A public transport integration and evacuation method based on multi-source data fusion

Abstract

The application discloses a public transport integration evacuation method based on multi-source data fusion, relates to the technical field of public transport integration evacuation, and comprises the following steps: acquiring multi-source traffic operation data, passenger flow distribution data and traffic state data, and performing time alignment and space mapping to construct a standardized traffic data set; passenger flow density, transport capacity occupation and traffic efficiency indexes are extracted to form multi-dimensional traffic state characteristics; a traffic correlation graph is constructed, node correlation weights are calculated, neighborhood characteristics are aggregated in combination with space distance and traffic time, and fused traffic characteristics are generated; a passenger flow demand and transport capacity supply matching relationship is constructed based on the fused traffic characteristics, an evacuation state characteristic is formed, a scheduling strategy and a connection scheduling scheme are generated; after scheduling execution, a historical evacuation scheduling case library is established, scheduling strategies are optimized through similar case retrieval and review suggestions, the historical evacuation scheduling case library is updated, and continuous optimization of the scheduling strategy is realized.

Description

Technical Field

[0001] This invention relates to the field of integrated public transportation transportation technology, specifically to an integrated public transportation transportation method based on multi-source data fusion. Background Technology

[0002] With the continuous expansion of public transportation systems and the highly dynamic spatiotemporal distribution of passenger flow, traditional dispatching methods based on static rules or single indicators are insufficient to effectively characterize the dynamic evolution of passenger flow in the spatiotemporal dimension, and also struggle to achieve refined matching between transport capacity resources and passenger demand. This results in insufficient adaptability of dispatching strategies and limited overall transport efficiency. As the ability to collect multi-source traffic data improves, analytical methods based on multi-source data fusion and network modeling are gradually being applied to the field of public transportation dispatching. By integrating passenger flow distribution data, operational status data, and road network structure information, the ability to express the state of the transportation system can be improved to some extent. However, existing methods still have shortcomings in spatiotemporal alignment and fusion of multi-source heterogeneous data, characterization of topological relationships between stations, and dynamic weight modeling, making it difficult to achieve a unified expression and collaborative analysis of passenger flow transfer relationships and road segment connectivity relationships.

[0003] However, existing public transportation scheduling methods typically lack a hierarchical characterization mechanism for supply and demand relationships, time fluctuation characteristics, and inter-station coupling relationships. They fail to provide a unified analysis of the differences between passenger demand and transport capacity supply, as well as the network-based propagation characteristics between stations. This results in insufficient ability to express complex traffic conditions, low precision in scheduling decisions, and a lack of effective mechanisms for accumulating and reusing historical scheduling experience during execution. It is difficult to form a structured case library for passenger transport scheduling, and there is a lack of systematic review and analysis of scheduling results. This prevents continuous optimization of scheduling strategies based on historical experience, leading to insufficient optimization capabilities and reduced overall passenger transport efficiency. Therefore, a public transportation integrated passenger transport method and system based on multi-source data fusion is needed to address the aforementioned problems. Summary of the Invention

[0004] To address the aforementioned technical problems, this paper provides an integrated public transportation dispatching method based on multi-source data fusion. This technical solution solves the problem that existing public transportation dispatching methods, as mentioned in the background, typically lack a hierarchical characterization mechanism for supply and demand relationships, time fluctuation characteristics, and inter-station coupling relationships. They are unable to uniformly analyze the differences between passenger flow demand and transport capacity supply, as well as the correlation propagation characteristics between stations based on the road network structure. This results in insufficient ability to express complex traffic conditions, low precision in dispatching decisions, and a lack of effective accumulation and reuse mechanisms for historical dispatching experience during dispatching execution. It is difficult to form a structured dispatching case library, and there is also a lack of systematic review and analysis of dispatching execution results. It is impossible to continuously optimize dispatching strategies based on historical experience, thus causing insufficient dispatching strategy optimization capabilities and reduced overall dispatching efficiency.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A public transportation integrated distribution method based on multi-source data fusion includes: Public transportation data within the target area is acquired and time-aligned and spatially mapped to obtain a standardized traffic dataset. The public transportation data includes public transportation operation data, passenger flow distribution data, and traffic status data. Based on a standardized traffic dataset, passenger flow density, capacity occupancy, and traffic efficiency indicators are extracted and fused to obtain a multi-dimensional traffic state feature set. Based on a multidimensional traffic state feature set, a traffic association graph structure is constructed, which maps traffic stations as nodes, passenger flow transfer relationships and road segment connection relationships as edges, and encodes passenger flow density index, capacity occupancy index and traffic efficiency index as node features to obtain traffic graph feature data. A graph neural network model is constructed based on traffic map feature data to obtain the association weights between nodes. The association weights are modulated by combining the spatial distance and travel time corresponding to the edges to obtain a weighted association matrix. The neighborhood features of the nodes are then weighted and aggregated to generate a fused traffic feature vector. Based on the fusion of traffic feature vectors, a matching relationship between passenger flow demand intensity and transport capacity supply is constructed to obtain the supply-demand deviation value, time fluctuation deviation value and connection deviation value of each transportation station, and combine them to form a hierarchical coupled transport status feature. A set of scheduling strategies is generated based on the characteristics of the transportation status, and a point-to-point connection scheduling scheme is obtained according to each scheduling strategy. The scheduling execution instruction is output to drive the scheduling execution. Then, historical public transportation scheduling records are obtained, a historical transportation scheduling case library is established, the scheduling strategies are optimized, the optimized scheduling scheme is obtained, and the historical transportation scheduling case library is updated.

[0006] In an optional embodiment, the step of extracting passenger flow density, capacity occupancy, and traffic efficiency indicators based on a standardized traffic dataset and fusing them to obtain a multi-dimensional traffic state feature set specifically includes: Data analysis is performed on public transportation operation data, passenger flow distribution data, and traffic status data to obtain a data set containing time stamps and spatial location information. The traffic status data includes information related to road traffic status, travel time, and congestion status. Based on the time stamps in the dataset, the public transportation operation data, passenger flow distribution data, and traffic status data are resampled according to a unified time granularity to obtain a time-aligned dataset. Based on the spatial location information in the time-aligned data set, public transportation operation data, passenger flow distribution data, and traffic status data are mapped to a unified spatial coordinate system, and spatial affiliation is divided according to transportation stations to obtain a spatial mapping data set; The spatial mapping dataset is aggregated according to transportation stations to obtain a standardized transportation dataset. Based on the passenger flow distribution data of each transportation station in the standardized traffic dataset, the average passenger flow per unit time is extracted to obtain the passenger flow density index. Based on the public transportation operation data of each transportation station in the standardized transportation dataset, the ratio of actual carrying capacity to transport capacity supply is determined to obtain the transport capacity occupancy index. Based on the centralized traffic status data in the standardized traffic dataset, the ratio of travel time to reference travel time is determined to obtain the traffic efficiency index. By applying a unified scale mapping to passenger flow density, capacity occupancy, and traffic efficiency indicators, a standardized set of indicators is obtained. An outlier identification and removal process is performed based on a standardized set of indicators to obtain a set of denoising indicators. The denoising index set is mapped according to the corresponding relationship of traffic stations to generate a station-level multidimensional state feature set, which is then spliced ​​and fused to obtain a multidimensional traffic state feature set.

[0007] In an optional embodiment, the construction of a traffic association graph structure based on a multi-dimensional traffic state feature set maps traffic stations as nodes, passenger flow transfer relationships and road segment connection relationships as edges, and encodes passenger flow density indicators, capacity occupancy indicators, and traffic efficiency indicators as node features to obtain traffic graph feature data, specifically including: Based on a multidimensional traffic state feature set, traffic station information in the feature set is obtained to determine the station set; The station set is subjected to structural parsing to extract the spatial location information and station identification information of each transportation station; Node identifiers are generated based on the uniqueness rules of station identifiers, and node mapping is completed after determining the spatial attribute association relationship of nodes in combination with spatial location information, thus mapping each transportation station to a node unit in the graph structure. Based on passenger flow distribution data, the passenger flow relationship between stations is analyzed to obtain the passenger flow transfer relationship between stations; Based on public transportation operation data, the route topology information is extracted and associated with stations to form a set of road segment connection relationships between stations; The consistency matching process between passenger flow transfer relationships and road segment connection relationships is performed to generate the edge relationship structure between nodes; Passenger flow density indicators are extracted from multidimensional traffic state features and mapped to each node to form the first type of node attributes. Capacity occupancy indicators are extracted from the multidimensional traffic state feature set and mapped to each node to form the second type of node attributes. At the same time, traffic efficiency indicators are extracted and mapped to each node to form the third type of node attributes. The three types of node attributes are uniformly encoded to form node feature vector representations, which are then fused and assembled with the edge relationship structure to obtain traffic map feature data.

[0008] In an optional embodiment, the step of constructing a graph neural network model based on traffic map feature data to obtain the association weights between nodes, modulating the association weights by combining the spatial distance and travel time corresponding to the edges to obtain a weighted association matrix, and weighted aggregation of the node neighborhood features to generate a fused traffic feature vector, specifically includes: A graph neural network model is constructed based on traffic map feature data; Based on the edge relationship structure in the graph neural network model, identify the nodes that have a connection relationship and form a set of node pairs; For any node pair in the set of node pairs, extract the node feature vectors of the two corresponding nodes and obtain the difference vector between the two node feature vectors; Based on the node feature vectors, the covariance relationship between each feature dimension is statistically obtained, and the node feature covariance matrix is ​​constructed. Based on the difference vector and the node feature covariance matrix, the difference vector is standardized by distance measurement to obtain the feature difference values ​​between nodes; Based on the feature differences between nodes, basic association weights are generated using an exponential decay function. Obtain the spatial distance data and travel time data corresponding to the node pairs in the edge relationship structure, and perform normalization processing to obtain the distance coefficient and time coefficient; Based on the distance coefficient, the basic association weights are spatially corrected to obtain the distance-corrected association weights; The distance-corrected correlation weights are modulated by applying a time coefficient to obtain modulated correlation weights. The modulation association weights are normalized so that the sum of all association weights corresponding to the same node is one, thus obtaining the normalized association weights. Based on the correspondence between node pairs in the node pair set, the normalized association weights are arranged into a matrix to obtain the node association matrix. The neighboring nodes of each node are determined based on the node association matrix, thus obtaining the set of neighboring nodes; Based on the set of adjacent nodes, the node feature vectors of the corresponding nodes are extracted to form a node neighborhood feature set; Using the normalized association weights in the node association matrix as coefficients, the node neighborhood feature set is weighted and summed to obtain the node neighborhood aggregated feature vector. The neighborhood aggregation feature vector of a node is concatenated with the node feature vector of the corresponding node to generate a fused traffic feature vector. The formula for calculating the feature difference value is as follows: In the formula, For nodes With nodes The characteristic difference values ​​between them Represents a node The corresponding node feature vector, Represents a node The corresponding node feature vector, The node feature covariance matrix, This is a vector transpose operation; The formula for calculating the basic association weight is as follows: In the formula, Based on the association weight, This is the preset attenuation coefficient.

