) represents the l+1 layer
  • A represents the matrix containing spatial structure information
  • W (l) is the weight matrix that needs to be trained for the l layer
  • ⁇ represents the activation function.
  • the matrix A is the matrix A containing spatial structure information that enables the graph convolutional neural network model to effectively extract spatial features on topological graphs (such as social networks and subway networks). Since the matrix A containing the spatial structure is the key to the graph convolutional neural network model, different matrices A represent different spatial information.
  • the matrix A is set as the proximity matrix of the site for the first time, so that the network module can obtain the characteristic information of the surrounding sites of each site.
  • Step Sc The fully connected layer is activated after receiving the input of the new feature matrix to obtain a higher-level passenger flow feature matrix, and extract abstract passenger flow features.
  • Step Sd The fusion layer of the third layer receives the higher-level passenger flow characteristic matrix, and repeats Sb-Sc.
  • the passenger flow characteristic matrix and the neighboring matrix undergo the function of the two-layer graph convolution layer, and continuously undergo linear and non-linear transformations to obtain a more advanced and abstract passenger flow characteristic matrix, which will get closer and closer to the true value.
  • Step Se The output layer obtains the passenger flow prediction value of the next time period of the target station according to the passenger flow feature matrix output by the fully connected layer in the third layer, and compares it with the passenger flow prediction target obtained in step S14 to obtain a reliable Output after the passenger flow forecast value.
  • the passenger flow forecast value is compared with the passenger flow forecast target, and the loss value is calculated by calculating the loss function.
  • the larger the loss value the greater the difference between the passenger flow forecast value and the passenger flow forecast target; the smaller the loss value, It shows that the gap between the two is smaller.
  • the model will change the weight parameters in the model towards the direction of decreasing loss, and keep repeating until the gap reaches the minimum value, so as to make the prediction effect of the model better and obtain a reliable passenger flow prediction value.
  • the present invention also provides a short-term subway passenger flow prediction system.
  • the prediction system includes an acquisition module, a feature extraction module, a graph convolutional neural network module, and an output module.
  • the collection module is used to collect subway source data, including subway smart transaction card data sources and subway route maps.
  • the feature extraction module extracts a passenger flow feature matrix, constructs a directed right graph of the subway, and constructs a neighboring matrix of the target station.
  • the graph convolutional neural network module predicts the passenger flow of the target station in the next time period according to the extracted passenger flow feature matrix and the proximity matrix of the target station.
  • the output module is used to output the passenger flow forecast value of the target station in the next time period.
  • the feature extraction module includes an extraction module, a building module, and a building module.
  • the extraction module is used to extract a passenger flow feature matrix and a passenger flow prediction target, which specifically includes:
  • the subway line graph is regarded as a directed right graph, and edges are established between two adjacent stations on the same subway line, and no edges are established between non-adjacent stations.
  • the edges are bidirectional, because adjacent stations can communicate with each other, and the weight of an edge is the time spent between stations.
  • the building module used to build the proximity matrix of the target station, which represents the shortest time it takes for all the other stations to reach the target station, and is used to characterize the spatial structure of the subway, including:
  • the shortest path predicted from the remaining stations to the target station is calculated .
  • the shortest path between subway stations refers to the departure from station A to station B.
  • the line with the shortest time is the shortest path from station A to station B, and the time taken is the shortest path length.
  • the obtained shortest path from each station to the target station is stratified according to the time granularity V set in advance, which can be divided into n levels, namely 0-V, V-2V, ... (n-1)V-nV , which indicates the spatial relationship between other sites and the target site.
  • n levels namely 0-V, V-2V, ... (n-1)V-nV , which indicates the spatial relationship between other sites and the target site.
  • 0-V belongs to the first-tier neighbors of the target site
  • V-2V belongs to the second-tier neighbors of the target site
  • the shortest paths from the remaining sites to the target site are layered according to the time granularity of 5 minutes, that is, the first-tier neighbors belonging to the target site within 0-5 minutes, and the second-tier neighbors belonging to the target site within 5-10 minutes. And so on.
  • a proximity matrix is constructed for the target site to indicate the shortest time it takes for all the remaining sites to reach the target site.
  • the first row of the proximity matrix represents the first-tier neighbors of the target site
  • the i-th row represents the i-th neighbors of the target site. Since the proximity matrix of the target station contains the time taken by all other stations to the target station, it can represent the spatial structure of the subway.
  • the graph convolutional neural network module includes an input module, a graph convolution module and a passenger flow prediction module, and the graph convolution module includes a fusion module and a fully connected module.
  • the input module used to receive the passenger flow feature matrix and input it to the fusion module.
  • the fusion module is used to receive the passenger flow feature matrix and the proximity matrix of the target station, and multiply the two to obtain a new fusion feature matrix. Since the passenger flow feature matrix contains time information, and the neighbor matrix contains space information, the new feature matrix integrates time and space information.
  • the model formula of the new feature matrix is as follows:
  • H (l) represents the lth layer
  • H (l+1) represents the l+1 layer
  • A represents the matrix containing spatial structure information
  • W (l) is the weight matrix that needs to be trained for the l layer
  • ⁇ represents the activation function.
  • the matrix A is the matrix A containing spatial structure information that enables the graph convolutional neural network module to effectively extract spatial features on topological graphs (such as social networks and subway networks). Since the matrix A containing the spatial structure is the key of the graph convolutional neural network module, different matrices A represent different spatial information.
  • the matrix A is set as the proximity matrix of the site for the first time, so that the network module can obtain the characteristic information of the surrounding sites of each site.
  • the fully connected module is used to activate after receiving a new feature matrix input to obtain a more advanced passenger flow feature matrix, and extract abstract passenger flow features.
  • the passenger flow prediction module used to compare the more advanced passenger flow feature matrix with the passenger flow prediction target, predict the passenger flow of the target station in the next time period, and obtain the passenger flow prediction value.
  • the Dijkstra algorithm is used to calculate the shortest path length from the target station to other stations, and based on the shortest path, a neighboring matrix of the target station is constructed, which contains the spatial relationship between the target station and surrounding stations .
  • the present invention applies the graph convolutional neural network model to the subway passenger flow prediction scene for the first time, which includes a four-layer structure, the first layer is the input layer, the second and third layers are both graph convolutional layers, and the fourth layer It is the output layer.
  • the graph convolution layer includes a fusion layer and a fully connected layer. In the fusion layer, the proximity matrix of the target station is multiplied by the passenger flow feature matrix to obtain a new feature matrix containing spatiotemporal information. Therefore, the present invention not only takes into account the time factor but also the space factor of the subway, it more effectively predicts the characteristics of the short-term passenger flow, and improves the accuracy of the short-term subway passenger flow prediction.
  • Fig. 7 is a schematic diagram of the hardware device structure of the short-term subway passenger flow prediction method provided by the present invention.
  • the device includes one or more processors and memory. Taking a processor as an example, the device may also include: an input system and an output system.
  • the processor, the memory, the input system, and the output system may be connected by a bus or in other ways.
  • the connection by a bus is taken as an example.
  • the memory can be used to store non-transitory software programs, non-transitory computer executable programs, and modules.
  • the processor executes various functional applications and data processing of the electronic device by running non-transitory software programs, instructions, and modules stored in the memory, that is, realizing the processing methods of the foregoing method embodiments.
  • the memory may include a program storage area and a data storage area, where the program storage area can store an operating system and an application program required by at least one function; the data storage area can store data and the like.
  • the memory may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices.
  • the memory may optionally include a memory remotely provided with respect to the processor, and these remote memories may be connected to the processing system through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
  • the input system can receive input digital or character information, and generate signal input.
  • the output system may include display devices such as a display screen.
  • the one or more modules are stored in the memory, and when executed by the one or more processors, the following operations of any of the foregoing method embodiments are performed:
  • Step a Collect subway source data
  • Step b The feature extraction module extracts the passenger flow feature matrix, constructs the directed power map of the subway, and builds the proximity matrix of the target station;
  • Step c According to the extracted passenger flow feature matrix and the proximity matrix of the target station, establish a graph convolutional neural network model to predict the passenger flow of the target station in the next time period;
  • Step d Output the passenger flow forecast value of the target station in the next time period.
  • the embodiment of the present application provides a non-transitory (non-volatile) computer storage medium, the computer storage medium stores computer executable instructions, and the computer executable instructions can perform the following operations:
  • Step a Collect subway source data
  • Step b The feature extraction module extracts the passenger flow feature matrix, constructs the directed power map of the subway, and builds the proximity matrix of the target station;
  • Step c According to the extracted passenger flow feature matrix and the proximity matrix of the target station, establish a graph convolutional neural network model to predict the passenger flow of the target station in the next time period;
  • Step d Output the passenger flow forecast value of the target station in the next time period.
  • the embodiment of the present application provides a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, when the program instructions are executed by a computer To make the computer do the following:
  • Step a Collect subway source data
  • Step b The feature extraction module extracts the passenger flow feature matrix, constructs the directed power map of the subway, and builds the proximity matrix of the target station;
  • Step c According to the extracted passenger flow feature matrix and the proximity matrix of the target station, establish a graph convolutional neural network model to predict the passenger flow of the target station in the next time period;
  • Step d Output the passenger flow forecast value of the target station in the next time period.
  • the short-term subway passenger flow prediction method, system and electronic equipment provided by the embodiments of the application not only take into account the time factor but also the space factor of the subway, it can effectively predict the short-term subway passenger flow and improve the accuracy of short-term subway passenger flow prediction.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • General Engineering & Computer Science (AREA)
  • Marketing (AREA)
  • Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Development Economics (AREA)
  • Evolutionary Computation (AREA)
  • Game Theory and Decision Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Biophysics (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Train Traffic Observation, Control, And Security (AREA)

