A large airport passenger flow spatio-temporal distribution prediction method based on STMGCN
By constructing the STMGCN framework and combining GCN, GRU, WDTW, and GLU, the problem of accuracy in predicting the spatial distribution of passengers at large airports was solved, achieving efficient spatiotemporal distribution prediction and improving the utilization efficiency of urban transportation resources.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHEAST UNIV
- Filing Date
- 2024-05-24
- Publication Date
- 2026-05-12
AI Technical Summary
Existing technologies are not very accurate in predicting the spatial distribution of passengers at large airports. Traditional methods have limited applicability and are difficult to acquire, resulting in low efficiency in the utilization of urban transportation resources.
A deep learning framework based on STMGCN is constructed, which combines graph convolutional networks (GCN) and gated recurrent units (GRU) to integrate airport passenger flow distribution matrix, influencing factor matrix and adjacency matrix. Passenger flow is corrected through weighted dynamic time warping (WDTW) and gated linear units (GLU) to achieve spatiotemporal distribution prediction.
It has improved the accuracy and precision of airport passenger flow distribution forecasting, provided a scientific basis for urban traffic planning and management, and enhanced the utilization efficiency of urban traffic resources.
Smart Images
Figure CN118446425B_ABST
Abstract
Description
Technical Field
[0001] This application relates to a method for predicting the spatiotemporal passenger flow distribution, and more particularly to a method for predicting the spatiotemporal passenger flow distribution of large airports based on STMGCN, belonging to the field of passenger flow prediction technology. Background Technology
[0002] With global economic development and accelerated urbanization, large airports, as crucial components of urban transportation hubs, are experiencing rapid growth in passenger traffic. Predicting the spatial distribution of passengers at large airports within cities is a fundamental basis for optimizing urban transportation resources. Passenger flow forecasting not only provides a basis for airport macro-planning, passenger flow control, and guidance, but also serves as an important reference for urban traffic management.
[0003] Currently, research on the spatial distribution of passengers at large airports is relatively limited and the methods used are quite traditional, falling under the traditional four-stage traffic distribution method. Due to the traditional nature of the problem and the difficulty in data acquisition, there are problems such as low accuracy in passenger flow prediction and insufficient precision in spatial distribution. For the highly complex traffic scenarios of large airports, the applicability of traditional methods is limited. Therefore, a more accurate and efficient method is needed to predict the spatial distribution of passengers at large airports to improve the utilization efficiency of urban transportation resources. Deep learning has been widely applied to passenger flow prediction in urban transportation such as subways and buses. Deep learning models, by establishing connections between multi-dimensional data and uncovering complex relationships between data, can effectively improve prediction accuracy. Summary of the Invention
[0004] The technical problem to be solved by this invention is to provide a method for predicting the spatiotemporal distribution of airport passenger flow. By considering the influencing factors of airport passenger flow distribution, a deep learning framework STMGCN based on GCN and GRU is constructed to realize the prediction of the spatiotemporal distribution of airport passenger flow, providing a scientific basis for airport and city traffic planning and management.
[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:
[0006] Airport passenger flow distribution prediction model includes the following steps:
[0007] (1) Obtain the travel origin and destination of airport passengers, obtain historical airport passenger flow distribution data, and construct a passenger flow distribution matrix; obtain data on external influencing factors of airport passenger flow distribution and construct an influencing factor matrix; obtain the adjacency relationship of each traffic zone and construct an adjacency matrix.
[0008] (2) Integrate Graph Convolutional Network (GCN) and Gated Recurrent Unit (GRU) to construct a spatiotemporal passenger flow prediction module; use the passenger flow distribution matrix and influencing factor matrix as inputs to the spatiotemporal passenger flow prediction module to obtain preliminary predicted passenger flow. .
[0009] (3) A passenger flow correction module is constructed by integrating weighted dynamic time warping (WDTW) and gated linear unit (GLU). WDTW is used to calculate the similarity between historical arrival passenger flow and passenger flow during the period to be predicted. The similarity and historical passenger flow distribution are used as inputs to GLU to obtain the corrected passenger flow. .
[0010] (4) and After weighted summation, the final prediction result is obtained. .
[0011] As a preferred embodiment of the present invention, the specific process of step (1) is as follows:
[0012] The airport passenger travel origin-destination (OD) data is obtained from mobile signaling data statistics. The airport passenger travel OD data is mapped to urban traffic zones and preprocessed with normalization to obtain urban area passenger flow distribution data. The external influencing factor data includes weather matrix, distance matrix from traffic zone to airport, and traffic zone function matrix. The adjacency matrix is an undirected adjacency matrix.