[0009] In an optional embodiment, the step of constructing a matching relationship between passenger demand intensity and transport capacity based on fused traffic feature vectors, obtaining supply-demand deviation values, time fluctuation deviation values, and connection deviation values ​​for each transportation station, and combining them to form a hierarchically coupled transport status feature, specifically includes: The integrated traffic feature vector is obtained and its features are analyzed to obtain the comprehensive passenger flow density feature component, comprehensive transport capacity occupancy feature component, and comprehensive traffic efficiency feature component corresponding to each traffic station. The comprehensive passenger flow density feature component is used as the passenger flow demand intensity of each transportation station, and a passenger flow demand intensity set is formed. The comprehensive passenger flow density feature component is a comprehensive representation result formed by superimposing the passenger flow density feature of the node itself with the aggregated passenger flow density feature of the neighborhood. The standard capacity of each transportation station is obtained, and the capacity supply capacity of each transportation station is determined by combining the comprehensive capacity occupancy feature component, and a capacity supply capacity set is formed. The comprehensive capacity occupancy feature component is a comprehensive representation result formed by averaging the capacity occupancy feature of the node itself and the aggregated capacity occupancy feature of the neighborhood. The difference between passenger demand intensity and transport capacity is calculated to obtain the supply-demand deviation value of each transportation station. The supply-demand deviation value is the difference between passenger demand intensity and transport capacity. Based on the comprehensive traffic efficiency feature component in the fused traffic feature vector, the change sequence of each traffic station within a continuous time window is extracted, and the change difference between adjacent time windows is obtained as the time fluctuation deviation value. The comprehensive traffic efficiency feature component is a comprehensive representation result formed by selecting the minimum value of the node's own traffic efficiency feature and the neighborhood aggregated traffic efficiency feature. Based on the edge relationship structure in the traffic map feature data, the connection relationship between adjacent traffic stations is extracted, and the supply and demand deviation value of the corresponding station is obtained. Based on the supply-demand deviation values ​​corresponding to the connected transportation stations, the absolute difference in supply-demand deviation values ​​between adjacent stations is calculated to obtain the connection deviation value. The connection deviation values ​​between stations are weighted and aggregated according to the connection relationship between adjacent stations to obtain the connection deviation value. The supply and demand deviation, time fluctuation deviation, and connection deviation are combined and processed to form a layered and coupled transportation status characteristic. The formula for calculating the transport capacity supply is as follows: In the formula, For the capacity supply of the a-th transportation station, Let be the characteristic component of the comprehensive transport capacity occupancy of the a-th transportation station. Let be the standard transport capacity of the a-th transportation station.

[0010] In an optional embodiment, the process of generating a set of scheduling strategies based on the characteristics of the transportation status, obtaining a point-to-point connection scheduling scheme according to each scheduling strategy, outputting scheduling execution instructions to drive scheduling execution, then acquiring historical public transportation scheduling records, establishing a historical transportation scheduling case library, optimizing the scheduling strategies to obtain an optimized scheduling scheme, and updating the historical transportation scheduling case library, specifically includes: Obtain the characteristics of the transportation status and extract the supply and demand deviation, time fluctuation deviation, and connection deviation values. After normalizing the supply-demand deviation, time fluctuation deviation, and connection deviation, the values ​​are weighted and summed according to preset weight coefficients to obtain the evacuation priority index of each transportation station. The transportation stations are then ranked according to the size of the evacuation priority index to form a station priority sequence. Based on the station priority sequence, high-priority transportation stations are extracted, and the corresponding adjacent transportation stations are determined by combining the edge relationship structure, forming a candidate evacuation station set; Using the site priority sequence as a constraint, candidate evacuation sites are sequentially arranged to generate multiple candidate site service orders and form a combination of site access order combinations. For each station access sequence combination, the corresponding capacity allocation ratio is determined by combining the supply and demand deviation value of each transportation station with the standard capacity capacity, thereby forming a set of scheduling strategies that include the station service sequence and the corresponding capacity allocation ratio. Based on the station service order and capacity allocation ratio in each scheduling strategy, a point-to-point shuttle scheduling scheme is obtained; The point-to-point connection scheduling scheme is converted into scheduling execution instructions and output in chronological order to drive scheduling execution; Historical public transportation scheduling records are obtained, and the scheduling process under different time periods and different traffic operation states is sorted and classified. The corresponding scheduling scenarios, traffic station status and historical scheduling strategies are extracted to obtain historical scheduling basic data. The execution results of each historical scheduling strategy are analyzed to generate historical review suggestions, which are then structured to obtain review suggestion information. These review suggestions are opinions on adjusting scheduling strategies formed after analyzing the results of scheduling execution, and are used to guide the optimization of subsequent scheduling strategies. Historical dispatching data and review suggestions are linked and integrated to form a historical evacuation dispatching case library; After the scheduling is completed, obtain the scheduling scenario and traffic station status corresponding to the current scheduling task, search for similar cases in the historical evacuation scheduling case library, and obtain matching historical scheduling strategies and historical review suggestions. Based on historical scheduling strategies and historical review suggestions, the current scheduling scheme is adjusted and optimized to obtain the optimized scheduling scheme. The optimized scheduling scheme is executed, and the results are analyzed to generate post-optimization review suggestions. The current scheduling scenarios, traffic station status, optimized scheduling schemes, and corresponding post-optimization review suggestions are structured and organized, and the historical evacuation scheduling case library is updated to form an updated evacuation scheduling case library.

[0011] Furthermore, a public transportation integrated distribution system based on multi-source data fusion is proposed to implement any of the above-mentioned distribution methods, including: The data acquisition and processing module is used to acquire public transportation operation data, passenger flow distribution data and traffic status data within the target area, and to perform time alignment and spatial mapping processing on the data to generate a standardized traffic dataset. The feature extraction module is used to extract passenger flow density indicators, capacity occupancy indicators and traffic efficiency indicators from a standardized traffic dataset, and perform unified scale mapping and noise reduction processing to form a multi-dimensional traffic state feature set. The graph construction module is used to construct a traffic association graph structure based on a multi-dimensional traffic state feature set, map traffic stations as nodes, map passenger flow transfer relationships and road segment connection relationships as edges, and generate traffic graph feature data. The graph neural network processing module is used to construct a graph neural network model based on traffic map feature data, calculate the association weights between nodes, modulate the association weights by combining spatial distance and travel time, generate a weighted association matrix, and perform weighted aggregation of the node neighborhood features to generate a fused traffic feature vector. The status analysis module is used to construct a matching relationship between passenger flow demand intensity and transport capacity supply based on the fused traffic feature vector, generate supply and demand deviation values, time fluctuation deviation values ​​and connection deviation values, and form evacuation status characteristics. The scheduling and case optimization module is used to generate a set of scheduling strategies and a point-to-point connection scheduling scheme based on the characteristics of the transportation status, and to obtain historical public transportation scheduling records to establish a historical transportation scheduling case library. Similar cases are searched in the historical transportation scheduling case library, the scheduling strategies are optimized, and the historical transportation scheduling case library is updated.

[0012] In an optional embodiment, the data acquisition and processing module includes: The data acquisition unit is used to acquire public transportation route operation status data and station capacity carrying capacity data. Passenger flow collection unit, which is used to acquire passenger flow distribution data and passenger flow time distribution data at each transportation station; A status acquisition unit is used to acquire traffic status data related to the road network traffic status. A time alignment unit is used to perform unified time-granularity resampling processing on multi-source data; A spatial mapping unit is used to map data to a unified spatial coordinate system and divide it according to station affiliation to generate a standardized traffic dataset.

[0013] In an optional embodiment, the graph construction module includes: A node mapping unit, wherein the node mapping unit is used to map traffic stations as graph structure nodes; An edge construction unit is used to parse the passenger flow transfer relationship and road segment connection relationship between stations; A relationship fusion unit is used to match and fuse passenger flow transfer relationships with road segment connection relationships to generate an edge relationship structure; An attribute encoding unit is used to encode passenger flow density indicators, capacity occupancy indicators and traffic efficiency indicators into node feature vectors. The graph generation unit is used to fuse node feature vectors with edge relationship structures to generate traffic map feature data.

[0014] In an optional embodiment, the scheduling and case optimization module includes: A strategy generation unit is used to generate a set of candidate evacuation sites and a combination of site access order based on a site priority sequence. A scheduling generation unit is used to generate a point-to-point connection scheduling scheme based on the combination of site access order and capacity allocation ratio. Case construction unit, which is used to acquire historical public transportation scheduling records and generate a historical evacuation scheduling case library; A case retrieval unit, which is used to retrieve similar cases from a historical evacuation and dispatch case database; A strategy optimization unit is used to optimize the scheduling strategy based on historical scheduling strategies and historical review suggestions. The case update unit is used to analyze the results of the optimized scheduling scheme, generate review suggestions, and update the historical evacuation scheduling case library.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This proposal presents an integrated public transportation distribution method based on multi-source data fusion. It achieves data standardization and unified expression by performing time alignment and spatial mapping on multi-source traffic data, improving the consistency and utilization efficiency of multi-source data fusion. By constructing multi-dimensional traffic state features and introducing a traffic association graph structure, it enhances the unified characterization of passenger flow transfer relationships between stations and road network topology. By calculating node association weights and modulating them in conjunction with spatial distance and travel time, it improves the accuracy of traffic association modeling. By constructing supply-demand deviation, time fluctuation deviation, and connection deviation, it achieves a hierarchical expression of the relationship between passenger flow demand and transport capacity supply. By establishing a historical distribution scheduling case library and introducing review suggestions, it optimizes scheduling strategies and updates the case library, thereby improving the adaptability of scheduling strategies and overall distribution efficiency. Attached Figure Description

[0016] Figure 1 This is a flowchart of a public transportation integrated evacuation method based on multi-source data fusion proposed in this invention; Figure 2 A flowchart illustrating the construction of multidimensional traffic state features in this invention; Figure 3 This is a flowchart of the evacuation status and scheduling optimization process in this invention; Figure 4 This is a system framework diagram of a remote automated operation and maintenance system for a big data storage server proposed in this invention. Detailed Implementation

[0017] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0018] Reference Figure 1 - Figure 4 As shown in the figure, an integrated public transportation evacuation method based on multi-source data fusion in an embodiment of the present invention includes: Public transportation data within the target area is acquired and time-aligned and spatially mapped to obtain a standardized traffic dataset. The public transportation data includes public transportation operation data, passenger flow distribution data, and traffic status data. Based on a standardized traffic dataset, passenger flow density, capacity occupancy, and traffic efficiency indicators are extracted and fused to obtain a multi-dimensional traffic state feature set. Based on a multidimensional traffic state feature set, a traffic association graph structure is constructed, which maps traffic stations as nodes, passenger flow transfer relationships and road segment connection relationships as edges, and encodes passenger flow density index, capacity occupancy index and traffic efficiency index as node features to obtain traffic graph feature data. A graph neural network model is constructed based on traffic map feature data to obtain the association weights between nodes. The association weights are modulated by combining the spatial distance and travel time corresponding to the edges to obtain a weighted association matrix. The neighborhood features of the nodes are then weighted and aggregated to generate a fused traffic feature vector. Based on the fusion of traffic feature vectors, a matching relationship between passenger flow demand intensity and transport capacity supply is constructed to obtain the supply-demand deviation value, time fluctuation deviation value and connection deviation value of each transportation station, and combine them to form a hierarchical coupled transport status feature. A set of scheduling strategies is generated based on the characteristics of the transportation status, and a point-to-point connection scheduling scheme is obtained according to each scheduling strategy. The scheduling execution instruction is output to drive the scheduling execution. Then, historical public transportation scheduling records are obtained, a historical transportation scheduling case library is established, the scheduling strategies are optimized, the optimized scheduling scheme is obtained, and the historical transportation scheduling case library is updated.