Abstract

一种短期地铁客流预测方法、系统及电子设备，该预测方法首先采集地铁源数据(a)；特征提取模块提取客流特征矩阵，构建地铁的有向有权图，并搭建目标站点的邻近矩阵(b)；根据所提取的客流特征矩阵和目标站点的邻近矩阵，建立图卷积神经网络模型，对目标站点在下一个时间段的客流进行预测(c)；最后输出目标站点在下一个时间段的客流预测值(d)。不仅考虑到时间因素也考虑到地铁的空间因素，更加有效地预测了短时客流特征，提高了短期地铁客流预测的精度。

Description

一种短期地铁客流预测方法、系统及电子设备 技术领域
本发明涉及时空数据挖掘或智慧交通技术领域，更具体的说，特别涉及一种短期地铁客流预测方法、系统及电子设备。
背景技术
随着我国城镇化的推进，人口不断向大城市集中，为了解决交通拥挤和环境恶化的城市问题，我国己步入城市轨道交通高速发展期。地铁以其绿色、快捷方便、运量大和安全性高等特点，成为大城市最重要的出行方式。近年来，各大城市大量建成地铁线路并投入营运，但线路管理和调度等问题凸显。如早晚高峰时段的线路车次供给与需求不匹配，导致车次载满率过高而过度拥挤，降低承运安全性；而平峰和低峰车辆载满率过低而导致运能浪费。因此，需要一种能有效预测短期客流分布的方案来满足市民出行的需求，减少乘客的出行时间，为铁路企业进行资源(包括服务人员、应急物资等)调配，安全措施部署等提供决策依据。
以时间粒度划分，客流预测包括以时段、日、月和季度等粒度的客流预测。时间粒度越小，客流量数据的随机性越强，预测的难度越大。短期客流预测的研究始于20世纪，目前已取得丰富的研究成果。研究方法分为以下几种：第一种是传统的线性预测模型，包括时间序列预测法、卡尔曼滤波模型和线性回归模型等。第二种是非线性预测模型，包括传统的基于小波理论的模型、基于混沌理论的模型、基于非参数回归的模型等模型，还包括支持向量机、神经网络等。第三种是以仿真技术为基础的预测模型，包括交通模拟预测法、动态交通分配预测法及元胞自动机预测法等。
已有的模型，包括预测精度最高的神经网络类模型，在预测地铁站点的客流时，大部分仅仅使用了所预测站点过去各个时段的客流特征去预测该站点的未来各时段的客流，没有考虑到其它站点的客流对该站点的客流影响，即考虑了时间因素但忽略了空间因素。而地铁可以是看作一个网络，单个站点某个时间段内的进(出)客流，不仅受到 最近过去时段该站点的客流状况影响，也受到其它站点的客流状况的影响。据观察，其它各个站点的客流状况对当前站点的客流影响与站点之间的距离有关，距离越近则影响越大，距离越远则影响越小。
发明内容
本发明的目的在于针对现有技术存在的技术问题，提供一种短期地铁客流预测方法、系统及电子设备，能够对短期内地铁网络中的各站点的客流量(包括流入量和流出量)做出有效的实时预测。
为了解决以上提出的问题，本发明采用的技术方案为：
一种短期地铁客流预测方法，该预测方法包括：
步骤a：采集地铁源数据；
步骤b：特征提取模块提取客流特征矩阵，构建地铁的有向有权图，并搭建目标站点的邻近矩阵；
步骤c：根据所提取的客流特征矩阵和目标站点的邻近矩阵，建立图卷积神经网络模型，对目标站点在下一个时间段的客流进行预测；
步骤d：输出目标站点在下一个时间段的客流预测值。
进一步地，所述特征提取模块中具体方法步骤如下：
步骤S1：根据所采集的地铁源数据，提取所有站点在设定时间段内的入闸人数和出闸人数作为客流特征矩阵，及提取下一时间段内目标站点的出闸人数作为客流预测目标；
步骤S2：构建地铁的有向有权图G＝(V，E)，其中V是图的顶点集，E是图的边集，图的顶点即地铁的站点；
步骤S3：根据所构建的有向有权图，搭建目标站点的邻近矩阵，表示其余所有站点到目标站点的所花费的最短时间。
进一步地，所述步骤S1中，提取客流特征矩阵和客流预测目标，具体包括如下：
步骤S11：根据获取到的地铁源数据，统计所有站点乘客的入闸人数和出闸人数；
步骤S12：以时间粒度t小时(t>0)进行时间段划分；
步骤S13：提取所划分的时间段内所有站点的入闸人数和出闸人数作为客流特征矩阵；
步骤S14：提取t+1时间段内目标站点的出闸人数作为客流预测目标。
进一步地，所述步骤S3中，搭建目标站点的邻近矩阵，具体包括如下：
步骤S31：根据地铁的有向有权图，对需要预测其未来客流的目标站点，通过使用迪杰斯特拉算法，计算出其余各个站点到目标站点所预测的最短路径；
步骤S32：将所得到各个站点到达目标站点的最短路径，按照提前设定的时间粒度进行分层，表示其它各个站点与目标站点的空间关系；
步骤S33：根据所得到各个站点与目标站点的空间关系，为目标站点搭建一个邻近矩阵，表示其余所有站点到目标站点的所花费的最短时间。
进一步地，所述图卷积神经网络模型包括输入层、图卷积层和输出层，其中图卷积层包括融合层和全连接层；所述图卷积层采用两层，模型的第二层和第三层均为图卷积层，第一层和第四层分别为输入层和输出层；其对目标站点在下一个时间段的客流进行预测的方法步骤如下：
步骤Sa：输入层接收所述客流特征矩阵，并将其输入给第二层的融合层；
步骤Sb：所述融合层还接收目标站点的邻近矩阵，并将所述客流特征矩阵和目标站点的邻近矩阵相乘，得到融合的新特征矩阵，并输出给对应的全连接层；
步骤Sc：所述全连接层接收新特征矩阵输入后进行激活，得到更高级的客流特征矩阵，提取抽象的客流特征；
步骤Sd：所述第三层的融合层接收所述更高级的客流特征矩阵，并重复Sb-Sc；
步骤Se：所述输出层根据第三层中全连接层输出的客流特征矩阵，得到目标站点下一个时间段的客流预测值，并将其与步骤S14得到的客流预测目标进行对比，得到可靠的客流预测值后输出。
进一步地，所述步骤Sb中，所述新特征矩阵的模型公式如下：
H (l+1)＝σ(AH (l)W (l))
其中，H (l)代表第l层，H (l+1)代表第l+1层，A代表包含空间结构信息的矩阵，W (l)为第l层需要训练的权重矩阵，σ代表激活函数。
一种短期地铁客流预测系统，该预测系统包括采集模块、特征提取模块、图卷积神经网络模块和输出模块；
所述采集模块用于采集地铁源数据，包括地铁智能交易卡数据源和地铁线路图；
