Transfer connection optimization method in multi-mode traffic integrated planning

Abstract

The invention relates to the technical field of traffic planning, in particular to a transfer connection optimization method in multi-mode traffic integrated planning, which comprises the steps of data acquisition and demand analysis, multi-mode transfer network topology design, multi-mode timetable collaborative optimization, transfer service system development and facility transport capacity dynamic matching. According to the transfer connection optimization method in the multi-mode traffic integrated planning, a dynamic and static combined multi-source data acquisition framework is constructed based on a multi-dimensional data fusion system, the data precision can be improved, the network can be perceived in real time, association modeling of weather events and demand fluctuation is realized based on construction of a multi-agent simulation system, and the optimization of the transfer connection in the multi-mode traffic integrated planning is realized. A dynamic weight mechanism is introduced based on a hypergraph model, the robustness of the algorithm is improved, a layered optimization framework processes global optimization through upper-layer integer programming, dynamic adjustment is achieved through lower-layer deep reinforcement learning, a buffer time intelligent insertion strategy can improve the late absorption capacity, and a facility utility function model is established to improve the transfer efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic planning, and specifically provides an optimization method for transfer connection in multi-modal traffic integration planning. Background Art

[0002] Transfer connection optimization refers to scientifically planning and dynamically adjusting transfer nodes, timetables, facility layouts, and transport capacity configurations between different transportation modes, enabling passengers to achieve seamless connection with the shortest time, lowest cost, and optimal experience when switching between different traffic modes (such as subways, buses, shared bicycles, walking, etc.).

[0003] After retrieval, a flat peak period train connection coordination optimization method, system, device, and medium are disclosed in the invention patent with the Chinese patent publication number CN117669826A. The train connection coordination optimization method in this invention patent constructs a flat peak period train connection coordination optimization model with the method of translating train operation lines and the objective function of the shortest total waiting time for passengers to transfer, assigns a certain translation amount to the train operation lines, obtains the optimization order according to the importance of the line and the importance of the transfer station, and adjusts the arrival and departure times of the trains to achieve the transfer connection coordination between trains.

[0004] However, this flat peak period train connection coordination optimization method can only be applied to the connection scheduling of trains during the flat peak period, and cannot achieve transfer connection coordination for different time periods in different traffic modes. Moreover, this connection coordination optimization method cannot associate weather events with demand fluctuations. Therefore, there is still room for further improvement in realizing the global optimization of traffic transfer connection. Thus, an optimization method for transfer connection in multi-modal traffic integration planning is proposed. Summary of the Invention

[0005] (1) Technical Problems to be Solved

[0006] Aiming at the deficiencies of the prior art, the present invention provides an optimization method for transfer connection in multi-modal traffic integration planning, which has the advantages of realizing the correlation between weather events and demand fluctuations and jointly optimizing by combining timetables and transport capacity matching, and solves the problem that the existing flat peak period train connection coordination optimization method in the above background art cannot achieve global optimization for different time periods in different traffic modes.

[0007] (2) Technical Solutions

[0008] To achieve the above purpose of realizing the correlation between weather events and demand fluctuations and jointly optimizing by combining timetables and transport capacity matching, the present invention provides the following technical solutions: An optimization method for transfer connection in multi-modal traffic integration planning, including the following steps:

[0009] S1. Data collection and demand analysis: Build the dynamic data foundation of the multi-modal transportation network and conduct demand modeling and analysis;

[0010] S2. Topological design of the multi-modal transfer network: Optimize the physical layout and functional coordination of transfer nodes;

[0011] S3. Coordinated optimization of multi-modal timetables: Reduce the waiting time for cross-modal transfers and improve the robustness of timetables;

[0012] S4. Development of the transfer service system: Provide real-time transfer decision support;

[0013] S5. Dynamic matching of facility capacity: Ensure the real-time adaptation of the capacity of transfer facilities to passenger flow demand.