[0019] Furthermore, based on standardized traffic datasets, passenger flow density, capacity utilization, and traffic efficiency indicators are extracted and fused to obtain a multi-dimensional traffic state feature set, specifically including: Data analysis is performed on public transportation operation data, passenger flow distribution data, and traffic status data to obtain a data set containing time stamps and spatial location information. Traffic status data includes information related to road traffic status, travel time, and congestion status. Based on the time stamps in the dataset, the public transportation operation data, passenger flow distribution data, and traffic status data are resampled according to a unified time granularity to obtain a time-aligned dataset. Based on the spatial location information in the time-aligned data set, public transportation operation data, passenger flow distribution data, and traffic status data are mapped to a unified spatial coordinate system, and spatial affiliation is divided according to transportation stations to obtain a spatial mapping data set; The spatial mapping dataset is aggregated according to transportation stations to obtain a standardized transportation dataset. Specifically, when analyzing public transportation operation data, passenger flow distribution data, and traffic status data, the first step is to perform structured field identification and unified parsing processing on the multi-source heterogeneous data. This extracts the time stamp and spatial location information corresponding to each data point. The time stamp represents the point in time or time interval in which the data was generated, while the spatial location information represents the geographical location or transportation station relationship corresponding to the data. Public transportation operation data generally includes vehicle arrival and departure times, route trajectories, and actual passenger load information. Passenger flow distribution data generally includes passenger flow records at various stations during different time periods. Traffic status data generally includes information related to road traffic conditions, travel time, and congestion status. Through the above parsing processing, the original heterogeneous data is transformed into a unified intermediate representation, possessing both time and spatial dimensions, thus forming a dataset containing time stamps and spatial location information. This provides the input basis for subsequent unified spatiotemporal benchmark processing. Subsequently, based on the time identifier in the dataset, various types of data are resampled according to a preset uniform time granularity. The uniform time granularity is used to map data with different sampling frequencies to the same time window scale, making them comparable under the same time benchmark. For high-frequency sampled data, statistical aggregation within the time window is used for downsampling, and for low-frequency or irregular sampled data, interpolation is used for time alignment, thereby eliminating the differences in time resolution between different data sources and obtaining a time-aligned dataset.

[0020] After time alignment, based on the spatial location information in the time-aligned dataset, various data types are uniformly mapped to a unified spatial coordinate system. This unified spatial coordinate system provides a consistent spatial reference benchmark, enabling the spatial locations of data from different sources to be expressed and compared within the same coordinate system. The spatial mapping process is achieved by establishing a mapping relationship between the original spatial coordinates and standard transportation spatial coordinates, thereby uniformly converting public transportation operation data, passenger flow distribution data, and traffic status data into a standard spatial representation. Subsequently, spatial division is performed based on the attribution relationship between spatial location information and transportation stations. When a data point falls within the service area of ​​a transportation station or the area of ​​its nearest neighbor, it is assigned to the corresponding station unit, thus forming a data allocation structure with stations as the basic unit, resulting in a spatially mapped dataset. Based on this, the spatial mapping dataset is aggregated according to the dimension of transportation stations. This involves summarizing and statistically processing multi-source data of the same station within the same time window. For example, passenger flow data is summed or averaged, operational status data is comprehensively statistically expressed, and auxiliary traffic data is averaged. This reduces the randomness caused by single-point fluctuations, making the data expression at the station level more stable and consistent. Finally, a standardized traffic dataset is obtained, providing a unified input basis for subsequent multi-dimensional traffic status feature extraction and graph structure construction.

[0021] Based on the passenger flow distribution data of each transportation station in the standardized traffic dataset, the average passenger flow per unit time is extracted to obtain the passenger flow density index. Based on the public transportation operation data of each transportation station in the standardized transportation dataset, the ratio of actual carrying capacity to transport capacity supply is determined to obtain the transport capacity occupancy index. Based on the centralized traffic status data in the standardized traffic dataset, the ratio of travel time to reference travel time is determined to obtain the traffic efficiency index. By applying a unified scale mapping to passenger flow density, capacity occupancy, and traffic efficiency indicators, a standardized set of indicators is obtained. An outlier identification and removal process is performed based on a standardized set of indicators to obtain a set of denoising indicators. The denoising index set is mapped according to the corresponding relationship of traffic stations to generate a station-level multidimensional state feature set, which is then spliced ​​and fused to obtain a multidimensional traffic state feature set.

[0022] Specifically, when extracting passenger flow density indicators based on standardized traffic datasets, the first step is to statistically summarize the passenger flow distribution data of each transportation station within a unit time window. This passenger flow distribution data characterizes the changes in passenger flow entering and exiting the station over different time periods. The average passenger flow per unit time refers to the average value calculated by accumulating the total passenger flow within the time window at a preset time granularity and then dividing it by the length of the time window. This process eliminates the randomness caused by short-term passenger flow fluctuations, thus obtaining a passenger flow density indicator that reflects the demand intensity of the station. For example, if a station accumulates 1,000 passenger visits within a ten-minute time window, its average passenger flow per unit time characterizes the passenger flow density of that station at that time scale. This method enables comparability between different stations at a unified time scale, providing a basic quantitative basis for subsequent comprehensive status analysis.

[0023] In calculating the capacity occupancy index, the ratio of actual carrying capacity to capacity supply is calculated based on public transportation operation data from various transportation stations in a standardized traffic dataset. Actual carrying capacity characterizes the actual passenger flow carried by vehicles or routes within a given time window, while capacity supply characterizes the maximum transport capacity that a station can provide within that time window, such as the number of available seats or the total dispatchable capacity per unit time. By calculating the ratio of actual carrying capacity to capacity supply, the capacity occupancy index is obtained. This index reflects the current utilization level of transport resources. A ratio close to one indicates that capacity is approaching saturation, while a low ratio indicates a surplus of capacity, thus providing a direct representation of the station's capacity strain. The traffic efficiency index is constructed based on traffic state data, calculated by comparing actual travel time with a reference travel time. The reference travel time is typically determined based on the average travel time under historical smooth traffic conditions or the designed travel time, serving as a benchmark. When the actual travel time is significantly higher than the reference travel time, the traffic efficiency index increases, indicating a decrease in traffic efficiency, which can reflect changes in the traffic operation status around the road segment or station.

[0024] After obtaining passenger flow density, capacity occupancy, and traffic efficiency indicators, a unified scale mapping process is applied to these three types of indicators. This unified scale mapping maps indicators with different dimensions and numerical ranges to a unified numerical interval, eliminating the impact of dimensional differences on subsequent fusion analysis. Common methods include normalization or interval scaling to ensure comparability of indicators at the same numerical scale. Based on this, outlier identification and removal are performed on the standardized indicator set. Outlier identification involves judging data deviating from the normal range based on statistical distribution characteristics or historical benchmarks. For example, if the passenger flow density at a station experiences a sudden change far exceeding the historical average within a short period, it is identified as an abnormal fluctuation and removed or corrected, thus forming a more stable and reliable set of denoised indicators. Subsequently, the denoised indicator set is mapped according to the traffic stations, and the passenger flow density indicator, capacity occupancy indicator and traffic efficiency indicator corresponding to each station are combined to form a station-level multidimensional state feature set. Then, the indicators of different dimensions are jointly expressed by feature splicing and fusion, so as to obtain a multidimensional traffic state feature set. This enables each traffic station to have multidimensional representation capabilities such as demand intensity, capacity status and operation efficiency in the same feature space, providing a unified input basis for subsequent graph structure modeling and correlation analysis.

[0025] Furthermore, a traffic association graph structure is constructed based on a multi-dimensional traffic state feature set. Traffic stations are mapped as nodes, passenger flow transfer relationships and road segment connection relationships are mapped as edges, and passenger flow density, capacity occupancy, and traffic efficiency indicators are encoded as node features to obtain traffic graph feature data, specifically including: Based on a multidimensional traffic state feature set, traffic station information in the feature set is obtained to determine the station set; The station set is subjected to structural parsing to extract the spatial location information and station identification information of each transportation station; Node identifiers are generated based on the uniqueness rules of station identifiers, and node mapping is completed after determining the spatial attribute association relationship of nodes in combination with spatial location information, thus mapping each transportation station to a node unit in the graph structure. Based on passenger flow distribution data, the passenger flow relationship between stations is analyzed to obtain the passenger flow transfer relationship between stations; Based on public transportation operation data, the route topology information is extracted and associated with stations to form a set of road segment connection relationships between stations; Understandably, when constructing graph structure nodes based on a multidimensional traffic state feature set, the first step is to extract traffic station information from this feature set to form a station set. Station information refers to the identifying data in the multidimensional traffic state feature set that corresponds one-to-one with a specific physical station, used to uniquely determine the location of each bus or rail transit station within the overall transportation network. After the station set is formed, it undergoes structural parsing processing. This process involves the field-level decomposition and reorganization of the station data to extract the spatial location information and station identification information corresponding to each traffic station. The spatial location information is typically represented by latitude and longitude coordinates or map projection coordinates, reflecting the geographical distribution of the station. The station identification information is used to distinguish the unique identity of different stations, such as a station code or a standardized unique identifier for the station name. After extracting the aforementioned information, node identifiers are generated based on the uniqueness rules of station identifiers. These uniqueness rules ensure that each station corresponds to only one node number through a pre-defined coding system or rule mapping method, thus avoiding duplicate or conflicting nodes during graph structure construction. Simultaneously, spatial location information is combined to determine the spatial attribute relationships between nodes. These relationships characterize the relative proximity or topological adjacency probability of different nodes in space, providing a spatial constraint basis for subsequent edge relationship construction. For example, adjacent subway stations, which are spatially close and usually have direct line connections, exhibit a high correlation tendency in their node spatial attributes. Through this process, transportation stations are mapped one by one to node units in the graph structure, giving each node both unique identifier attributes and spatial attribute expression capabilities.

[0026] After completing the node mapping, the passenger flow relationship between stations is analyzed based on passenger flow distribution data. The passenger flow distribution data is used to describe the changes in passenger entry and exit between stations within different time windows. By matching and analyzing the entry and exit flow within continuous time windows, the actual transfer path of passengers from one station to another can be identified, thus forming the passenger flow transfer relationship between stations. This relationship is used to characterize the strength of the real demand connection between stations. For example, the continuous one-way passenger flow from residential area stations to business area stations during the morning peak can constitute a stable passenger flow transfer relationship. Meanwhile, route topology information is extracted from public transportation operation data. This information is a structured representation of the route's operating path and station stop order, used to describe the connection and arrangement relationships of each station on different routes. By analyzing the route's operating trajectory, it is possible to determine which stations have direct adjacent stop relationships or reachable path relationships. These relationships are then matched and integrated with the station information to form a set of road segment connection relationships between stations. This set is used to characterize the physical connection structure in the transportation network, ensuring that the edges in the subsequent graph structure not only reflect passenger flow logic relationships but also actual route topology connection relationships. This guarantees that the graph structure can simultaneously express the dual constraints of demand flow relationships and physical traffic structure.

[0027] The consistency matching process between passenger flow transfer relationships and road segment connection relationships is performed to generate the edge relationship structure between nodes; Passenger flow density indicators are extracted from multidimensional traffic state features and mapped to each node to form the first type of node attributes. Capacity occupancy indicators are extracted from the multidimensional traffic state feature set and mapped to each node to form the second type of node attributes. At the same time, traffic efficiency indicators are extracted and mapped to each node to form the third type of node attributes. The three types of node attributes are uniformly encoded to form node feature vector representations, which are then fused and assembled with the edge relationship structure to obtain traffic map feature data.