所述特征提取模块提取客流特征矩阵，构建地铁的有向有权图，并搭建目标站点的邻近矩阵；
所述图卷积神经网络模块根据所提取的客流特征矩阵和目标站点的邻近矩阵，对目标站点在下一个时间段的客流进行预测；
所述输出模块用于输出目标站点在下一个时间段的客流预测值。
进一步地，所述特征提取模块包括提取模块、构建模块和搭建模块；
所述提取模块：用于提取客流特征矩阵，得到所有站点在设定时间段内的入闸人数和出闸人数；还用于提取客流预测目标，得到下一时间段内目标站点的出闸人数；
所述构建模块：用于根据所得到的地铁线路图，构建地铁的有向有权图G＝(V，E)，其中V是图的顶点集，E是图的边集，图的顶点即地铁的站点；
所述搭建模块：用于搭建目标站点的邻近矩阵，表示其余所有站点到目标站点的所花费的最短时间。
进一步地，所述图卷积神经网络模块包括输入模块、图卷积模块和客流预测模块，所述图卷积模块包括融合模块和全连接模块；
所述输入模块：用于接收客流特征矩阵，并将其输入给融合模块；
所述融合模块：用于接收所述客流特征矩阵和目标站点的邻近矩阵，并将两者相乘，得到融合后的新特征矩阵；
所述全连接模块：用于接收新特征矩阵输入后进行激活，得到更高级的客流特征矩阵，提取抽象的客流特征；
所述客流预测模块：用于将更高级的客流特征矩阵和客流预测目标进行对比，预测目标站点在下一个时间段的客流量，得到客流预测值；
所述新特征矩阵的模型公式如下：
H (l+1)＝σ(AH (l)W (l))
其中，H (l)代表第l层，H (l+1)代表第l+1层，A代表包含空间结构信息的矩阵，W (l) 为第l层需要训练的权重矩阵，σ代表激活函数。
一种电子设备，包括：
至少一个处理器；以及
与所述至少一个处理器通信连接的存储器；其中，
所述存储器存储有可被所述一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行上述1至6任一项所述的短期地铁客流预测方法的以下操作：
步骤a：采集地铁源数据；
步骤b：特征提取模块提取客流特征矩阵，构建地铁的有向有权图，并搭建目标站点的邻近矩阵；
步骤c：根据所提取的客流特征矩阵和目标站点的邻近矩阵，建立图卷积神经网络模型，对目标站点在下一个时间段的客流进行预测；
步骤d：输出目标站点在下一个时间段的客流预测值。
与现有技术相比，本发明的有益效果在于：
本发明通过采集地铁源数据，提取客流特征矩阵并搭建目标站点的邻近矩阵，融合地铁网络的空间结构、时间因素，以及全网各个站点客流信息建立图卷积神经网络模型，不仅考虑到时间因素也考虑到地铁的空间因素，能够对短时地铁客流进行有效预测，也提高了短期地铁客流预测的精度。
附图说明
图1为本发明短期地铁客流预测方法的原理图。
图2为本发明短期地铁客流预测方法的流程图。
图3为本发明中特征提取模块的原理流程图。
图4为本发明中图卷积神经网络模型的原理图。
图5为本发明中图卷积神经网络模型的流程图。
图6为本发明短期地铁客流预测系统的原理图。
图7为本发明提供的短期地铁客流预测方法的硬件设备结构示意图。
具体实施方式
为了便于理解本发明，下面将参照相关附图对本发明进行更全面的描述。附图中给出了本发明的较佳实施例。但是，本发明可以以许多不同的形式来实现，并不限于本文所描述的实施例。相反地，提供这些实施例的目的是使对本发明的公开内容的理解更加透彻全面。
除非另有定义，本文所使用的所有的技术和科学术语与属于本发明的技术领域的技术人员通常理解的含义相同。本文中在本发明的说明书中所使用的术语只是为了描述具体的实施例的目的，不是旨在于限制本发明。
参阅图1和图2所示，本发明提供一种短期地铁客流预测方法，该预测方法包括：
步骤a：采集地铁源数据，包括地铁智能交易卡数据源和地铁线路图。
步骤b：特征提取模块提取客流特征矩阵，构建地铁的有向有权图，并搭建目标站点的邻近矩阵。
步骤c：根据所提取的客流特征矩阵和目标站点的邻近矩阵，建立图卷积神经网络模型，对目标站点在下一个时间段的客流进行预测。
步骤d：输出目标站点在下一个时间段的客流预测值。
参阅图3所示，所述特征提取模块根据所得到的地铁源数据，进行如下操作：
步骤S1：提取客流特征矩阵和客流预测目标，具体包括(继续参阅图3所示)：
步骤S11：根据采集到的地铁智能交易卡数据源，统计所有站点乘客的入闸人数和出闸人数。
步骤S12：以时间粒度t小时(t>0)进行时间段划分，假定地铁早上P时开始营运，晚上Q时停止，则(Q-P)/t+1＝m，即共划分m个时间段。
步骤S13：在所划分的时间段内即t-m+1，t-m+2，…，t共m个时间段内，提取所有站点的入闸人数和出闸人数作为客流特征矩阵。
步骤S14：提取t+1时间段内目标站点的出闸人数作为客流预测目标，所述客流预测目标为t+1时间段内客流的真实值。
本步骤S1中，假设预测的时间粒度t为半小时，即预测目标站点未来半小时的出 站客流。首先，将深圳通卡交易数据以半个小时为时间间隔统计每个站点乘客的入闸人数和出闸人数。每天从早上6时起，至晚上23.5时结束，划分为(23.5－6)/0.5+1＝35个时间段，统计全年每天每个时间段内，每个乘客站点的入闸人数和出闸人数。由于某些站点的开始营运时间可能比6时更晚，或者结束营运时间比23.5时更早，因而存在缺失的时间段。这时，可视为该站点在缺失的时间段内无乘客入闸或者无乘客出闸，将客流量设置为0。
步骤S2：根据所得到的地铁线路图，构建地铁的有向有权图G＝(V，E)，其中V是图的顶点集，E是图的边集，图的顶点即地铁的站点。
本步骤S2中，将地铁线路图视为一个有向有权图，同一条地铁线路上的相邻两个站点之间建立边，不相邻的站点之间不建立边。边是双向的，因为相邻站点可相互通车，边的权重是站点之间花费的时间。
步骤S3：搭建目标站点的邻近矩阵，表示其余所有站点到目标站点的所花费的最短时间，用以表征地铁的空间结构，具体包括如下(继续参阅图3所示)：
步骤S31：根据地铁的有向有权图，对需要预测其未来客流的目标站点，通过使用迪杰斯特拉算法(也可以采用其它算法进行计算)，计算出其余各个站点到目标站点所预测的最短路径。