[0013] The historical airport passenger flow distribution data tensor is ( ),in, , representing the number of passengers in n traffic zones during the t-th time period;
[0014] Normalization uses min-max standardization, and its calculation formula is as follows:
[0015]
[0016] in, The original data, The data is after normalization;
[0017] The weather matrix is , representing the temporal changes in weather conditions for n traffic zones (each traffic zone has m time periods);
[0018] The distance matrix from the traffic zone to the airport is as follows: , where n is the number of traffic zones; the traffic zone function matrix is as follows: ;
[0019] The adjacency matrix is: , Indicates a traffic community cluster. Represents an edge set.
[0020] As a preferred embodiment of the present invention, the specific process of step (2) is as follows:
[0021] 2.1 Traffic Zone Feature Matrix
[0022] The historical passenger flow distribution data tensor ( Weather matrix Distance matrix Traffic Community Function Matrix The feature matrix S = ( ) is merged into a traffic zone , …… , );
[0023] 2.2 Spatial Feature Capture
[0024] The characteristic matrix S and the adjacency matrix As input to GCN, spatial features are captured to obtain the output. ;
[0025] 2.3 Temporal Feature Capture
[0026] Will As input to GRU, preliminary passenger flow estimates are obtained. :
[0027]
[0028]
[0029]
[0030]
[0031] The hidden state used to determine the current time step should be based on historical passenger flow data. and current information The degree of updating; This determines how much information from the previous time step should be ignored in the hidden state of the current time step; after obtaining information from the update and reset gates, historical passenger flows are merged. and current input Calculate the candidate hidden state Finally, combined with the updated door and candidate hidden state This gives us the hidden state at the current time step. ( ).
[0032] As a preferred embodiment of the present invention, the specific process of step (3) is as follows:
[0033] Using WDTW to calculate historical arrival passenger flow and the predicted arrival time of passenger flow The similarity; the cost matrix represents the similarity, and the cost matrix is:
[0034] , It is a time period and The weight value;
[0035] The cost matrix is used as the input parameter matrix for GLU to obtain the passenger flow correction result:
[0036]
[0037] in, For historical visitor flow, , , , Let be the parameter matrix of the gated linear unit. This is the activation function.
[0038] As a preferred embodiment of the present invention, the specific process of step (4) is as follows:
[0039] Output of the spatiotemporal prediction module and the output of the schedule correction module After weighting and Weighted summation, combined to form the final output :
[0040]
[0041] in, , These are the weighting coefficients. It is the sigmoid activation function.
[0042] The beneficial effects of this invention are: by incorporating factors influencing airport passenger flow, it can capture the spatiotemporal characteristics of passenger flow distribution and achieve accurate prediction of airport passenger flow distribution. Attached Figure Description
[0043] Figure 1 This is a flowchart of the airport passenger flow distribution prediction model of the present invention;
[0044] Figure 2 This is a schematic diagram of the actual model structure of the present invention;
[0045] Figure 3 This is a distribution diagram of the prediction result error in an embodiment of the present invention, wherein the left figure is the distribution of the mean absolute error (MAE) and the right figure is the distribution of the root mean square error (RMSE). Detailed Implementation
[0046] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0047] like Figure 1 As shown, this invention provides a method for predicting the spatiotemporal distribution of passenger flow at large airports based on STMGCN, which mainly includes the following steps:
[0048] Step (1): Obtain the travel origin-destination (OD) data of airport passengers, obtain historical airport passenger flow distribution data, and construct a passenger flow distribution matrix; obtain data on external influencing factors of airport passenger flow distribution and construct an influencing factor matrix; obtain the adjacency relationship of each traffic zone and construct an adjacency matrix.
[0049] Step (2): Integrate Graph Convolutional Network (GCN) and Gated Recurrent Unit (GRU) to construct a spatiotemporal passenger flow prediction module; use the passenger flow distribution matrix and influencing factor matrix as inputs to the spatiotemporal passenger flow prediction module to obtain preliminary predicted passenger flow. .
[0050] Step (3): Integrate Weighted Dynamic Time Warping (WDTW) and Gated Linear Unit (GLU) to construct a passenger flow correction module. Use WDTW to calculate the similarity between historical arrival passenger flow and passenger flow during the period to be predicted. Use the similarity and historical passenger flow distribution as input to GLU to obtain the corrected passenger flow. .
[0051] Step (4), and After weighted summation, the final prediction result is obtained. .