[0028] Specifically, when performing consistency matching on passenger flow transfer relationships and road segment connection relationships, the passenger flow direction and intensity between stations described in the passenger flow transfer relationship are first used as the basis. This is then compared and verified with the actual connectivity topology between stations represented in the road segment connection relationship set. The passenger flow transfer relationship is used to represent the direction and relative scale of passenger flow inflow and outflow between different transportation stations per unit time, while the road segment connection relationship is used to represent whether there is an actual passable route connection between each transportation station and its connection direction attribute. By matching the two types of relationships one by one in the station pair dimension, relationship pairs that do not have actual connectivity in the road segment connection relationship but have abnormal flow direction in the passenger flow transfer relationship are eliminated. This generates an edge relationship structure between nodes that conforms to the actual connectivity constraints. This edge relationship structure is used to represent the effective passenger flow transmission channel formed between transportation stations via public transportation lines, and the connection relationship and its direction attribute are recorded with station pairs as the basic unit.

[0029] After forming the edge relationship structure, the passenger flow density index extracted from the multi-dimensional traffic state features is mapped one by one to the corresponding transportation stations. The passenger flow density index is used to characterize the average passenger flow passing through or gathering at a certain transportation station per unit time, and is assigned as the first type of node attribute reflecting the demand intensity of the station. At the same time, the capacity occupancy index is mapped to each transportation station to form the second type of node attribute. The capacity occupancy index is used to characterize the ratio between the actual carrying capacity and the capacity supply to reflect the current load level of public transportation resources at the station. Furthermore, the traffic efficiency index is mapped to each transportation station to form the third type of node attribute. The traffic efficiency index is used to characterize the ratio between the actual travel time and the reference travel time to reflect the smoothness of operation of the station and its associated travel routes. Subsequently, the first type of node attributes, the second type of node attributes, and the third type of node attributes are uniformly encoded. That is, by standardizing the mapping of indicators with different dimensions and value ranges, they are made into node feature vectors that can be jointly expressed under the same numerical scale. The node feature vectors are then fused and assembled with the obtained edge relationship structure to obtain traffic map feature data used to characterize the status of each traffic station and their interaction relationships.

[0030] Furthermore, a graph neural network model is constructed based on traffic map feature data to obtain the association weights between nodes. These association weights are modulated by combining the spatial distance and travel time corresponding to the edges to obtain a weighted association matrix. Finally, the neighborhood features of the nodes are weighted and aggregated to generate a fused traffic feature vector, specifically including: A graph neural network model is constructed based on traffic map feature data; Based on the edge relationship structure in the graph neural network model, identify the nodes that have a connection relationship and form a set of node pairs; For any node pair in the set of node pairs, extract the node feature vectors of the two corresponding nodes and obtain the difference vector between the two node feature vectors; Based on the node feature vectors, the covariance relationship between each feature dimension is statistically obtained, and the node feature covariance matrix is ​​constructed. Based on the difference vector and the node feature covariance matrix, the difference vector is standardized by distance measurement to obtain the feature difference values ​​between nodes; Based on the feature differences between nodes, basic association weights are generated using an exponential decay function. Specifically, when constructing a graph neural network model based on traffic map feature data, the node feature vectors in the traffic map feature data are used as initial input features, the edge relationship structure is used as the connection topology constraint in the graph neural network model, and traffic stations are used as node units for graph structure modeling. Each node corresponds to a multi-dimensional state feature expression of a traffic station. During the construction process, adjacency connections are established around nodes, enabling connected nodes to form information transmission paths in the model, thus allowing node features to be expressed in a structured manner within their neighborhood. Based on the structural definition of this graph neural network model, its internal node set, edge relationship structure, and node feature vectors are parsed to separate the corresponding structural components. Nodes represent each traffic station, the edge relationship structure represents the connection topology between stations, and the node feature vectors describe the state information of each station in multiple dimensions, including passenger flow density, capacity occupancy, and traffic efficiency. Based on the edge relationship structure, nodes with connections are paired up to form a set of node pairs for subsequent quantitative analysis of state differences between stations.

[0031] For any pair of nodes in the node pair set, the node feature vectors of the corresponding two nodes are extracted, and the difference between the two vectors in each feature dimension is calculated, thus obtaining the node feature difference vector. This difference vector is used to characterize the degree of deviation between the two stations in multidimensional states. Furthermore, the node feature covariance matrix refers to a matrix structure that statistically characterizes the joint change relationship between different feature dimensions based on the node feature vectors of all nodes. Its core purpose is to describe whether there is a statistical relationship of synchronous or inverse changes between different indicators. For example, in actual statistics, when the passenger flow density increases from approximately 500 passengers per hour to approximately 900 passengers per hour in a certain period, the capacity utilization usually also increases from approximately 0.4 to approximately 0.8, showing a relatively obvious positive correlation between the two; while the traffic efficiency index may decrease from approximately 1.2 to approximately 0.9, accompanied by an increase in capacity utilization, showing a negative correlation. By statistically analyzing the characteristic data of all nodes in their historical or current states, for example, with a scale of approximately one hundred stations, the three types of indicators—passenger flow density, capacity utilization, and traffic efficiency—are combined in pairs for calculation. This yields results such as a covariance of approximately 0.6 between passenger flow density and capacity utilization, approximately -0.5 between passenger flow density and traffic efficiency, and approximately -0.4 between capacity utilization and traffic efficiency. This forms a complete covariance matrix, which is used to characterize the overall coupling structure between multidimensional indicators, thereby avoiding ignoring the correlation between indicators in subsequent distance calculations.

[0032] Building upon this, the node feature difference vector is combined with the node feature covariance matrix. A standardized distance metric is used to scale the difference vector, enabling comparisons of feature dimensions with different dimensions and those exhibiting correlation at a unified scale. This yields the feature difference value between nodes, which comprehensively reflects the overall degree of difference between two sites in their multidimensional states. Subsequently, based on this feature difference value, an exponential decay function is introduced for mapping transformation. When the state difference between two nodes is small, the exponential function outputs a larger basic association weight; conversely, as the difference increases, the association weight decays rapidly in an exponential manner, thus generating the basic association weight between nodes to characterize the connection strength between sites based on their overall state similarity.

[0033] Obtain the spatial distance data and travel time data corresponding to the node pairs in the edge relationship structure, and perform normalization processing to obtain the distance coefficient and time coefficient; Based on the distance coefficient, the basic association weights are spatially corrected to obtain the distance-corrected association weights; The distance-corrected correlation weights are modulated by applying a time coefficient to obtain modulated correlation weights. The modulation association weights are normalized so that the sum of all association weights corresponding to the same node is one, thus obtaining the normalized association weights. Based on the correspondence between node pairs in the node pair set, the normalized association weights are arranged into a matrix to obtain the node association matrix. Specifically, when modeling the spatial and temporal influences of node pairs in an edge-relationship structure, the spatial distance and travel time data between each node pair are first obtained. These data are then normalized to map to a unified dimensionless interval, yielding corresponding distance and time coefficients. Based on these distance coefficients, the basic association weights obtained in the previous steps are spatially corrected. Specifically, the distance coefficient and the basic association weights are multiplicatively adjusted to reflect the attenuation or strengthening effect of spatial distance on node association strength at the weight level. For example, when the basic association weight of a node pair is 0.8 and the distance coefficient is 0.5, the distance-corrected association weight is 0.4. Subsequently, the temporal influence is introduced based on the distance-corrected association weights, and the time coefficient is multiplicatively modulated to further constrain the association strength. For example, when the distance-corrected association weight is 0.4 and the time coefficient is 0.7, the modulated association weight is 0.28, thus obtaining a comprehensive weight expression that integrates both spatial and temporal influences.

[0034] After calculating the modulation association weights, the modulation association weights corresponding to all node pairs are normalized. The normalization method involves scaling the sum of the modulation association weights of all target nodes corresponding to the same source node, ensuring that the sum of all association weights originating from the same node equals 1. This eliminates the scale bias caused by differences in the degree of different nodes. Finally, according to the correspondence of each node pair in the node pair set, the normalized association weights are organized into a matrix according to the order of the source and target nodes, forming a node association matrix. This matrix is ​​used to uniformly represent the association strength structure between nodes under the dual constraints of spatial distance and travel time. For example, if a node corresponds to three adjacent nodes with normalized association weights of 0.5, 0.3, and 0.2, the corresponding row vector in the matrix represents the ordered arrangement of these three values.

[0035] The neighboring nodes of each node are determined based on the node association matrix, thus obtaining the set of neighboring nodes; Based on the set of adjacent nodes, the node feature vectors of the corresponding nodes are extracted to form a node neighborhood feature set; Using the normalized association weights in the node association matrix as coefficients, the node neighborhood feature set is weighted and summed to obtain the node neighborhood aggregated feature vector. The neighborhood aggregation feature vector of a node is concatenated with the node feature vector of the corresponding node to generate a fused traffic feature vector. Specifically, when determining the adjacency relationship of each node based on the node association matrix, the normalized node association weight is used as the basis for adjacency determination. Non-zero or high-weight connected nodes corresponding to the same node in the node association matrix are considered its adjacent nodes, thus forming a corresponding set of adjacent nodes. This set of adjacent nodes can be understood as a combination of nodes with effective association strength with the target node after comprehensively considering the influence of spatial distance and travel time. For example, if the normalized association weights of a node with three candidate nodes are 0.6, 0.3, and 0.1 respectively, the two nodes with the higher weights are usually included in its set of adjacent nodes, thus reflecting a neighborhood structure that prioritizes strong associations. Based on this set of adjacent nodes, node feature vectors are extracted from the original node feature data. These node feature vectors characterize the node's own traffic attribute information, such as passenger flow intensity, traffic capacity, or operational status, among other multi-dimensional attributes. All node feature vectors corresponding to adjacent nodes are organized in a set according to their adjacency relationship, forming a node neighborhood feature set. This ensures that each target node corresponds to a set of feature descriptions reflecting its neighborhood structure.

[0036] After obtaining the node neighborhood feature set, the normalized association weights in the node association matrix are used as weighting coefficients to perform a weighted summation of the feature vectors of each node within the neighborhood. This process ensures that neighboring nodes with higher weights contribute a greater proportion to the aggregation result, thus generating a node neighborhood aggregated feature vector. For example, if the feature vectors of a node's three neighboring nodes correspond to weights of 0.5, 0.3, and 0.2 respectively, the final aggregation result is dominated by the feature vector of the node with the higher weight. Based on this, the node neighborhood aggregated feature vector is concatenated with the corresponding node's own node feature vector, i.e., sequentially connected at the feature dimension level. This allows the node's own attributes and neighborhood aggregated information to form a unified expression, generating a fused traffic feature vector. This fused vector can simultaneously reflect the node's local attributes and its neighborhood influence characteristics in the traffic network structure. For example, it includes both the node's own passenger flow information and the comprehensive operational status information of its neighboring nodes, thereby enhancing the ability of subsequent analysis to express the overall traffic structure relationship.

[0037] The formula for calculating the feature difference value is as follows: In the formula, For nodes With nodes The characteristic difference values ​​between them Represents a node The corresponding node feature vector, Represents a node The corresponding node feature vector, The node feature covariance matrix, This is a vector transpose operation; It is understandable that measuring the feature differences between nodes essentially involves quantifying the overall deviation between the two nodes in a multi-dimensional traffic state space. Specifically, the differences between the two nodes in multi-dimensional traffic features such as passenger flow density, capacity utilization, and traffic efficiency are first comprehensively calculated to obtain a basic difference vector, which is used to characterize the degree of deviation between the two nodes at the original state level. Since different feature dimensions have different fluctuation ranges and correlations in actual traffic networks—for example, some features fluctuate significantly across the entire network while others remain relatively stable—directly weighting the differences of each dimension can lead to the high-fluctuation dimension having an excessively dominant effect on the results, thereby weakening the objectivity and stability of the difference evaluation. Furthermore, in actual operation, there is also a certain correlation between different features. If this is not addressed, it may cause double-counting, causing the difference results to deviate from the true structural distance.