本步骤S31中，所述地铁站点之间的最短路径是指站点A出发到站点B，两者之间可能存在多条线路，其中所花费时间最短的线路即从站点A到站点B的最短路径，而花费的时间是最短路径的长度。
步骤S32：将所得到各个站点到达目标站点的最短路径，按照设定的时间粒度V进行分层，可以划分为n个层次即0-V、V-2V、……(n-1)V-nV，从而表示其它各个站点与目标站点的空间关系。其中，0-V内属于目标站点的第一层邻居，V－2V内属于目标站点的第二层邻居，以此类推。
本步骤S32中，例如按照5分钟的时间粒度将其余各站点到达目标站点的最短路径分层，即0－5分钟内属于目标站点的第一层邻居，5－10分钟内属于目标站点的第二层邻居，以此类推。
步骤S33：根据所得到各个站点与目标站点的空间关系，为目标站点搭建一个邻近 矩阵，表示其余各个站点到目标站点所花费的最短时间。
本步骤S33中，所述邻近矩阵的第一行代表目标站点的第一层邻居，第i行代表目标站点的第i层邻居。由于目标站点的邻近矩阵都包含了其余所有站点到目标站点所花费时间，因此它能代表地铁的空间结构。
参阅图4所示，所述图卷积神经网络模型包括输入层、图卷积层和输出层，其中图卷积层包括融合层和全连接层。本实施例中，所述图卷积层优选为两层，即所述图卷积神经网络模型包括四层，第一层为输入层，第二层和第三层均为图卷积层，第四层为输出层。
上述中，所述神经网络模型在全连接神经网络的基础上，在每个全连接层前插入一个融合层，最后一层为输出层。所述图卷积层可以包括一层或一层以上，其层数可以根据实际需要进行增加或减少，本实施例经过多次实验得到层数越多，效果并没有变得更好，反而训练模型的时间增加了，因此采用效果最好的两层。
上述中，通过所述图卷积神经网络模型，对目标站点在下一个时间段的客流进行预测的方法步骤如下：
步骤Sa：输入层接收所述客流特征矩阵，并将其输入给第二层的融合层。
步骤Sb：所述融合层还接收目标站点的邻近矩阵，并将所述客流特征矩阵和目标站点的邻近矩阵相乘，得到融合的新特征矩阵，并输出给对应层的全连接层。由于所述客流特征矩阵中包含时间信息，所述邻近矩阵包含空间信息，因而新的特征矩阵融合了时间和空间信息。
本步骤Sb中，所述新特征矩阵的模型公式如下：
H (l+1)＝σ(AH (l)W (l))
其中，H (l)代表第l层，H (l+1)代表第l+1层，A代表包含空间结构信息的矩阵，W (l)为第l层需要训练的权重矩阵，σ代表激活函数。
对比全连神经网络模块的模型公式：
H (l+1)＝σ(H (l)W (l))
可以发现，上述图卷积神经网络模型和全连接神经网络模块的不同之处在于矩阵A。正是有了包含空间结构信息的矩阵A，才让所述图卷积神经网络模型能在拓扑图(如社 交网络、地铁网络)上有效地提取空间特征。由于包含空间结构的矩阵A是图卷积神经网络模型的关键，不同的矩阵A代表不同的空间信息。在本发明中，第一次将矩阵A设置为站点的邻近矩阵，这样能让网络模块获取到每个站点的周围站点的特征信息。
步骤Sc：所述全连接层接收新特征矩阵输入后进行激活，得到更高级的客流特征矩阵，提取抽象的客流特征。
步骤Sd：所述第三层的融合层接收所述更高级的客流特征矩阵，并重复Sb-Sc。
上述中，所述客流特征矩阵和邻近矩阵经过两层图卷积层的作用，不断经过了线性和非线性的变换，得到更高级和抽象的客流特征矩阵，其会越来越接近真实值。
步骤Se：所述输出层根据第三层中全连接层输出的客流特征矩阵，得到目标站点下一个时间段的客流预测值，并将其与步骤S14得到的客流预测目标进行对比，得到可靠的客流预测值后输出。
本步骤Se中，将客流预测值和客流预测目标进行比较，通过计算损失函数计算损失值，损失值越大，则表明客流预测值和客流预测目标两者的差距越大；损失值越小，则表明两者的差距越小。而模型会朝着损失变小的方向改变模型中的权重参数，不断重复直到差距取得最小值，从而让模型的预测效果变得更好，得到可靠的客流预测值。
参阅图6所示，本发明还提供一种短期地铁客流预测系统，该预测系统包括采集模块、特征提取模块、图卷积神经网络模块和输出模块。
所述采集模块用于采集地铁源数据，包括地铁智能交易卡数据源和地铁线路图。
所述特征提取模块提取客流特征矩阵，构建地铁的有向有权图，并搭建目标站点的邻近矩阵。
所述图卷积神经网络模块根据所提取的客流特征矩阵和目标站点的邻近矩阵，对目标站点在下一个时间段的客流进行预测。
所述输出模块用于输出目标站点在下一个时间段的客流预测值。
进一步地，所述特征提取模块包括提取模块、构建模块和搭建模块。
所述提取模块：用于提取客流特征矩阵和客流预测目标，具体包括：
根据获取到的地铁源数据，统计所有站点乘客的入闸人数和出闸人数。
以时间粒度t小时(t>0)进行时间段划分，假定地铁早上P时开始营运，晚上Q 时停止，则(Q-P)/t+1＝m，即共划分m个时间段。
在所划分的时间段内即t-m+1，t-m+2，…，t共m个时间段内，提取所有站点的入闸人数和出闸人数作为客流特征矩阵，并提取t+1时间段内目标站点的出闸人数作为客流预测目标。
所述构建模块：用于根据所得到的地铁线路图，构建地铁的有向有权图G＝(V，E)，其中V是图的顶点集，E是图的边集，图的顶点即地铁的站点。进一步地，将地铁线路图视为一个有向有权图，同一条地铁线路上的相邻两个站点之间建立边，不相邻的站点之间不建立边。边是双向的，因为相邻站点可相互通车，边的权重是站点之间花费的时间。
所述搭建模块：用于搭建目标站点的邻近矩阵，表示其余所有站点到目标站点的所花费的最短时间，用以表征地铁的空间结构，具体包括：
根据地铁的有向有权图，对需要预测其未来客流的目标站点，通过使用迪杰斯特拉算法(也可以采用其它算法进行计算)，计算出其余各个站点到目标站点所预测的最短路径。其中地铁站点之间的最短路径是指站点A出发，到站点B，可能存在多条线路，其中所花费时间最短的线路即从站点A到站点B的最短路径，而花费的时间是最短路径的长度。