[0014] Preferably, multi-source data is obtained based on the transportation network, and the multi-source data includes:

[0015] 1) Static data: a. Transportation network topology: The line planning distribution, station location distribution, and transfer node location of different transportation modes; b. Timetable operation: The starting and ending times of departure, the location distribution of stops, and the headway duration; c. Facility attributes: Station capacity, parking lot capacity, and the location distribution of shared vehicle drop-off points; d. Transfer rules: The transfer distance and transfer time threshold of different transportation modes;

[0016] 2) Dynamic data: a. Real-time traffic conditions: The traffic flow, driving speed and direction of different lines, and road congestion; b. Passenger flow data: The passenger flow, population distribution and flow direction of different stations, and travel purposes; c. External factors: Weather conditions and traffic emergencies; d. User individual data: Travel trajectory preferences and transfer tendencies;

[0017] The specific steps for obtaining multi-source data include:

[0018] 1) Static data acquisition: a. Enter the data open platform of the urban traffic management department, download the GTFS dataset to obtain the timetables and line station information under different transportation modes; b. Obtain the road network distribution, station coordinates, and spatial data of different transfer channels through OpenStreetMap; c. Obtain the location distribution of drop-off points and service areas of shared travel enterprises;

[0019] 2) Dynamic data acquisition: a. Obtain the real-time positions of different means of transportation based on the vehicle-mounted positioning system; b. Obtain the passenger flow density data based on the cameras and WIFI probe modules set at the platforms; c. Obtain the real-time traffic conditions based on the data interface of the transportation system and third-party map services; d. Record the user transfer query data based on the user terminal, and analyze the starting and ending points and route distribution of user travel.

[0020] Preferably, data processing is performed based on the acquired static data and dynamic data, including:

[0021] 1) Data cleaning: a. Based on GPS drift points, outliers in the dynamic data are removed. By importing the pandas and matplotlib databases, they are used for data processing and visual display; GPS data is loaded through a CSV file and invalid data containing null value rows is removed; based on the calculation of distance and time difference, and by setting distance thresholds and time thresholds to determine the drift point criteria; drift points exceeding the set threshold range are removed for visual comparison;

[0022] b. Calculate the average value of sensor data, and fill in the missing sensor data according to the average interpolation method;

[0023] c. Start the Arc workstation environment, enter the edit mode by inputting the ArcEdit command, load the data using the ec+ command and the Coverage file path, display the hanging lines based on the dearcnodedangle command, and repair errors according to nodecdangle and draw;

[0024] 2) Data standardization: a. Standardize the coordinate data in different maps based on a unified coordinate system; b. Standardize the timetable data of different traffic modes based on timestamp alignment;

[0025] 3) Feature extraction: a. Extract the site topological relationship and transfer channel length from the static data; b. Calculate the real-time passenger flow density and vehicle delay duration for the dynamic data.

[0026] Preferably, based on the extracted features of the static data and dynamic data, the dynamic data is accurately mapped to the static road network. By establishing the causal relationship and coordination rules between the dynamic and static data, a joint analysis framework is constructed, and a dynamic-static data fusion model is established to fuse and process the static data and dynamic data, specifically including:

[0027] 1) Map matching: Match the vehicle GPS trajectory points to the nearest road or track according to the hidden Markov model, specifically:

[0028] a. For each GPS point o t Set to search for all road segments {r t,i} within a surrounding radius R, project the GPS point onto the nearest point of the candidate road segment, and record the projected point coordinates and the distance d t,i ;

[0029] b. Set the actual road segment r t,i where the vehicle is located as the hidden state. According to the GPS point o t Set the observation probability P(o t,i generated by the road segment r t|r t,i ), and construct the observation model, expressed as:

[0030] Where σ is the standard deviation of GPS error;

[0031] c. According to the road section r from the previous moment t-1,i Transfer to current section r t,j The possibility of setting the transition probability P(r t,j |r t-1,i ), based on the path connectivity judgment, if r t,j With r t-1,i Direct connection has a higher probability. The distance ΔS between the two projection points is calculated based on the motion consistency, the time difference Δt with the GPS point and the maximum speed v of the vehicle. max By comparison, we have Where λ is the adjustment parameter;

[0032] d. Through the observation probability P(o t |r t,i ) directly determines the initial probability P(r1,i) of the first GPS point o1, and calculates the maximum probability path of each candidate section at each moment according to the recursive formula, which is expressed as:

[0033] δ t (j) = max i [δ t-1 (i)·P(r t,j |r t-1,i )]·P(o t |r t,j ), where δ t (j) represents the maximum cumulative probability of road segment j at time t. We trace back from the maximum probability state at the final moment to obtain the optimal road segment sequence and achieve map matching.

[0034] 2) Spatiotemporal gridding: Divide the city into a 500m×500m grid structure, aggregate dynamic data and static data, and import them into the grid structure for distribution;

[0035] 3) Define nodes and edges based on static data, assign edge weights using dynamic data, build a transfer relationship graph, and use graph neural networks to combine static topology with dynamic features to predict demand propagation across the entire network;

[0036] 4) Take static data as system state and dynamic data as observation value, update state estimation through Kalman filtering, establish state space model, capture spatial topology and time series at the same time, predict transfer demand in future time period based on spatiotemporal graph convolution network, and integrate static road network and dynamic data into virtual simulation environment with digital twin platform to simulate multimodal traffic flow in real time.