[0038] Based on this, a statistical relation matrix is ​​introduced to characterize the overall feature distribution structure, and the basic difference vector is modified. Its function is to unify the scale and eliminate correlations of the influence of different feature dimensions. For example, the difference contribution of features with large fluctuation ranges is appropriately compressed, while the difference influence of features with small fluctuations but sensitive to the system's operating state is appropriately enhanced, thus enabling different dimensions to participate in the comprehensive calculation under a unified evaluation scale. Physically, this feature difference value is used to characterize the degree of structural deviation between two nodes in the comprehensive traffic operating state, rather than simply the sum of numerical differences. For example, even if two nodes have similar values ​​on some indicators, as long as there is a deviation in key operational constraint features, the difference value will still increase significantly, reflecting a clear difference in their actual operating behavior and state structure. Conversely, even if some indicators differ greatly, if they belong to low-sensitivity or high-fluctuation dimensions in the system, their influence will be moderately weakened. Therefore, this feature difference value can be understood as a modified quantitative result of the true structural distance between nodes in the multi-dimensional traffic state space, used to more accurately characterize the similarity and deviation between nodes, and to provide a unified evaluation basis for subsequent neighborhood division and association weight calculation.

[0039] The formula for calculating the basic association weight is: In the formula, Based on the association weight, This is the preset attenuation coefficient.

[0040] Understandably, when calculating the basic association weights between nodes, an exponential decay method is used to map the feature difference values ​​to the association strength. The calculation formula is that the basic association weight equals an exponential function with the natural constant as its base, and the negative value of the product of the feature difference value and a preset decay coefficient is exponentially applied. Here, the feature difference value characterizes the overall deviation between two nodes in multi-dimensional traffic conditions such as passenger flow demand, transport capacity supply, and traffic efficiency. A larger value indicates weaker structural similarity between nodes, and therefore, their collaborative association strength should decrease accordingly. The role of the exponential function here is to transform this "greater difference, weaker association" relationship into a non-linear decay form, so that when the difference is small, the association weight remains at a high level, while as the difference gradually increases, the association weight decreases rapidly, thus conforming to the actual changing patterns of node collaborative relationships in a transportation network.

[0041] Based on this, a preset attenuation coefficient is used to control the sensitivity of differences to the influence of association strength. This preset coefficient is determined during the system modeling phase based on historical operational data. Specifically, firstly, the distribution of characteristic differences among a large number of historical node pairs and the corresponding changes in actual collaborative relationships are statistically analyzed. The attenuation trend range of association strength as differences increase is analyzed, and an overall mapping relationship between difference changes and association attenuation is constructed. Based on this, multiple candidate attenuation coefficients are tested. The calculated basic association weight is compared with the observed changes in node collaborative strength. The coefficient with the smallest error or highest matching degree is selected as the preset attenuation coefficient, using it as the evaluation criterion. This achieves a unified calibration of the sensitivity of global network associations. Physically, this basic association weight characterizes the similarity connection strength or closeness of collaboration between two nodes in the overall traffic operation state. Its changing pattern is as follows: when the characteristic differences between nodes are small, the system considers them to have strong collaborative potential, so the association weight is close to a high level; as the differences gradually increase, the possibility of collaboration decreases rapidly, and the association weight exhibits an exponential attenuation trend. By introducing a preset attenuation coefficient, the attenuation rate can be further adjusted, so that it can avoid the problem of insufficient discrimination when the difference is too small, and avoid the situation of excessive weakening of association when the difference is too large. This ensures that the entire node association matrix has stability and comparability under different traffic scenarios, and provides a unified quantitative basis for subsequent neighborhood construction and traffic management decisions.

[0042] Furthermore, based on the fused traffic feature vectors, a matching relationship between passenger demand intensity and transport capacity is constructed, obtaining the supply-demand deviation value, time fluctuation deviation value, and connection deviation value of each transportation station, and combining them to form a hierarchically coupled transport status feature, specifically including: The integrated traffic feature vector is obtained and its features are analyzed to obtain the comprehensive passenger flow density feature component, comprehensive transport capacity occupancy feature component, and comprehensive traffic efficiency feature component corresponding to each traffic station. The comprehensive passenger flow density feature component is used as the passenger flow demand intensity of each transportation station, and a set of passenger flow demand intensity is formed. The comprehensive passenger flow density feature component is the comprehensive representation result formed by superimposing the node's own passenger flow density feature with the neighboring aggregated passenger flow density feature. The standard capacity of each transportation station is obtained, and the capacity supply capacity of each transportation station is determined by combining the comprehensive capacity occupancy feature component. A set of capacity supply capacity is formed. The comprehensive capacity occupancy feature component is a comprehensive representation result formed by averaging the capacity occupancy feature of the node itself and the aggregated capacity occupancy feature of the neighborhood. The difference between passenger demand intensity and transport capacity is calculated to obtain the supply-demand deviation value of each transportation station. The supply-demand deviation value is the difference between passenger demand intensity and transport capacity. Based on the comprehensive traffic efficiency feature component in the fused traffic feature vector, the change sequence of each traffic station within a continuous time window is extracted, and the change difference between adjacent time windows is obtained as the time fluctuation deviation value. The comprehensive traffic efficiency feature component is the comprehensive representation result formed by selecting the minimum value of the node's own traffic efficiency feature and the neighborhood aggregated traffic efficiency feature. Specifically, when analyzing the integrated traffic feature vector, it is divided into three independent representation components according to a preset feature dimension division rule: comprehensive passenger flow density feature component, comprehensive capacity occupancy feature component, and comprehensive traffic efficiency feature component. The comprehensive passenger flow density feature component characterizes the overall passenger flow demand intensity of a transportation station under the influence of the local network. This component is obtained by superimposing the node's own passenger flow density feature with its neighboring aggregated passenger flow density feature. For example, if a station's own passenger flow density feature value is 60 and its neighboring aggregated passenger flow density feature value is 40, then its comprehensive passenger flow density feature component is 100, thus reflecting the station's total demand level under the influence of the local network. Based on this comprehensive passenger flow density feature component, it is used as the passenger flow demand intensity of each transportation station, and these are aggregated to form a passenger flow demand intensity set, used to uniformly represent the demand differences between different stations.

[0043] In supply-side analysis, the standard capacity of each transportation station is first obtained. This standard capacity characterizes the maximum transport capacity that a station can carry per unit time. The actual capacity supply capacity is then determined by combining this with a comprehensive capacity occupancy characteristic component. This comprehensive capacity occupancy characteristic component is obtained by averaging the node's own capacity occupancy characteristic and the aggregated capacity occupancy characteristic of its neighborhood. For example, if the node's own capacity occupancy characteristic is 0.7 and the aggregated capacity occupancy characteristic of its neighborhood is 0.5, then the comprehensive capacity occupancy characteristic component is 0.6. This allows the capacity supply capacity to simultaneously reflect local load and neighborhood influence. Subsequently, the difference between passenger demand intensity and capacity supply capacity is calculated to obtain the supply-demand deviation value. This deviation value characterizes the degree of supply-demand imbalance at the station level. For example, when passenger demand intensity is 100 and capacity supply capacity is converted to 80, the supply-demand deviation value is 20, reflecting the potential congestion pressure at that station.

[0044] Simultaneously, a comprehensive traffic efficiency feature component is extracted from the fused traffic feature vector. This component is obtained by minimizing the node's own traffic efficiency feature and the neighborhood aggregated traffic efficiency feature. It is used to emphasize the constraint-side impact of traffic capacity in the local network. For example, if the node's own traffic efficiency is 0.8 and the neighborhood aggregated traffic efficiency is 0.6, then the comprehensive traffic efficiency feature component is set to 0.6. Based on this component, the change sequence of each traffic station within a continuous time window is extracted, and the difference between adjacent time windows is calculated as a time fluctuation deviation value. This value is used to characterize the fluctuation intensity of the station's traffic status over time. For example, if the traffic efficiency of a station decreases from 0.7 to 0.5 within two adjacent time windows, then the time fluctuation deviation value is 0.2, thus reflecting the instability of its operating status.

[0045] Based on the edge relationship structure in the traffic map feature data, the connection relationship between adjacent traffic stations is extracted, and the supply and demand deviation value of the corresponding station is obtained. Based on the supply-demand deviation values ​​corresponding to the connected transportation stations, the absolute difference in supply-demand deviation values ​​between adjacent stations is calculated to obtain the connection deviation value. The connection deviation values ​​between stations are weighted and aggregated according to the connection relationship between adjacent stations to obtain the connection deviation value. The supply and demand deviation, time fluctuation deviation, and connection deviation are combined and processed to form a layered and coupled transportation status characteristic. Specifically, when extracting the connection relationships between adjacent transportation stations based on the edge relationship structure in the traffic map feature data, node pairs with effective normalized association weights in the node association matrix are identified as adjacent transportation station pairs. The supply-demand deviation value of the corresponding station is then obtained. For example, a supply-demand deviation value of 20 indicates that demand is significantly higher than supply, while -10 indicates supply redundancy. Based on this adjacency relationship, the absolute difference between the supply-demand deviation values ​​of each pair of adjacent stations is calculated, i.e., the absolute value of the difference between the two station supply-demand deviation values ​​is taken, thus obtaining the connection deviation value. This value reflects the spatial inconsistency of the degree of supply-demand imbalance between adjacent stations. For example, when the supply-demand deviation values ​​of two adjacent stations are 20 and 8 respectively, their connection deviation value is 12.

[0046] After obtaining the connection deviation values ​​for each pair of adjacent stations, they are weighted and aggregated according to the connection relationships between nodes. The weighting coefficients are taken from the normalized association weights in the node association matrix. For example, when the weights for a connection relationship are 0.6, 0.3, and 0.1, they correspond to the influence strength of different station pairs, allowing station pairs with higher connection strengths to have a greater impact on the overall result. This yields the connection deviation value, which is used to characterize the concentration of supply and demand differences within a local connection area. For example, if the weighted result for a certain area is 15, it indicates a high degree of overall connection imbalance. Subsequently, when combining the supply and demand deviation value, time fluctuation deviation value, and connection deviation value, it is necessary to perform scale coordination processing on the three types of indicators. The purpose is to ensure that different indicators can participate reasonably in the combination process, and that no single indicator will have an excessively dominant effect on the result due to differences in numerical magnitude. It is understandable that different types of deviation values ​​typically exhibit significant differences in numerical representation. For example, the supply-demand deviation value might be 20 or 30, the connection deviation value might be 15, while the time fluctuation deviation value is usually 0.3. If directly combined, the supply-demand deviation value and the connection deviation value would have a greater impact on the overall result, while the effect of the time fluctuation deviation value would be weakened. Therefore, in this step, by coordinating the participation of different indicators in the combination process, such as through structural layering or independent component expression, the three types of indicators are placed within a comparable expression framework when combined, rather than being directly subjected to numerical compression, thereby achieving a unified expression at the scale level. After completing the above scale coordination, the supply-demand deviation value, the time fluctuation deviation value, and the connection deviation value are combined as state representation components at different levels. The supply-demand deviation value is used to reflect the degree of supply-demand imbalance of a single station at the current moment, for example, 20; the time fluctuation deviation value is used to reflect the magnitude of change in the station's operating status within a continuous time window, for example, 0.3; and the connection deviation value is used to reflect the transmission and superposition effect of supply-demand differences between adjacent stations, for example, 15. These three elements are collaboratively expressed in a hierarchical structure, forming a layered and coupled evacuation status feature. This feature can simultaneously retain local node state information, temporal variation characteristics, and adjacency relationship influence information, without performing single numerical processing on them. For example, if a station corresponds to three types of deviation values ​​of 20, 0.3, and 15, then the evacuation status feature of that station is composed of these three sets of values. These values ​​are then used in the subsequent priority calculation step for unified scale mapping and weighting, thereby ensuring the integrity of information expression and the rationality of subsequent processing.