将所得到各个站点到达目标站点的最短路径，按照提前设定的时间粒度V进行分层，可以划分为n个层次，即0-V、V-2V、……(n-1)V-nV，从而表示其它各个站点与目标站点的空间关系。其中，0-V内属于目标站点的第一层邻居，V－2V内属于目标站点的第二层邻居，以此类推。例如按照5分钟的时间粒度将其余各站点到达目标站点的最短路径分层，即0－5分钟内属于目标站点的第一层邻居，5－10分钟内属于目标站点的第二层邻居，以此类推。
根据所得到各个站点与目标站点的空间关系，为目标站点搭建一个邻近矩阵，表示其余所有站点到目标站点的所花费的最短时间。其中，所述邻近矩阵的第一行代表目标站点的第一层邻居，第i行代表目标站点的第i层邻居。由于目标站点的邻近矩阵都包含了其余所有站点到目标站点所花费时间，因此它能代表地铁的空间结构。
继续参阅图6所示，所述图卷积神经网络模块包括输入模块、图卷积模块和客流预 测模块，所述图卷积模块包括融合模块和全连接模块。
所述输入模块：用于接收客流特征矩阵，并将其输入给融合模块。
所述融合模块：用于接收所述客流特征矩阵和目标站点的邻近矩阵，并将两者相乘，得到融合后的新特征矩阵。由于所述客流特征矩阵中包含时间信息，所述邻近矩阵包含空间信息，因而新的特征矩阵融合了时间和空间信息。
其中，所述新特征矩阵的模型公式如下：
H (l+1)＝σ(AH (l)W (l))
其中，H (l)代表第l层，H (l+1)代表第l+1层，A代表包含空间结构信息的矩阵，W (l)为第l层需要训练的权重矩阵，σ代表激活函数。
对比全连神经网络模块的模型公式：
H (l+1)＝σ(H (l)W (l))
可以发现，上述图卷积神经网络模块和全连接神经网络模块的不同之处在于矩阵A。正是有了包含空间结构信息的矩阵A，才让所述图卷积神经网络模块能在拓扑图(如社交网络、地铁网络)上有效地提取空间特征。由于包含空间结构的矩阵A是图卷积神经网络模块的关键，不同的矩阵A代表不同的空间信息。在本发明中，第一次将矩阵A设置为站点的邻近矩阵，这样能让网络模块获取到每个站点的周围站点的特征信息。
所述全连接模块：用于接收新特征矩阵输入后进行激活，得到更高级的客流特征矩阵，提取抽象的客流特征。
所述客流预测模块：用于将更高级的客流特征矩阵和客流预测目标进行对比，预测目标站点在下一个时间段的客流量，得到客流预测值。
本发明中在地铁客流预测中，采用迪杰斯特拉算法计算目标站点到其他站点的最短路径长度，并基于最短路径构建了目标站点的邻近矩阵，该矩阵包含目标站点和周围站点的空间关系。此外，本发明首次将图卷积神经网络模型应用在地铁客流预测场景中，其包含四层结构，第一层是输入层，第二层、第三层均是图卷积层，第四层是输出层。所述图卷积层包含融合层和全连接层，在融合层中，目标站点的邻近矩阵和客流特征矩阵相乘，得到包含时空信息的新特征矩阵。因此，本发明不仅考虑到时间因素也考虑到地铁的空间因素，更加有效地预测了短时客流特征，提高了短期地铁客流预测的精度。
图7是本发明提供的短期地铁客流预测方法的硬件设备结构示意图。如图7所示，该设备包括一个或多个处理器以及存储器。以一个处理器为例，该设备还可以包括：输入系统和输出系统。
处理器、存储器、输入系统和输出系统可以通过总线或者其他方式连接，图7中以通过总线连接为例。
存储器作为一种非暂态计算机可读存储介质，可用于存储非暂态软件程序、非暂态计算机可执行程序以及模块。处理器通过运行存储在存储器中的非暂态软件程序、指令以及模块，从而执行电子设备的各种功能应用以及数据处理，即实现上述方法实施例的处理方法。
存储器可以包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需要的应用程序；存储数据区可存储数据等。此外，存储器可以包括高速随机存取存储器，还可以包括非暂态存储器，例如至少一个磁盘存储器件、闪存器件、或其他非暂态固态存储器件。在一些实施例中，存储器可选包括相对于处理器远程设置的存储器，这些远程存储器可以通过网络连接至处理系统。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
输入系统可接收输入的数字或字符信息，以及产生信号输入。输出系统可包括显示屏等显示设备。
所述一个或者多个模块存储在所述存储器中，当被所述一个或者多个处理器执行时，执行上述任一方法实施例的以下操作：
步骤a：采集地铁源数据；
步骤b：特征提取模块提取客流特征矩阵，构建地铁的有向有权图，并搭建目标站点的邻近矩阵；
步骤c：根据所提取的客流特征矩阵和目标站点的邻近矩阵，建立图卷积神经网络模型，对目标站点在下一个时间段的客流进行预测；
步骤d：输出目标站点在下一个时间段的客流预测值。
上述产品可执行本申请实施例所提供的方法，具备执行方法相应的功能模块和有益效果。未在本实施例中详尽描述的技术细节，可参见本申请实施例提供的方法。
本申请实施例提供了一种非暂态(非易失性)计算机存储介质，所述计算机存储介质存储有计算机可执行指令，该计算机可执行指令可执行以下操作：
步骤a：采集地铁源数据；
步骤b：特征提取模块提取客流特征矩阵，构建地铁的有向有权图，并搭建目标站点的邻近矩阵；
步骤c：根据所提取的客流特征矩阵和目标站点的邻近矩阵，建立图卷积神经网络模型，对目标站点在下一个时间段的客流进行预测；
步骤d：输出目标站点在下一个时间段的客流预测值。
本申请实施例提供了一种计算机程序产品，所述计算机程序产品包括存储在非暂态计算机可读存储介质上的计算机程序，所述计算机程序包括程序指令，当所述程序指令被计算机执行时，使所述计算机执行以下操作：
步骤a：采集地铁源数据；
步骤b：特征提取模块提取客流特征矩阵，构建地铁的有向有权图，并搭建目标站点的邻近矩阵；
步骤c：根据所提取的客流特征矩阵和目标站点的邻近矩阵，建立图卷积神经网络模型，对目标站点在下一个时间段的客流进行预测；
步骤d：输出目标站点在下一个时间段的客流预测值。
本申请实施例提供的短期地铁客流预测方法、系统及电子设备，不仅考虑到时间因素也考虑到地铁的空间因素，能够对短时地铁客流进行有效预测，也提高了短期地铁客流预测的精度。
上述实施例为本发明较佳的实施方式，但本发明的实施方式并不受上述实施例的限制，其他的任何未背离本发明的精神实质与原理下所作的改变、修饰、替代、组合、简化，均应为等效的置换方式，都包含在本发明的保护范围之内。