[0037] Preferably, travel demand characteristics are extracted and multi-modal demand is modeled, including:

[0038] 1) Static demand modeling: Based on historical data analysis of periodic demands such as commuting and going to school, OD distribution at the station level, peak hours, and transfer hot spot data are extracted, and similar travel patterns are identified based on cluster analysis;

[0039] 2) Dynamic demand modeling: Based on the LSTM model, the passenger flow at the station in the next 30 minutes is predicted for real-time passenger flow prediction, and combined with event-driven demand, the data of sudden increase in temporary demand is displayed in real-time. Based on time series analysis, trends and periodicities are captured, and the correlation between weather, events and demand changes is analyzed;

[0040] 3) Demand decomposition and correlation: The contributions of different transportation modes are distinguished for mode splitting, a transfer selection probability model is established, considering walking time, waiting time, and comfort factors, and the weights are dynamically adjusted;

[0041] 4) Network-level demand modeling: The transportation network is modeled as a graph structure, and the dynamic demand is used as the node feature to predict the propagation of the whole network demand. According to the minimization of the whole network travel time and transfer waiting time, an objective function is constructed, and constraint conditions are set according to vehicle capacity, timetable coordination, and resource scheduling upper limit to construct a collaborative optimization model.

[0042] Preferably, the steps of the multi-modal transfer network topology design specifically include:

[0043] 1) Transfer node division and hierarchy construction:

[0044] a. Node definition: The stations of a single transportation mode are set as physical nodes, the multi-modal intersection points are set as transfer hubs, and multiple stations within a short distance are merged into logical nodes and then virtually processed;

[0045] b. Hierarchy division: Hierarchical processing is carried out according to transportation mode and function;

[0046] 2) Transfer edge connection and weight calculation:

[0047] a. Edge type design: The connections within the same transportation mode are set as intra-mode edges, the transfer connections between different transportation modes are set as inter-mode edges, and the walking paths are set as virtual transfer edges.

[0048] b. Weight modeling: The time weight calculation is expressed as transfer time = walking time + waiting time, the transfer cost is set as the economic weight, and multiple indicators are fused by the entropy weight method to calculate the comprehensive weight. The formula is expressed as:

[0049] W ij =αT ij +βC ij +γ(1 - Sij )

[0050] where T ij is time, C ij is cost, S ij is comfort level (0 - 1), and α, β, γ are weight coefficients;

[0051] 3) Construction of network topology model: Establish a hypergraph model, generate a topology table based on the adjacency list, calculate indicators such as degree distribution, clustering coefficient, and average path length, and analyze the characteristics of complex networks;

[0052] 4) Transfer optimization and dynamic adjustment:

[0053] a. Static optimization: Solve the optimal transfer path through the Dijkstra algorithm, analyze the critical path, and simulate the design of redundant paths under node failure to enhance robustness;

[0054] b. Dynamic adjustment: Dynamically adjust the weights of edges according to passenger flow data, predict future passenger flow changes based on the LSTM network, and dynamically add or delete nodes or edges to achieve topological evolution.

[0055] Preferably, the steps of the multi - mode timetable collaborative optimization specifically include:

[0056] 1) Construction of a collaborative optimization model:

[0057] a. Design an objective function based on minimizing the average transfer waiting time across the network, and perform multi - objective trade - off by balancing time efficiency and operating cost through Pareto optimization, expressed as:

[0058]

[0059] where is the transfer waiting time, P ij is the transfer flow, and the range of departure interval fluctuations is restricted to maximize the robustness of the timetable;

[0060] b. Set the vehicle turnover time ≤ t min and the driver's working time less than the legal upper limit as physical constraint conditions, set the effective transfer time window as a transfer constraint condition, and satisfy:

[0061] where Δt tolerance is the passenger's tolerated waiting time;

[0062] 3) Algorithm optimization:

[0063] a. Static collaborative optimization: Discretize the timetable into time slots based on integer programming, use the departure time as the decision variable, and solve the global optimal solution;

[0064] b. Dynamic real-time adjustment: Define the state space, action space, and reward function, and train an agent to dynamically adjust the timetable based on the PPO algorithm.