[0047] The formula for calculating transport capacity supply is as follows: In the formula, For the capacity supply of the a-th transportation station, Let be the characteristic component of the comprehensive transport capacity occupancy of the a-th transportation station. Let be the standard transport capacity of the a-th transportation station.

[0048] Understandably, when characterizing the capacity supply of transportation stations, it is necessary to consider both the station's maximum capacity limit and its current occupancy level. The standard capacity is used to represent the maximum transportation capacity that a given transportation station can provide under ideal conditions, such as the passenger flow handling capacity or vehicle turnover capacity per unit time; essentially, it reflects the station's basic service ceiling. The comprehensive capacity occupancy characteristic component characterizes the proportion of capacity currently occupied at the station during actual operation; a larger value indicates greater resource consumption and a smaller remaining available scheduling space.

[0049] Based on this, the comprehensive capacity occupancy characteristic component is processed to convert it into a proportion of available capacity. This proportion is then multiplied by the standard capacity to obtain the capacity supply capacity of the th transportation station. The physical meaning of this calculation method is as follows: when the comprehensive capacity occupancy characteristic component is small, it indicates that the station's current occupancy level is low, and there is a large amount of remaining capacity available for dispatching, thus the calculation result is close to the standard capacity. As the comprehensive capacity occupancy characteristic component gradually increases, it indicates that the station's operating load is continuously rising, the available resources are correspondingly decreasing, and the capacity supply capacity decreases linearly. When the comprehensive capacity occupancy characteristic component approaches its upper limit, it indicates that the station is under high load or even close to saturation, and its capacity available for external dispatching approaches zero. Through this attenuation mapping relationship based on occupancy ratio, the capacity supply capacity can more accurately reflect the station's remaining capacity level during dynamic operation, thus providing a stable quantitative basis for subsequent dispatching.

[0050] Furthermore, a set of scheduling strategies is generated based on the characteristics of the transportation status, and a point-to-point connection scheduling scheme is obtained according to each scheduling strategy. Scheduling execution instructions are output to drive scheduling execution. Then, historical public transportation scheduling records are obtained, a historical transportation scheduling case library is established, the scheduling strategies are optimized, an optimized scheduling scheme is obtained, and the historical transportation scheduling case library is updated. Specifically, this includes: Obtain the characteristics of the transportation status and extract the supply and demand deviation, time fluctuation deviation, and connection deviation values. After normalizing the supply-demand deviation, time fluctuation deviation, and connection deviation, the values ​​are weighted and summed according to preset weight coefficients to obtain the evacuation priority index of each transportation station. The transportation stations are then ranked according to the size of the evacuation priority index to form a station priority sequence. Based on the station priority sequence, high-priority transportation stations are extracted, and the corresponding adjacent transportation stations are determined by combining the edge relationship structure, forming a candidate evacuation station set; Using the site priority sequence as a constraint, candidate evacuation sites are sequentially arranged to generate multiple candidate site service orders and form a combination of site access order combinations. For each station access sequence combination, the corresponding capacity allocation ratio is determined by combining the supply and demand deviation value of each transportation station with the standard capacity capacity, thereby forming a set of scheduling strategies that include the station service sequence and the corresponding capacity allocation ratio. Specifically, after obtaining the characteristics of the transportation status, supply-demand deviation, time fluctuation deviation, and connection deviation are extracted. The supply-demand deviation represents the difference between passenger demand intensity and transportation capacity. For example, if the passenger demand intensity of a station is 120 and the transportation capacity is 100, the supply-demand deviation is 20. The time fluctuation deviation represents the magnitude of changes in the station's operating status within a continuous time window. For example, if the traffic efficiency decreases from 0.7 to 0.5, the time fluctuation deviation is 0.2. The connection deviation represents the degree of transmission of supply-demand differences between adjacent stations in the topological connection. Subsequently, the three types of deviation values ​​are normalized. The normalization adopts a linear mapping method based on historical statistical intervals, that is, using the historical maximum and minimum values ​​of the same type of indicator as boundaries. For example, if the historical maximum value of a certain type of supply-demand deviation is 40 and the minimum value is 0, then the value 20 is normalized to 0.5, thereby unifying the differences in the dimensions and scales of different indicators. Furthermore, the normalization results are weighted and summed according to preset weight coefficients. These preset weight coefficients are determined by back-calculation of historical transportation performance data. Specifically, the improvement in transportation efficiency within multiple time windows is statistically analyzed, and the contribution percentage of three types of deviations to the improvement results is calculated. For example, supply and demand deviation contributes 0.5, time fluctuation contributes 0.3, and connection and connection contribute 0.2. This yields the transportation priority index for each station, which is then sorted in descending order to form a station priority sequence. For example, the indices for five stations in a certain area are 0.82, 0.76, 0.55, 0.43, and 0.31, respectively.

[0051] After forming the site priority sequence, high-priority sites are selected as core distribution nodes using a preset threshold. This preset threshold is determined through quantile statistics of the historical priority index distribution; for example, the threshold is set to 0.7, corresponding to the top 30% quantile, ensuring that sites with a priority higher than 0.7 enter the core set. Subsequently, the adjacent sites of the core nodes are determined by combining the non-zero normalized weighted connections in the node association matrix of the edge relationship structure, thus forming a candidate distribution site set. For example, if a core site connects to 3 adjacent sites, a candidate set containing 4 sites is formed. Using the site priority sequence as a constraint, the candidate sites are sequentially arranged, generating multiple access order combinations while maintaining the priority order. For example, 4 candidate sites can form several different path arrangements. For each access sequence combination, the capacity allocation ratio is determined based on the ratio between the supply-demand deviation value of each station and the standard capacity capacity, where the standard capacity capacity is the maximum carrying capacity of the station per unit time, for example, set to 100. If the supply-demand deviations of the three stations in a combination are 20, 10, and 5 respectively, the basic ratios are 0.20, 0.10, and 0.05, and these are normalized to sum to 1, resulting in the final capacity allocation ratio. Finally, the access sequence combinations of each station and their corresponding capacity allocation ratios are summarized to form a scheduling strategy set, which is used for subsequent comparison of transportation schemes and determination of execution strategies.

[0052] Based on the station service order and capacity allocation ratio in each scheduling strategy, a point-to-point shuttle scheduling scheme is obtained; The point-to-point connection scheduling scheme is converted into scheduling execution instructions and output in chronological order to drive scheduling execution; Historical public transportation scheduling records are obtained, and the scheduling process under different time periods and different traffic operation states is sorted and classified. The corresponding scheduling scenarios, traffic station status and historical scheduling strategies are extracted to obtain historical scheduling basic data. The execution results of each historical scheduling strategy are analyzed to generate historical review suggestions, which are then structured to obtain review suggestion information. These review suggestions are opinions on adjusting scheduling strategies formed after analyzing the results of scheduling execution, and are used to guide the optimization of subsequent scheduling strategies. Historical dispatching data and review suggestions are linked and integrated to form a historical evacuation dispatching case library; Specifically, after obtaining the point-to-point shuttle scheduling scheme based on the station service order and capacity allocation ratio in each scheduling strategy, the scheduling scheme is converted into an executable scheduling execution instruction and output in chronological order. This instruction drives the corresponding vehicles or capacity resources to be scheduled and operated between different transportation stations. The station service order is used to determine the order of vehicle access, and the capacity allocation ratio is used to determine the scale of capacity input between each station. For example, in a certain scheduling scenario, the capacity allocation ratio from station A to station B is 60%, and the capacity allocation ratio from station A to station C is 40%. In the corresponding generated scheduling execution instruction, more vehicles are prioritized to travel from station A to station B, and capacity is added to station C in subsequent time windows, thereby responding to the differences in passenger flow demand between different stations. After the scheduling is completed, historical public transportation scheduling records are obtained, and the scheduling process in different time periods and under different traffic operation states is sorted and classified. For example, it is classified according to the morning peak congestion state, the off-peak stable state, and the sudden passenger flow growth state. At the same time, the scheduling scenarios, traffic station status, and historical scheduling strategies corresponding to each type of scheduling process are extracted to obtain historical scheduling basic data with a unified structure. The scheduling scenario is used to describe the overall traffic environment at that time, the traffic station status is used to reflect the passenger flow distribution and supply and demand relationship of each station, and the historical scheduling strategy is used to record the actual scheduling plan executed.

[0053] Simultaneously, after obtaining historical scheduling data, the execution results of each historical scheduling strategy are analyzed to identify the performance of the scheduling strategy in actual operation. For example, by comparing changes in the number of stranded passengers, passenger flow speed, and capacity utilization at each station before and after scheduling execution, it is determined whether the scheduling strategy effectively alleviated the supply-demand mismatch problem. Based on this, historical review suggestions are generated. These review suggestions are adjustments to the scheduling strategy formed after analyzing the effects of the scheduling execution results. For example, in a historical case, if the number of stranded passengers decreased from 100 to 40 after increasing the capacity of a transfer node, a review suggestion of "prioritizing the increase of capacity configuration at this node" is generated. Or, if a detour on a certain route leads to a decrease in traffic efficiency, a review suggestion of "adjusting the service order of stations to shorten the route" is generated. Subsequently, the historical scheduling data and review suggestion information are linked and integrated to ensure that each historical scheduling strategy corresponds to its execution results and review suggestions, thereby forming a structured historical evacuation scheduling case library for retrieval and reference in subsequent scheduling processes.

[0054] After the scheduling is completed, obtain the scheduling scenario and traffic station status corresponding to the current scheduling task, search for similar cases in the historical evacuation scheduling case library, and obtain matching historical scheduling strategies and historical review suggestions. Based on historical scheduling strategies and historical review suggestions, the current scheduling scheme is adjusted and optimized to obtain the optimized scheduling scheme. The optimized scheduling scheme is executed, and the results are analyzed to generate post-optimization review suggestions. The current scheduling scenarios, traffic station status, optimized scheduling schemes, and corresponding post-optimization review suggestions are structured and organized, and the historical evacuation scheduling case library is updated to form an updated evacuation scheduling case library.

[0055] Specifically, after the scheduling is completed, the scheduling scenario and traffic station status corresponding to the current scheduling task are first obtained. The scheduling scenario describes the overall traffic operation environment within the current time window, such as the morning rush hour congestion stage or a local area passenger flow surge stage. The traffic station status reflects the current passenger flow distribution and supply-demand imbalance at each station, such as a station having a capacity shortage while adjacent stations have excess capacity. Based on this, the current scheduling scenario and traffic station status are used as search conditions to search for similar cases in the historical evacuation scheduling case database. By comparing the corresponding scheduling scenarios and traffic station status in historical cases, historical cases with similar characteristics to the current situation are selected, and the corresponding historical scheduling strategies and historical review suggestions are obtained. For example, when a transfer node in the current scenario is congested and adjacent nodes have excess capacity, the same type of congestion transfer cases in history are retrieved, and the historical scheduling strategies that achieved evacuation by increasing capacity or adjusting connecting routes, as well as the corresponding review suggestions, are extracted, thus providing a reference for the current scheduling.