Claims (10)

  1. 一种短期地铁客流预测方法，其特征在于：该预测方法包括：
    步骤a：采集地铁源数据；
    步骤b：特征提取模块提取客流特征矩阵，构建地铁的有向有权图，并搭建目标站点的邻近矩阵；
    步骤c：根据所提取的客流特征矩阵和目标站点的邻近矩阵，建立图卷积神经网络模型，对目标站点在下一个时间段的客流进行预测；
    步骤d：输出目标站点在下一个时间段的客流预测值。
  2. 根据权利要求1所述的短期地铁客流预测方法，其特征在于：所述特征提取模块中具体方法步骤如下：
    步骤S1：根据所采集的地铁源数据，提取所有站点在设定时间段内的入闸人数和出闸人数作为客流特征矩阵，及提取下一时间段内目标站点的出闸人数作为客流预测目标；
    步骤S2：构建地铁的有向有权图G＝(V，E)，其中V是图的顶点集，E是图的边集，图的顶点即地铁的站点；
    步骤S3：根据所构建的有向有权图，搭建目标站点的邻近矩阵，表示其余所有站点到目标站点的所花费的最短时间。
  3. 根据权利要求2所述的短期地铁客流预测方法，其特征在于：所述步骤S1中，提取客流特征矩阵和客流预测目标，具体包括如下：
    步骤S11：根据获取到的地铁源数据，统计所有站点乘客的入闸人数和出闸人数；
    步骤S12：以时间粒度t小时(t>0)进行时间段划分；
    步骤S13：提取所划分的时间段内所有站点的入闸人数和出闸人数作为客流特征矩阵；
    步骤S14：提取t+1时间段内目标站点的出闸人数作为客流预测目标。
  4. 根据权利要求2所述的短期地铁客流预测方法，其特征在于：所述步骤S3中，搭建目标站点的邻近矩阵，具体包括如下：
    步骤S31：根据地铁的有向有权图，对需要预测其未来客流的目标站点，通过使用迪杰斯特拉算法，计算出其余各个站点到目标站点所预测的最短路径；
    步骤S32：将所得到各个站点到达目标站点的最短路径，按照提前设定的时间粒度 进行分层，表示其它各个站点与目标站点的空间关系；
    步骤S33：根据所得到各个站点与目标站点的空间关系，为目标站点搭建一个邻近矩阵，表示其余所有站点到目标站点的所花费的最短时间。
  5. 根据权利要求2－4任一项所述的短期地铁客流预测方法，其特征在于：所述图卷积神经网络模型包括输入层、图卷积层和输出层，其中图卷积层包括融合层和全连接层；所述图卷积层采用两层，模型的第二层和第三层均为图卷积层，第一层和第四层分别为输入层和输出层；其对目标站点在下一个时间段的客流进行预测的方法步骤如下：
    步骤Sa：输入层接收所述客流特征矩阵，并将其输入给第二层的融合层；
    步骤Sb：所述融合层还接收目标站点的邻近矩阵，并将所述客流特征矩阵和目标站点的邻近矩阵相乘，得到融合的新特征矩阵，并输出给对应的全连接层；
    步骤Sc：所述全连接层接收新特征矩阵输入后进行激活，得到更高级的客流特征矩阵，提取抽象的客流特征；
    步骤Sd：所述第三层的融合层接收所述更高级的客流特征矩阵，并重复Sb-Sc；
    步骤Se：所述输出层根据第三层中全连接层输出的客流特征矩阵，得到目标站点下一个时间段的客流预测值，并将其与步骤S14得到的客流预测目标进行对比，得到可靠的客流预测值后输出。
  6. 根据权利要求5所述的短期地铁客流预测方法，其特征在于：所述步骤Sb中，所述新特征矩阵的模型公式如下：
    H (l+1)＝σ(AH (l)W (l))
    其中，H (l)代表第l层，H (l+1)代表第l+1层，A代表包含空间结构信息的矩阵，W (l)为第l层需要训练的权重矩阵，σ代表激活函数。
  7. 一种短期地铁客流预测系统，其特征在于：该预测系统包括采集模块、特征提取模块、图卷积神经网络模块和输出模块；
    所述采集模块用于采集地铁源数据，包括地铁智能交易卡数据源和地铁线路图；
    所述特征提取模块提取客流特征矩阵，构建地铁的有向有权图，并搭建目标站点的邻近矩阵；
    所述图卷积神经网络模块根据所提取的客流特征矩阵和目标站点的邻近矩阵，对目 标站点在下一个时间段的客流进行预测；
    所述输出模块用于输出目标站点在下一个时间段的客流预测值。
  8. 根据权利要求7所述的短期地铁客流预测系统，其特征在于：所述特征提取模块包括提取模块、构建模块和搭建模块；
    所述提取模块：用于提取客流特征矩阵，得到所有站点在设定时间段内的入闸人数和出闸人数；还用于提取客流预测目标，得到下一时间段内目标站点的出闸人数；
    所述构建模块：用于根据所得到的地铁线路图，构建地铁的有向有权图G＝(V，E)，其中V是图的顶点集，E是图的边集，图的顶点即地铁的站点；
    所述搭建模块：用于搭建目标站点的邻近矩阵，表示其余所有站点到目标站点的所花费的最短时间。
  9. 根据权利要求7所述的短期地铁客流预测系统，其特征在于：所述图卷积神经网络模块包括输入模块、图卷积模块和客流预测模块，所述图卷积模块包括融合模块和全连接模块；
    所述输入模块：用于接收客流特征矩阵，并将其输入给融合模块；
    所述融合模块：用于接收所述客流特征矩阵和目标站点的邻近矩阵，并将两者相乘，得到融合后的新特征矩阵；
    所述全连接模块：用于接收新特征矩阵输入后进行激活，得到更高级的客流特征矩阵，提取抽象的客流特征；
    所述客流预测模块：用于将更高级的客流特征矩阵和客流预测目标进行对比，预测目标站点在下一个时间段的客流量，得到客流预测值；
    所述新特征矩阵的模型公式如下：
    H (l+1)＝σ(AH (l)W (l))
    其中，H (l)代表第l层，H (l+1)代表第l+1层，A代表包含空间结构信息的矩阵，W (l)为第l层需要训练的权重矩阵，σ代表激活函数。
  10. 一种电子设备，包括：
    至少一个处理器；以及
    与所述至少一个处理器通信连接的存储器；其中，
    所述存储器存储有可被所述一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行上述1至6任一项所述的短期地铁客流预测方法的以下操作：
    步骤a：采集地铁源数据；
    步骤b：特征提取模块提取客流特征矩阵，构建地铁的有向有权图，并搭建目标站点的邻近矩阵；
    步骤c：根据所提取的客流特征矩阵和目标站点的邻近矩阵，建立图卷积神经网络模型，对目标站点在下一个时间段的客流进行预测；
    步骤d：输出目标站点在下一个时间段的客流预测值。
PCT/CN2020/128857 2019-11-19 2020-11-13 一种短期地铁客流预测方法、系统及电子设备 WO2021098619A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911132859.7A CN110866649B (zh) 2019-11-19 2019-11-19 一种短期地铁客流预测方法、系统及电子设备
CN201911132859.7 2019-11-19