[0052] The technical solution of the present invention will be further illustrated below with specific embodiments. The passenger flow at Beijing Daxing International Airport in China is taken as the research object. The research area covers 337 traffic zones in the main urban area of Beijing.
[0053] 1. Obtain the origin-destination (OD) data of airport passengers to obtain historical airport passenger flow distribution data and construct a passenger flow distribution matrix; obtain data on external influencing factors of airport passenger flow distribution and construct an influencing factor matrix; obtain the adjacency relationships of each traffic zone and construct an adjacency matrix.
[0054] The data includes the spatiotemporal distribution of airport passenger flow in October 2023, with a time granularity of 1 hour, and a passenger flow matrix. (31*24, 337); Weather conditions for various traffic areas in October 2023, weather matrix. (31*24, 337); Functions of each traffic zone (represented by the main point of interest type), function matrix (1, 337); Distances from each traffic zone to the airport, distance matrix (1, 337); Adjacency relationships of each traffic zone, adjacency matrix (337, 337).
[0055] Table 1 Example of historical passenger flow distribution data
[0056]
[0057] 2. Integrate Graph Convolutional Network (GCN) and Gated Recurrent Unit (GRU) to construct a spatiotemporal passenger flow prediction module (e.g., Figure 2 As shown in the figure), the passenger flow distribution data and the influencing factor matrix are used as inputs to the spatiotemporal passenger flow prediction module to obtain the preliminary predicted passenger flow y1.
[0058] 2.1 Feature Matrix Construction
[0059] Passenger Flow Matrix Weather Matrix Distance matrix Traffic Community Function Matrix The feature matrix S = ( ) is merged into a traffic zone , , );
[0060] 2.2 Spatial Feature Capture
[0061] The characteristic matrix S and the adjacency matrix As input to GCN, the batch size of the input data is 32, the time series length is 6, and the learning rate is set to 0.001, resulting in the output. ;
[0062] 2.3 Temporal Feature Capture
[0063] Will As input to GRU, preliminary passenger flow estimates are obtained. :
[0064]
[0065]
[0066]
[0067]
[0068] The hidden state used to determine the current time step should be based on historical passenger flow data. and current information The degree of updating; This determines how much information from the previous time step should be ignored in the hidden state of the current time step; after obtaining information from the update and reset gates, historical passenger flows are merged. and current input Calculate the candidate hidden state Finally, combined with the updated door and candidate hidden state This gives us the hidden state at the current time step. ( ).
[0069] 3. Integrate Weighted Dynamic Time Warping (WTTW) and Gated Linear Units (GLU) to construct a passenger flow correction module (e.g., Figure 2 As shown in the figure, the similarity between historical arrival passenger flow and passenger flow during the period to be predicted is calculated using WDTW. The similarity and historical passenger flow distribution are used as inputs to GLU to obtain the corrected passenger flow y2.
[0070] 4. Utilize WDTW to calculate historical arrival passenger flow at the airport. and the predicted arrival time of passenger flow The similarity; the cost matrix represents the similarity, and the cost matrix is:
[0071] , It is a time period and The weight value;
[0072] The cost matrix is used as the input parameter matrix for GLU to obtain the passenger flow correction result:
[0073]
[0074] in, For historical visitor flow, , , , Let be the parameter matrix of the gated linear unit. This is the activation function.
[0075] (4) Output of the spatiotemporal prediction module and the output of the schedule correction module After weighting and Weighted summation, combined to form the final output :
[0076]
[0077] in, , These are the weighting coefficients. It is the sigmoid activation function.
[0078] In this embodiment, to verify the effectiveness of the model, ARIMA, GBDT, Transformer, and LSTM are selected as comparison models. The mean absolute error (MAE) and root mean square error (RMSE) of the prediction results are set as accuracy evaluation metrics (e.g., ...). Figure 3 (As shown), the formula is as follows:
[0079]
[0080]
[0081] in, and It is the first Actual and predicted passenger flow at any given time. This represents the sample size of the passenger flow. The smaller the MAE and RMSE, the smaller the error and the better the prediction performance. The prediction results of each model are shown in the table below:
[0082]
[0083] The prediction results show that ARIMA, GBDT, Transformer, and LSTM are all inferior to the STMGCN model proposed in this invention, indicating that the model proposed in this invention can capture the spatiotemporal characteristics of airport passenger flow better and has a better prediction effect.