[0056] Simultaneously, after obtaining matching historical scheduling strategies and historical review suggestions, the current scheduling plan is adjusted and optimized based on the capacity allocation methods and station service sequences reflected in the historical scheduling strategies, and in conjunction with the adjustment opinions reflected in the historical review suggestions. For example, if the historical review suggestions indicate that "prioritizing the increase of capacity to a certain key node can significantly reduce the number of stranded people," the capacity allocation ratio of that node in the current scheduling plan is increased; or if the historical review suggestions indicate that "adjusting the service sequence can shorten the overall travel time," the station access sequence in the current scheduling plan is optimized, thus obtaining an optimized scheduling plan. Subsequently, the optimized scheduling plan is executed, and the execution results are analyzed, such as comparing the changes in the number of stranded people or the changes in passenger flow evacuation efficiency at each station before and after optimization, generating optimized review suggestions. Finally, the current scheduling scenario, traffic station status, optimized scheduling plan, and corresponding optimized review suggestions are structured and written into the historical evacuation scheduling case library to achieve continuous updates to the case library, enabling subsequent scheduling processes to be referenced and optimized based on continuously accumulated historical experience.

[0057] Subsequently, through the synergistic effect of scheduling execution results, historical review suggestions, and updates to the historical transportation scheduling case library, the scheduling execution results first provide feedback on changes in passenger flow at each transportation station after actual execution. For example, the number of people stranded at a certain station may decrease from 50 to 20, reflecting the actual impact of the scheduling strategy on the supply and demand relationship. Secondly, by analyzing the effects of the scheduling execution results, review suggestions are generated to summarize and correct problems in the scheduling strategy, such as unreasonable capacity allocation or suboptimal station service order. For example, it is identified that prioritizing the increase of capacity at a certain node in a specific scenario can effectively reduce congestion, thus forming reusable scheduling strategy adjustment opinions. At the same time, combined with the continuous updates to the historical transportation scheduling case library, new scheduling scenarios, transportation station statuses, scheduling plans, and corresponding review suggestions are constantly recorded and accumulated, enriching the historical experience data foundation and improving the reference capability for subsequent scheduling. Driven by the synergy of the aforementioned multiple types of information, the entire scheduling-execution-review-update process forms a continuously iterative operating mechanism. This operating mechanism is abstracted into a public transportation intelligent agent with continuous perception, state analysis, strategy optimization, and experience accumulation capabilities. This intelligent agent is essentially composed of a scheduling decision-making process, an execution result feedback process, and a strategy optimization process based on a case library. It can continuously receive multi-source traffic data within a continuous time window and dynamically adjust the scheduling strategy in combination with historical experience, thereby realizing adaptive optimization of the traffic network operation status and the expression of scheduling decision-making capabilities.

[0058] Furthermore, a public transportation integrated distribution system based on multi-source data fusion is proposed to implement any of the above-mentioned distribution methods, including: The data acquisition and processing module is used to acquire public transportation operation data, passenger flow distribution data, and traffic status data within the target area, and to perform time alignment and spatial mapping processing on the data to generate a standardized traffic dataset. The feature extraction module is used to extract passenger flow density indicators, capacity occupancy indicators and traffic efficiency indicators from the standardized traffic dataset, and perform unified scale mapping and noise reduction to form a multi-dimensional traffic state feature set. The graph construction module is used to construct a traffic association graph structure based on a multi-dimensional traffic state feature set, mapping traffic stations as nodes, passenger flow transfer relationships and road segment connection relationships as edges, and generating traffic graph feature data. The graph neural network processing module is used to construct a graph neural network model based on traffic map feature data, calculate the association weights between nodes, modulate the association weights by combining spatial distance and travel time, generate a weighted association matrix, and perform weighted aggregation of the neighborhood features of nodes to generate a fused traffic feature vector. The status analysis module is used to construct the matching relationship between passenger flow demand intensity and transport capacity supply based on the fused traffic feature vector, generate supply and demand deviation values, time fluctuation deviation values ​​and connection deviation values, and form the evacuation status characteristics. The scheduling and case optimization module is used to generate a set of scheduling strategies and a point-to-point connection scheduling scheme based on the characteristics of the transportation status. It also obtains historical public transportation scheduling records to establish a historical transportation scheduling case library, searches for similar cases in the historical transportation scheduling case library, optimizes the scheduling strategies, and updates the historical transportation scheduling case library.

[0059] Furthermore, the data acquisition and processing module includes: The data acquisition unit is used to acquire data on the operational status of public transportation routes and the capacity of stations. Passenger flow collection unit: The passenger flow collection unit is used to acquire passenger flow distribution data and passenger flow time distribution data at each transportation station; Status acquisition unit, which is used to acquire traffic status data related to the road network traffic status; Time alignment unit, used to perform unified time-granularity resampling processing on multi-source data; Spatial mapping units are used to map data to a unified spatial coordinate system and divide it according to station affiliation, generating standardized traffic datasets.

[0060] Furthermore, the graph building module includes: The node mapping unit is used to map transportation stations as graph structure nodes. The edge-construction unit is used to parse the passenger flow transfer relationship and road segment connection relationship between stations; The relationship fusion unit is used to match and fuse passenger flow transfer relationships with road segment connection relationships to generate an edge relationship structure; The attribute encoding unit is used to encode passenger flow density indicators, capacity occupancy indicators and traffic efficiency indicators into node feature vectors. The graph generation unit is used to fuse node feature vectors with edge relationship structures to generate traffic map feature data.

[0061] Furthermore, the scheduling and case optimization module includes: The strategy generation unit is used to generate a set of candidate evacuation sites and a combination of site access order based on the site priority sequence. The scheduling generation unit is used to generate a point-to-point shuttle scheduling scheme based on the combination of site access order and capacity allocation ratio. The case construction unit is used to obtain historical public transportation scheduling records and generate a historical evacuation scheduling case library. The case retrieval unit is used to search for similar cases in the historical evacuation and dispatch case database. The strategy optimization unit is used to optimize the scheduling strategy based on historical scheduling strategies and historical review suggestions. The case update unit is used to analyze the results of the optimized scheduling scheme, generate review suggestions, and update the historical evacuation scheduling case library.

[0062] In summary, the advantages of this invention are as follows: by performing time alignment and spatial mapping on multi-source traffic data, a unified and standardized expression and highly consistent fusion of data are achieved; by constructing a traffic association graph structure and calculating node association weights and a modulation mechanism based on spatial distance and travel time, the modeling ability of traffic network topology and passenger flow transfer relationships is enhanced; by constructing a matching relationship between passenger flow demand intensity and transport capacity supply capacity, and introducing a hierarchical characterization mechanism for supply-demand deviation, time fluctuation deviation, and connection deviation, a refined expression of multi-dimensional transport status is achieved; by establishing a historical transport scheduling case library and introducing review suggestions, the scheduling strategy is optimized and the case library is updated, enabling the scheduling process to have the ability to accumulate experience and continuously optimize, thereby improving the efficiency of public transport and the system's collaborative control capabilities.

[0063] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A public transportation integrated distribution method based on multi-source data fusion, characterized in that, include: Public transportation data within the target area is acquired and time-aligned and spatially mapped to obtain a standardized traffic dataset. The public transportation data includes public transportation operation data, passenger flow distribution data, and traffic status data. Based on a standardized traffic dataset, passenger flow density, capacity occupancy, and traffic efficiency indicators are extracted and fused to obtain a multi-dimensional traffic state feature set. Based on a multidimensional traffic state feature set, a traffic association graph structure is constructed, which maps traffic stations as nodes, passenger flow transfer relationships and road segment connection relationships as edges, and encodes passenger flow density index, capacity occupancy index and traffic efficiency index as node features to obtain traffic graph feature data. A graph neural network model is constructed based on traffic map feature data to obtain the association weights between nodes. The association weights are modulated by combining the spatial distance and travel time corresponding to the edges to obtain a weighted association matrix. The neighborhood features of the nodes are then weighted and aggregated to generate a fused traffic feature vector. Based on the fusion of traffic feature vectors, a matching relationship between passenger flow demand intensity and transport capacity supply is constructed to obtain the supply-demand deviation value, time fluctuation deviation value and connection deviation value of each transportation station, and combine them to form a hierarchical coupled transport status feature. A set of scheduling strategies is generated based on the characteristics of the transportation status, and a point-to-point connection scheduling scheme is obtained according to each scheduling strategy. The scheduling execution instruction is output to drive the scheduling execution. Then, historical public transportation scheduling records are obtained, a historical transportation scheduling case library is established, the scheduling strategies are optimized, the optimized scheduling scheme is obtained, and the historical transportation scheduling case library is updated.

2. The integrated public transportation distribution method based on multi-source data fusion according to claim 1, characterized in that, Based on a standardized traffic dataset, passenger flow density, capacity occupancy, and traffic efficiency indicators are extracted and fused to obtain a multi-dimensional traffic state feature set, specifically including: Data analysis is performed on public transportation operation data, passenger flow distribution data, and traffic status data to obtain a data set containing time stamps and spatial location information. The traffic status data includes information related to road traffic status, travel time, and congestion status. Based on the time stamps in the dataset, the public transportation operation data, passenger flow distribution data, and traffic status data are resampled according to a unified time granularity to obtain a time-aligned dataset. Based on the spatial location information in the time-aligned data set, public transportation operation data, passenger flow distribution data, and traffic status data are mapped to a unified spatial coordinate system, and spatial affiliation is divided according to transportation stations to obtain a spatial mapping data set; The spatial mapping dataset is aggregated according to transportation stations to obtain a standardized transportation dataset. Based on the passenger flow distribution data of each transportation station in the standardized traffic dataset, the average passenger flow per unit time is extracted to obtain the passenger flow density index. Based on the public transportation operation data of each transportation station in the standardized transportation dataset, the ratio of actual carrying capacity to transport capacity supply is determined to obtain the transport capacity occupancy index. Based on the centralized traffic status data in the standardized traffic dataset, the ratio of travel time to reference travel time is determined to obtain the traffic efficiency index. By applying a unified scale mapping to passenger flow density, capacity occupancy, and traffic efficiency indicators, a standardized set of indicators is obtained. An outlier identification and removal process is performed based on a standardized set of indicators to obtain a set of denoising indicators. The denoising index set is mapped according to the correspondence of traffic stations to generate a station-level multidimensional state feature set, which is then spliced ​​and fused to obtain a multidimensional traffic state feature set.

3. The integrated public transportation distribution method based on multi-source data fusion according to claim 1, characterized in that, The traffic association graph structure constructed based on the multi-dimensional traffic state feature set maps traffic stations as nodes, passenger flow transfer relationships and road segment connection relationships as edges, and encodes passenger flow density indicators, capacity occupancy indicators, and traffic efficiency indicators as node features to obtain traffic graph feature data, specifically including: Based on a multidimensional traffic state feature set, traffic station information in the feature set is obtained to determine the station set; The station set is subjected to structural parsing to extract the spatial location information and station identification information of each transportation station; Node identifiers are generated based on the uniqueness rules of station identifiers, and node mapping is completed after determining the spatial attribute association relationship of nodes in combination with spatial location information, thus mapping each transportation station to a node unit in the graph structure. Based on passenger flow distribution data, the passenger flow relationship between stations is analyzed to obtain the passenger flow transfer relationship between stations; Based on public transportation operation data, the route topology information is extracted and associated with stations to form a set of road segment connection relationships between stations; The consistency matching process between passenger flow transfer relationships and road segment connection relationships is performed to generate the edge relationship structure between nodes; Passenger flow density indicators are extracted from multidimensional traffic state features and mapped to each node to form the first type of node attributes. Capacity occupancy indicators are extracted from the multidimensional traffic state feature set and mapped to each node to form the second type of node attributes. At the same time, traffic efficiency indicators are extracted and mapped to each node to form the third type of node attributes. The three types of node attributes are uniformly encoded to form node feature vector representations, which are then fused and assembled with the edge relationship structure to obtain traffic map feature data.