Publications (1)

Publication Number Publication Date
WO2021098619A1 true WO2021098619A1 (zh) 2021-05-27

Family

ID=69655156

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/128857 WO2021098619A1 (zh) 2019-11-19 2020-11-13 一种短期地铁客流预测方法、系统及电子设备

Country Status (2)

Country Link
CN (1) CN110866649B (zh)
WO (1) WO2021098619A1 (zh)

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113537580A (zh) * 2021-06-28 2021-10-22 中科领航智能科技(苏州)有限公司 一种基于自适应图学习的公共交通客流预测方法及系统
CN113850295A (zh) * 2021-08-27 2021-12-28 北京工业大学 一种基于Weighted-LeaderRank和GMM聚类的站点选址预测方法
CN114117903A (zh) * 2021-11-22 2022-03-01 合肥工业大学 一种基于bp神经网络的轨道交通短时客流预测方法
CN114154393A (zh) * 2021-10-15 2022-03-08 广州市交通规划研究院 一种基于腹地理论的目标地机场群旅客吞吐量预测方法
CN114399726A (zh) * 2021-12-06 2022-04-26 上海市黄浦区城市运行管理中心(上海市黄浦区城市网格化综合管理中心、上海市黄浦区大数据中心) 实时智能监测客流及预警的方法和系统
CN114463978A (zh) * 2022-02-10 2022-05-10 深圳明弘电子科技有限公司 一种基于轨道交通信息处理终端的数据监测方法
CN114492951A (zh) * 2022-01-06 2022-05-13 青岛海信网络科技股份有限公司 一种基于客流预测的资源调配方法、装置及设备
CN114581846A (zh) * 2022-03-03 2022-06-03 北京城建设计发展集团股份有限公司 地铁车站全息客流实时监测方法及装置及计算机设备
CN114708013A (zh) * 2022-03-21 2022-07-05 思创数码科技股份有限公司 点击率预测方法、系统、计算机及可读存储介质
CN114819290A (zh) * 2022-04-04 2022-07-29 南京行者易智能交通科技有限公司 一种基于stacking的多环境下短时公交客流预测方法
CN114912683A (zh) * 2022-05-13 2022-08-16 中铁第六勘察设计院集团有限公司 一种智慧城市轨道交通异常大客流预测系统及方法
CN115031794A (zh) * 2022-04-29 2022-09-09 天津大学 一种多特征图卷积的新型气固两相流流量测量方法
CN115620525A (zh) * 2022-12-16 2023-01-17 中国民用航空总局第二研究所 基于时变动态贝叶斯网络的短时交通客运需求预测方法
CN117272848A (zh) * 2023-11-22 2023-12-22 上海随申行智慧交通科技有限公司 基于时空影响的地铁客流预测方法及模型训练方法
CN117973640A (zh) * 2024-03-29 2024-05-03 煤炭科学研究总院有限公司 换乘地铁站换乘人流量预测方法及装置
CN118297419A (zh) * 2024-04-02 2024-07-05 北京交通大学 考虑事故状态的城市轨道交通短时od客流预测方法及系统

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110866649B (zh) * 2019-11-19 2022-09-20 中国科学院深圳先进技术研究院 一种短期地铁客流预测方法、系统及电子设备
CN111738535A (zh) * 2020-08-25 2020-10-02 北京交通大学 轨道交通时空短时客流预测方法、装置、设备及存储介质
CN111814920B (zh) * 2020-09-04 2021-01-05 中国科学院自动化研究所 基于图网络的多粒度特征学习的图像精细分类方法及系统
CN112418508B (zh) * 2020-11-19 2021-10-08 中国科学院地理科学与资源研究所 基于物理空间与社交网络空间交互的人口分布预测方法
CN112687110B (zh) * 2020-12-23 2023-02-28 江苏高立泊车科技有限公司 一种基于大数据分析的车位级导航方法及其系统
CN112990555A (zh) * 2021-02-24 2021-06-18 上海交通大学 基于时空并行栅格神经网络的地铁人流量预测方法及系统
CN113657934A (zh) * 2021-08-16 2021-11-16 京东城市(北京)数字科技有限公司 客流量预测模型的训练方法和客流量的预测方法和装置
CN115392752B (zh) * 2022-09-01 2023-11-28 亿雅捷交通系统(北京)有限公司 一种地铁短时客流预测方法、系统、电子设备及存储介质
CN117131999B (zh) * 2023-09-14 2024-04-30 浙江非线数联科技股份有限公司 基于数字孪生的轨道交通客流量预测系统及其方法

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109754605A (zh) * 2019-02-27 2019-05-14 中南大学 一种基于注意力时态图卷积网络的交通预测方法
CN110264709A (zh) * 2019-05-06 2019-09-20 北京交通大学 基于图卷积网络的道路的交通流量的预测方法
CN110866649A (zh) * 2019-11-19 2020-03-06 中国科学院深圳先进技术研究院 一种短期地铁客流预测方法、系统及电子设备

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103700255B (zh) * 2013-12-30 2015-10-07 复旦大学 一种基于时空关联数据挖掘的交通流预测方法
WO2018021958A1 (en) * 2016-07-27 2018-02-01 Hitachi, Ltd. Passenger load prediction systems and methods for predicting passenger load
CN107529651B (zh) * 2017-08-18 2020-10-16 北京航空航天大学 一种基于深度学习的城市交通客流预测方法和设备
CN108108844A (zh) * 2017-12-25 2018-06-01 儒安科技有限公司 一种城市人流量预测方法及系统
CN108564227B (zh) * 2018-04-26 2021-08-06 重庆大学 一种基于时空特征的轨道交通客流量预测方法
CN109886444B (zh) * 2018-12-03 2023-07-11 深圳市北斗智能科技有限公司 一种交通短时客流预测方法、装置、设备和存储介质
CN110322064B (zh) * 2019-06-27 2022-04-15 清华大学 城市出行需求预测方法

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109754605A (zh) * 2019-02-27 2019-05-14 中南大学 一种基于注意力时态图卷积网络的交通预测方法
CN110264709A (zh) * 2019-05-06 2019-09-20 北京交通大学 基于图卷积网络的道路的交通流量的预测方法
CN110866649A (zh) * 2019-11-19 2020-03-06 中国科学院深圳先进技术研究院 一种短期地铁客流预测方法、系统及电子设备