[0065] c. Rolling horizon optimization: Resolve the timetable within the next 1 hour every 15 minutes and update it in combination with real-time data;

[0066] 4) Transfer time window synchronization:

[0067] a. Identification of key transfer points: Identify high-traffic transfer hubs through complex network centrality metrics (such as betweenness centrality), and use the DBSCAN algorithm to find high-frequency transfer OD pairs and time periods.

[0068] b. Timetable offset strategy: Adjust the departure time of secondary transportation modes so that their arrival times match the arrival times of the main transportation.

[0069] c. Buffer time insertion: Reserve time buffer at key transfer nodes to cope with delays, expressed as:

[0070] t buffer =μ delay +kσ delay where μ delay is the average delay and k is the confidence coefficient.

[0071] Preferably, the transfer service system includes: a knowledge graph module that integrates multi-source data by constructing a knowledge graph;

[0072] a dynamic path recommendation module that dynamically generates an optimal transfer plan based on reinforcement learning;

[0073] a user profile module that provides differentiated recommendations according to different user situations;

[0074] and an integrated service platform;

[0075] The dynamic matching of facility capacity includes:

[0076] 1) Construction of a dynamic demand-resource matching model: Set the objective function of minimizing the passenger waiting time and the facility congestion cost, and set the physical facility capacity limit and the feasibility of capacity scheduling as constraints, expressed as:

[0077] where is the waiting time at transfer point i, is the congestion penalty;

[0078] 2) Classification of matching strategies: Schedule logistics vehicles to make up for shortages in real time according to the outbound passenger flow, dynamically adjust the placement of shared bicycles, and add flexible bus schedules during peak transfer periods to achieve short-term response. Achieve long-term optimization by reallocating parking lot resources and adjusting subway formations;

[0079] 3) Design of dynamic matching algorithm:

[0080] a. Optimization of transport capacity scheduling: For dynamic vehicle scheduling (DVRP) according to real-time demand, use reinforcement learning (such as DQN) or heuristic algorithms for variants of the vehicle routing problem, adjust the bus departure interval based on queuing theory, and achieve coordinated adjustment of shifts, expressed as: (M / M / 1 model), where λ is the arrival rate and μ is the service rate;

[0081] b. Dynamic allocation of facility resources: Use mixed integer programming (MIP) to optimize the layout of transfer node facilities and achieve spatial reconfiguration, expressed as: max∑ i,j x ij ·u ij s.t.∑x ij ≤S i , where x ij is the allocation quantity of facility type j at node i, and u ij is the utility value;

[0082] 4) Edge computing and fast response:

[0083] a. Deployment of edge nodes: Deploy edge servers at transfer hubs to process local data in real time;

[0084] b. Control strategy: Trigger actions based on a rule engine and use fuzzy logic to handle uncertainties;

[0085] 5) Closed-loop feedback optimization:

[0086] a. Digital twin simulation: Rehearse the scheduling strategy in a virtual environment and implement it specifically after evaluating the effect;

[0087] b. Online training of reinforcement learning: Feed back the actual operation data to the model and continuously optimize the scheduling strategy.

[0088] (III) Beneficial effects

[0089] Compared with the prior art, the present invention provides a method for optimizing transfer connection in multimodal transportation integration, having the following beneficial effects:

[0090] 1. The method for optimizing transfer connection in multimodal transportation integration constructs a dynamic and static combined multi-source data acquisition framework based on a multi-dimensional data fusion system, which can improve data accuracy and real-time perception of the network.

[0091] 2. The method for optimizing transfer connection in multimodal transportation integration realizes the correlation modeling of weather events and demand fluctuations based on the construction of a multi-agent simulation system, and introduces a dynamic weight mechanism based on a hypergraph model to improve the robustness of the algorithm.

[0092] 3. The transfer connection optimization method in this multi-modal transportation integration plan designs a hierarchical optimization framework: the upper-layer integer programming is used to handle global optimization, the lower-layer deep reinforcement learning is used to achieve dynamic adjustment, the buffer time intelligent insertion strategy can improve the late arrival absorption ability, and the facility utility function model is established to improve the transfer efficiency. Brief Description of the Drawings

[0093] Figure 1 It is a flow chart of the transfer connection optimization method of the present invention. Detailed Embodiments

[0094] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the embodiments and drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0095] Furthermore, multi-source data is obtained based on the transportation network, and the multi-source data includes:

[0096] 1) Static data: a. Transportation network topology: the line planning distribution, station location distribution, and transfer node location of different transportation modes; b. Timetable operation: the start and end times of departure of shifts, the location distribution of stops, and the headway duration; c. Facility attributes: station capacity, parking lot capacity, and the location distribution of shared vehicle drop-off points; d. Transfer rules: the transfer distance and transfer time threshold of different transportation modes;