4. The integrated public transportation distribution method based on multi-source data fusion according to claim 1, characterized in that, The graph neural network model constructed based on traffic map feature data obtains the association weights between nodes. These association weights are modulated by combining the spatial distance and travel time corresponding to the edges to obtain a weighted association matrix. Furthermore, the neighborhood features of the nodes are weighted and aggregated to generate a fused traffic feature vector. Specifically, this includes: A graph neural network model is constructed based on traffic map feature data. Based on the edge relationship structure in the graph neural network model, identify the nodes that have a connection relationship and form a set of node pairs; For any node pair in the set of node pairs, extract the node feature vectors of the two corresponding nodes and obtain the difference vector between the two node feature vectors; Based on the node feature vectors, the covariance relationship between each feature dimension is statistically obtained, and the node feature covariance matrix is ​​constructed. Based on the difference vector and the node feature covariance matrix, the difference vector is standardized by distance measurement to obtain the feature difference values ​​between nodes; Based on the feature differences between nodes, basic association weights are generated using an exponential decay function. Obtain the spatial distance data and travel time data corresponding to the node pairs in the edge relationship structure, and perform normalization processing to obtain the distance coefficient and time coefficient; Based on the distance coefficient, the basic association weights are spatially corrected to obtain the distance-corrected association weights; The distance-corrected correlation weights are modulated by time coefficients to obtain modulated correlation weights. The modulation association weights are normalized so that the sum of all association weights corresponding to the same node is one, thus obtaining the normalized association weights. Based on the correspondence between node pairs in the node pair set, the normalized association weights are arranged into a matrix to obtain the node association matrix. The neighboring nodes of each node are determined based on the node association matrix, thus obtaining the set of neighboring nodes; The node feature vectors of the corresponding nodes are extracted based on the set of adjacent nodes to form a node neighborhood feature set; Using the normalized association weights in the node association matrix as coefficients, the node neighborhood feature set is weighted and summed to obtain the node neighborhood aggregated feature vector. The neighborhood aggregation feature vector of a node is concatenated with the node feature vector of the corresponding node to generate a fused traffic feature vector. The formula for calculating the feature difference value is as follows: In the formula, For nodes With nodes The characteristic difference values ​​between them Represents a node The corresponding node feature vector, Represents a node The corresponding node feature vector, The node feature covariance matrix, This is a vector transpose operation; The formula for calculating the basic association weight is as follows: In the formula, Based on the association weight, This is the preset attenuation coefficient.

5. The integrated public transportation distribution method based on multi-source data fusion according to claim 1, characterized in that, The method, based on fused traffic feature vectors, constructs a matching relationship between passenger demand intensity and transport capacity supply, obtaining supply-demand deviation values, time fluctuation deviation values, and connection deviation values ​​for each transportation station, and combines them to form a layered coupled transport status feature, specifically including: The integrated traffic feature vector is obtained and its features are analyzed to obtain the comprehensive passenger flow density feature component, comprehensive transport capacity occupancy feature component, and comprehensive traffic efficiency feature component corresponding to each traffic station. The comprehensive passenger flow density feature component is used as the passenger flow demand intensity of each transportation station, and a passenger flow demand intensity set is formed. The comprehensive passenger flow density feature component is a comprehensive representation result formed by superimposing the passenger flow density feature of the node itself with the aggregated passenger flow density feature of the neighborhood. The standard capacity of each transportation station is obtained, and the capacity supply capacity of each transportation station is determined by combining the comprehensive capacity occupancy feature component, and a capacity supply capacity set is formed. The comprehensive capacity occupancy feature component is a comprehensive representation result formed by averaging the capacity occupancy feature of the node itself and the aggregated capacity occupancy feature of the neighborhood. The difference between passenger demand intensity and transport capacity is calculated to obtain the supply-demand deviation value of each transportation station. The supply-demand deviation value is the difference between passenger demand intensity and transport capacity. Based on the comprehensive traffic efficiency feature component in the fused traffic feature vector, the change sequence of each traffic station within a continuous time window is extracted, and the change difference between adjacent time windows is obtained as the time fluctuation deviation value. The comprehensive traffic efficiency feature component is a comprehensive representation result formed by selecting the minimum value of the node's own traffic efficiency feature and the neighborhood aggregated traffic efficiency feature. Based on the edge relationship structure in the traffic map feature data, the connection relationship between adjacent traffic stations is extracted, and the supply and demand deviation value of the corresponding station is obtained. Based on the supply-demand deviation values ​​corresponding to the connected transportation stations, the absolute difference in supply-demand deviation values ​​between adjacent stations is calculated to obtain the connection deviation value. The connection deviation values ​​between stations are weighted and aggregated according to the connection relationship between adjacent stations to obtain the connection deviation value. The supply and demand deviation, time fluctuation deviation, and connection deviation are combined and processed to form a layered and coupled transportation status characteristic. The formula for calculating the transport capacity supply is as follows: In the formula, For the capacity supply of the a-th transportation station, Let be the characteristic component of the comprehensive transport capacity occupancy of the a-th transportation station. Let be the standard transport capacity of the a-th transportation station.

6. The integrated public transportation distribution method based on multi-source data fusion according to claim 1, characterized in that, The process involves generating a set of scheduling strategies based on the characteristics of the transportation status, obtaining point-to-point connection scheduling schemes according to each strategy, outputting scheduling execution instructions to drive scheduling execution, then acquiring historical public transportation scheduling records, establishing a historical transportation scheduling case library, optimizing the scheduling strategies to obtain optimized scheduling schemes, and updating the historical transportation scheduling case library. Specifically, this includes: Obtain the characteristics of the transportation status and extract the supply and demand deviation, time fluctuation deviation, and connection deviation values; After normalizing the supply-demand deviation, time fluctuation deviation, and connection deviation, the values ​​are weighted and summed according to preset weight coefficients to obtain the evacuation priority index of each transportation station. The transportation stations are then ranked according to the size of the evacuation priority index to form a station priority sequence. Based on the station priority sequence, high-priority transportation stations are extracted, and the corresponding adjacent transportation stations are determined by combining the edge relationship structure, forming a candidate evacuation station set; Using the site priority sequence as a constraint, candidate evacuation sites are sequentially arranged to generate multiple candidate site service orders and form a combination of site access order combinations. For each station access sequence combination, the corresponding capacity allocation ratio is determined by combining the supply and demand deviation value of each transportation station with the standard capacity capacity, thereby forming a set of scheduling strategies that include the station service sequence and the corresponding capacity allocation ratio. Based on the station service order and capacity allocation ratio in each scheduling strategy, a point-to-point shuttle scheduling scheme is obtained; The point-to-point connection scheduling scheme is converted into scheduling execution instructions and output in chronological order to drive scheduling execution; Historical public transportation scheduling records are obtained, and the scheduling process under different time periods and different traffic operation states is sorted and classified. The corresponding scheduling scenarios, traffic station status and historical scheduling strategies are extracted to obtain historical scheduling basic data. The execution results of each historical scheduling strategy are analyzed to generate historical review suggestions, which are then structured to obtain review suggestion information. These review suggestions are opinions on adjusting scheduling strategies formed after analyzing the results of scheduling execution, and are used to guide the optimization of subsequent scheduling strategies. Historical dispatching data and review suggestions are linked and integrated to form a historical evacuation dispatching case library; After the scheduling is completed, obtain the scheduling scenario and traffic station status corresponding to the current scheduling task, search for similar cases in the historical evacuation scheduling case library, and obtain matching historical scheduling strategies and historical review suggestions. Based on historical scheduling strategies and historical review suggestions, the current scheduling scheme is adjusted and optimized to obtain the optimized scheduling scheme. The optimized scheduling scheme is executed, and the results are analyzed to generate post-optimization review suggestions. The current scheduling scenarios, traffic station status, optimized scheduling schemes, and corresponding post-optimization review suggestions are structured and organized, and the historical evacuation scheduling case library is updated to form an updated evacuation scheduling case library.

7. A public transportation integrated distribution system based on multi-source data fusion, used to implement the distribution method as described in any one of claims 1-6, characterized in that, include: The data acquisition and processing module is used to acquire public transportation operation data, passenger flow distribution data and traffic status data within the target area, and to perform time alignment and spatial mapping processing on the data to generate a standardized traffic dataset. The feature extraction module is used to extract passenger flow density indicators, capacity occupancy indicators and traffic efficiency indicators from a standardized traffic dataset, and perform unified scale mapping and noise reduction processing to form a multi-dimensional traffic state feature set. The graph construction module is used to construct a traffic association graph structure based on a multi-dimensional traffic state feature set, map traffic stations as nodes, map passenger flow transfer relationships and road segment connection relationships as edges, and generate traffic graph feature data. The graph neural network processing module is used to construct a graph neural network model based on traffic map feature data, calculate the association weights between nodes, modulate the association weights by combining spatial distance and travel time, generate a weighted association matrix, and perform weighted aggregation of the node neighborhood features to generate a fused traffic feature vector. The status analysis module is used to construct a matching relationship between passenger flow demand intensity and transport capacity supply based on the fused traffic feature vector, generate supply and demand deviation values, time fluctuation deviation values ​​and connection deviation values, and form evacuation status characteristics. The scheduling and case optimization module is used to generate a set of scheduling strategies and a point-to-point connection scheduling scheme based on the characteristics of the transportation status, and to obtain historical public transportation scheduling records to establish a historical transportation scheduling case library. Similar cases are searched in the historical transportation scheduling case library, the scheduling strategies are optimized, and the historical transportation scheduling case library is updated.

8. A public transportation integrated distribution system based on multi-source data fusion according to claim 7, characterized in that, The data acquisition and processing module includes: The data acquisition unit is used to acquire public transportation route operation status data and station capacity carrying capacity data. Passenger flow collection unit, which is used to acquire passenger flow distribution data and passenger flow time distribution data at each transportation station; A status acquisition unit is used to acquire traffic status data related to the road network traffic status. A time alignment unit is used to perform unified time-granularity resampling processing on multi-source data; A spatial mapping unit is used to map data to a unified spatial coordinate system and divide it according to station affiliation to generate a standardized traffic dataset.

9. A public transportation integrated distribution system based on multi-source data fusion according to claim 7, characterized in that, The graph construction module includes: A node mapping unit, wherein the node mapping unit is used to map traffic stations as graph structure nodes; An edge construction unit is used to parse the passenger flow transfer relationship and road segment connection relationship between stations; A relationship fusion unit is used to match and fuse passenger flow transfer relationships with road segment connection relationships to generate an edge relationship structure; An attribute encoding unit is used to encode passenger flow density indicators, capacity occupancy indicators and traffic efficiency indicators into node feature vectors. The graph generation unit is used to fuse node feature vectors with edge relationship structures to generate traffic map feature data.

10. A public transportation integrated distribution system based on multi-source data fusion according to claim 7, characterized in that, The scheduling and case optimization module includes: A strategy generation unit is used to generate a set of candidate evacuation sites and a combination of site access order based on a site priority sequence. A scheduling generation unit is used to generate a point-to-point connection scheduling scheme based on the combination of site access order and capacity allocation ratio. Case construction unit, which is used to acquire historical public transportation scheduling records and generate a historical evacuation scheduling case library; A case retrieval unit, which is used to retrieve similar cases from a historical evacuation and dispatch case database; A strategy optimization unit is used to optimize the scheduling strategy based on historical scheduling strategies and historical review suggestions. The case update unit is used to analyze the results of the optimized scheduling scheme, generate review suggestions, and update the historical evacuation scheduling case library.

Images