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
HAN YONG, WANG SHUKANG, REN YIBIN, WANG CHENG, GAO PENG, CHEN GE: "Predicting Station-Level Short-Term Passenger Flow in a Citywide Metro Network Using Spatiotemporal Graph Convolutional Neural Networks", ISPRS INTERNATIONAL JOURNAL OF GEO-INFORMATION, vol. 8, no. 6, pages 243, XP055814061, DOI: 10.3390/ijgi8060243 *
LI, ZHISHUAI ET AL.: "Short-term Traffic Flow Prediction Based on Graph Convolutional Neural Network and Attention Mechanism", JOURNAL OF TRANSPORTATION ENGINEERING, vol. 19, no. 4, 31 August 2019 (2019-08-31), pages 15 - 19,28, XP055814309 *

Cited By (26)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113537580B (zh) * 2021-06-28 2024-04-09 中科领航智能科技(苏州)有限公司 一种基于自适应图学习的公共交通客流预测方法及系统
CN113537580A (zh) * 2021-06-28 2021-10-22 中科领航智能科技(苏州)有限公司 一种基于自适应图学习的公共交通客流预测方法及系统
CN113850295A (zh) * 2021-08-27 2021-12-28 北京工业大学 一种基于Weighted-LeaderRank和GMM聚类的站点选址预测方法
CN113850295B (zh) * 2021-08-27 2024-04-02 北京工业大学 一种基于Weighted-LeaderRank和GMM聚类的站点选址预测方法
CN114154393A (zh) * 2021-10-15 2022-03-08 广州市交通规划研究院 一种基于腹地理论的目标地机场群旅客吞吐量预测方法
CN114154393B (zh) * 2021-10-15 2024-04-12 广州市交通规划研究院 一种基于腹地理论的目标地机场群旅客吞吐量预测方法
CN114117903A (zh) * 2021-11-22 2022-03-01 合肥工业大学 一种基于bp神经网络的轨道交通短时客流预测方法
CN114399726B (zh) * 2021-12-06 2023-07-07 上海市黄浦区城市运行管理中心(上海市黄浦区城市网格化综合管理中心、上海市黄浦区大数据中心) 实时智能监测客流及预警的方法和系统
CN114399726A (zh) * 2021-12-06 2022-04-26 上海市黄浦区城市运行管理中心(上海市黄浦区城市网格化综合管理中心、上海市黄浦区大数据中心) 实时智能监测客流及预警的方法和系统
CN114492951A (zh) * 2022-01-06 2022-05-13 青岛海信网络科技股份有限公司 一种基于客流预测的资源调配方法、装置及设备
CN114463978B (zh) * 2022-02-10 2024-03-29 深圳明弘电子科技有限公司 一种基于轨道交通信息处理终端的数据监测方法
CN114463978A (zh) * 2022-02-10 2022-05-10 深圳明弘电子科技有限公司 一种基于轨道交通信息处理终端的数据监测方法
CN114581846A (zh) * 2022-03-03 2022-06-03 北京城建设计发展集团股份有限公司 地铁车站全息客流实时监测方法及装置及计算机设备
CN114581846B (zh) * 2022-03-03 2024-02-20 北京城建设计发展集团股份有限公司 地铁车站全息客流实时监测方法及装置及计算机设备
CN114708013A (zh) * 2022-03-21 2022-07-05 思创数码科技股份有限公司 点击率预测方法、系统、计算机及可读存储介质
CN114819290B (zh) * 2022-04-04 2024-08-27 南京行者易智能交通科技有限公司 一种基于stacking的多环境下短时公交客流预测方法
CN114819290A (zh) * 2022-04-04 2022-07-29 南京行者易智能交通科技有限公司 一种基于stacking的多环境下短时公交客流预测方法
CN115031794A (zh) * 2022-04-29 2022-09-09 天津大学 一种多特征图卷积的新型气固两相流流量测量方法
CN114912683B (zh) * 2022-05-13 2024-05-10 中铁第六勘察设计院集团有限公司 一种智慧城市轨道交通异常大客流预测系统及方法
CN114912683A (zh) * 2022-05-13 2022-08-16 中铁第六勘察设计院集团有限公司 一种智慧城市轨道交通异常大客流预测系统及方法
CN115620525B (zh) * 2022-12-16 2023-03-10 中国民用航空总局第二研究所 基于时变动态贝叶斯网络的短时交通客运需求预测方法
CN115620525A (zh) * 2022-12-16 2023-01-17 中国民用航空总局第二研究所 基于时变动态贝叶斯网络的短时交通客运需求预测方法
CN117272848B (zh) * 2023-11-22 2024-02-02 上海随申行智慧交通科技有限公司 基于时空影响的地铁客流预测方法及模型训练方法
CN117272848A (zh) * 2023-11-22 2023-12-22 上海随申行智慧交通科技有限公司 基于时空影响的地铁客流预测方法及模型训练方法
CN117973640A (zh) * 2024-03-29 2024-05-03 煤炭科学研究总院有限公司 换乘地铁站换乘人流量预测方法及装置
CN118297419A (zh) * 2024-04-02 2024-07-05 北京交通大学 考虑事故状态的城市轨道交通短时od客流预测方法及系统

Also Published As

Publication number Publication date
CN110866649B (zh) 2022-09-20
CN110866649A (zh) 2020-03-06

Similar Documents

Publication Publication Date Title
WO2021098619A1 (zh) 一种短期地铁客流预测方法、系统及电子设备
Toqué et al. Short & long term forecasting of multimodal transport passenger flows with machine learning methods
Li et al. Shelter location and transportation planning under hurricane conditions
Gong et al. Network-wide crowd flow prediction of sydney trains via customized online non-negative matrix factorization
Niu et al. Optimizing urban rail timetable under time-dependent demand and oversaturated conditions
D’Acierno et al. Methodology for determining dwell times consistent with passenger flows in the case of metro services
Wang et al. Data driven approaches for passenger train delay estimation
Lv et al. A risk-based method for planning of bus–subway corridor evacuation under hybrid uncertainties
Liu et al. Stochastic user equilibrium based spatial-temporal distribution prediction of electric vehicle charging load
Dewil et al. A minimum cost network flow model for the maximum covering and patrol routing problem
López-Ramos Integrating network design and frequency setting in public transportation networks: a survey
Yan Optimal layout and scale of charging stations for electric vehicles
Lv et al. An origin–destination passenger flow prediction system based on convolutional neural network and passenger source-based attention mechanism
Liu et al. A MRT daily passenger flow prediction model with different combinations of influential factors
Liang et al. Bike sharing demand prediction based on knowledge sharing across modes: A graph-based deep learning approach
Wang et al. A simulation-based metro train scheduling optimization incorporating multimodal coordination and flexible routing plans
Rashidi et al. Estimating bus dwell time: A review of the literature
Su et al. Data‐Driven Method for Passenger Path Choice Inference in Congested Subway Network
Liu et al. Data analytics approach for train timetable performance measures using automatic train supervision data
Yokoyama et al. A framework for large-scale train trip record analysis and its application to passengers’ flow prediction after train accidents
Tu et al. Forecasting subway passenger flow for station-level service supply
Brar et al. Ensuring service fairness in taxi fleet management
Caropreso et al. A long-term analysis of passenger flows on a regional rail line
Carrel et al. A framework for evaluating operations control on a metro line: integrating multiple perspectives and automatically collected train and passenger movement data
Othman et al. Enhancing realism in simulation through deep learning

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20891261

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20891261

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 09.01.2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20891261

Country of ref document: EP

Kind code of ref document: A1