[0084] This invention proposes a method for predicting the spatiotemporal distribution of passenger flow at large airports based on STMGCN. Through multi-source data analysis, it incorporates external factors such as weather, urban functions, urban regional connectivity, and distance to the airport. It also considers the impact of airport arrivals on passenger flow distribution within the city, resulting in a relatively accurate prediction of airport passenger flow distribution. The prediction results can provide scientific guidance for airport and urban transportation operations and management.
Claims
1. A method for predicting the spatiotemporal distribution of passenger flow at large airports based on STMGCN, characterized in that, The method includes the following steps: (1) Obtain the origin-destination (OD) data of airport passengers to obtain historical airport passenger flow distribution data and construct a passenger flow matrix; obtain data on external influencing factors of airport passenger flow distribution and construct an influencing factor matrix, including a weather matrix. Distance matrix Traffic Community Function Matrix Obtain the adjacency relationships of each traffic zone and construct an adjacency matrix; (2) Integrating Graph Convolutional Network (GCN) and Gated Recursive Unit (GRU) to construct a spatiotemporal passenger flow prediction module; and using the passenger flow matrix Weather Matrix Distance matrix Traffic Community Function Matrix The feature matrix S of the traffic zone is merged with the adjacency matrix. As input to GCN, the output is obtained. ,Will As input to GRU, preliminary passenger flow estimates are obtained. ; (3) A passenger flow correction module is constructed by integrating weighted dynamic time warping (WDTW) and gated linear unit (GLU). The similarity between historical arrival passenger flow and passenger flow during the period to be predicted is calculated using WDTW. The similarity and historical passenger flow distribution are used as inputs to GLU to obtain the corrected passenger flow. ; (4) and After weighted summation, the final prediction result is obtained. .
2. The method for predicting the spatiotemporal distribution of passenger flow in large airports based on STMGCN according to claim 1, characterized in that, The specific process of step (1) is as follows: The airport passenger travel origin-destination (OD) data is obtained from mobile signaling data statistics. The airport passenger travel OD data is mapped to urban traffic zones and preprocessed with normalization to obtain urban area passenger flow distribution data. The external influencing factor data includes weather matrix, distance matrix from traffic zone to airport, and traffic zone function matrix. The adjacency matrix is an undirected adjacency matrix.
3. The method for predicting the spatiotemporal distribution of passenger flow in large airports based on STMGCN according to claim 2, characterized in that, The historical airport passenger flow distribution data tensor is: ,in, , representing the Within a time period The number of passengers in each traffic zone; the weather matrix is... ,represent Traffic Community The temporal changes in weather conditions over a given period; the distance matrix from the traffic zone to the airport is... , The number of traffic zones; the functional matrix of the traffic zones is as follows: The adjacency matrix is: , Indicates a traffic community cluster. Represents an edge set.
4. The method for predicting the spatiotemporal distribution of passenger flow in large airports based on STMGCN according to claim 2, characterized in that, The specific process of step (2) is as follows: (2.1) Traffic zone feature matrix The historical passenger flow distribution data tensor ( Weather matrix Distance matrix Traffic Community Function Matrix The feature matrix S = ( ) is merged into a traffic zone , …… , ); (2.2) Spatial feature capture The characteristic matrix S and the adjacency matrix As input to GCN, spatial features are captured to obtain the output. ; (2.3) Temporal Feature Capture Will As input to GRU, preliminary passenger flow estimates are obtained. : The hidden state used to determine the current time step should be based on historical passenger flow data. and current input The degree of updating; This determines how much information from the previous time step should be ignored in the hidden state of the current time step; after obtaining information from the update and reset gates, historical passenger flows are merged. and current input Calculate the candidate hidden state Finally, combined with the updated door and candidate hidden state Get the hidden state at the current time step. ,Right now .
5. The method for predicting the spatiotemporal distribution of passenger flow in large airports based on STMGCN according to claim 1, characterized in that, The specific process of step (3) is as follows: Using WDTW to calculate historical arrival passenger flow Passenger flow during the predicted time period The similarity; the cost matrix represents the similarity, and the cost matrix is: , yes and The weight value; The cost matrix is used as the input parameter matrix for GLU to obtain the passenger flow correction result: in, Based on historical passenger flow distribution, , , , Let be the parameter matrix of the gated linear unit. This is the activation function.
6. The method for predicting the spatiotemporal distribution of passenger flow in large airports based on STMGCN according to claim 1, characterized in that, The specific process of step (4) is as follows: Output of the spatiotemporal passenger flow prediction module and the output of the passenger flow correction module After weighting and Weighted summation, combined to form the final output : in, , These are the weighting coefficients. It is the sigmoid activation function.