[0097] 2) Dynamic data: a. Real-time traffic conditions: the traffic flow, driving speed and direction of different lines, and the road congestion situation; b. Passenger flow data: the passenger flow, population distribution and flow direction of different stations, and the travel purpose; c. External force factors: weather conditions and traffic emergencies; d. User individual data: travel trajectory preferences and transfer tendencies;

[0098] The specific steps for obtaining multi-source data include:

[0099] 1) Static data acquisition: a. Enter the data open platform of the urban transportation management department, download the GTFS dataset to obtain the timetable and line station information under different transportation modes; b. Obtain the road network distribution, station coordinates, and spatial data of different transfer channels through OpenStreetMap; c. Obtain the location distribution of the drop-off points and service areas of shared travel enterprises;

[0100] 2) Dynamic data acquisition: a. Obtain the real-time locations of different transportation vehicles based on the vehicle positioning system; b. Obtain passenger flow density data based on the cameras and WIFI probe modules installed at stations; c. Obtain real-time traffic conditions based on the data interface of the transportation system and third-party map services; d. Record the user transfer query data on the user terminal, and analyze the starting and ending locations and route distributions of user trips.

[0101] Specifically, by integrating static data and dynamic data, construct a complete data portrait of the urban transportation system. Through systematic hierarchical design covering the physical layer, rule layer, dynamic layer, and user layer of the transportation system, a complete four-dimensional data system is formed, integrating government open data (GTFS), crowdsourced map data, enterprise proprietary data, and IoT real-time data, breaking through the limitations of a single data source. The static data provides basic constraints, and the dynamic data reflects real-time changes. The spatio-temporal alignment of the two supports dynamic decision-making.

[0102] Furthermore, perform data processing based on the obtained static data and dynamic data, including:

[0103] 1) Data cleaning: a. Remove outliers in the dynamic data based on GPS drift points. Import the pandas and matplotlib databases for data processing and visualization display; load GPS data through a CSV file and remove invalid data containing null value rows; calculate based on distance and time difference, and set distance thresholds and time thresholds to determine the drift point criteria; remove drift points exceeding the set threshold range and conduct visual comparison;

[0104] b. Calculate the average value of sensor data and fill in the missing sensor data according to the mean interpolation method;

[0105] c. Start the Arc workstation environment, enter the editing mode by inputting the ArcEdit command, load data using the ec+ command and the Coverage file path, display hanging lines based on the dearcnodedangle command, and repair errors according to nodecdangle and draw;

[0106] 2) Data standardization: a. Standardize the coordinate data in different maps based on a unified coordinate system; b. Standardize the timetable data of different transportation modes based on timestamp alignment;

[0107] 3) Feature extraction: a. Extract the topological relationship of stations and the length of transfer channels from static data; b. Calculate the real-time passenger flow density and vehicle delay duration from dynamic data.

[0108] Specifically, by constructing a progressive process of data cleaning - standardization - feature extraction, secondary pollution caused by process confusion is avoided. For GPS drift point determination, spatio - temporal double - threshold constraints are adopted, which can identify real anomalies more accurately than a single threshold.

[0109] Furthermore, according to the characteristics of the extracted static data and dynamic data, the dynamic data is accurately mapped to the static road network. By establishing the causal relationship and collaboration rules between dynamic and static data, a joint analysis framework is constructed, and a dynamic - static data fusion model is established to fuse and process static data and dynamic data, specifically including:

[0110] 1) Map matching: Match vehicle GPS trajectory points to the nearest road or track according to the Hidden Markov Model, specifically as follows:

[0111] a. For each GPS point o t Set to search for all road segments {r t,i} within a surrounding radius R, project the GPS point onto the nearest point of the candidate road segment, and record the projected point coordinates and the distance d to the road segment t,i ;

[0112] b. Set the actual road segment r t,i where the vehicle is located as the hidden state. According to the GPS point o t generated by the road segment r t,i set the observation probability P(o t |r t,i ), and construct an observation model, expressed as: where σ is the standard deviation of GPS error;

[0114] c. Set the transition probability P(r t-1,i |r t,j ) according to the probability of transferring from the road segment r t,j at the previous moment to the current road segment r t-1,i ). Based on path connectivity judgment, if r t,j and r t-1,i are directly connected, the probability is relatively high. Calculate the distance ΔS between the two projected points based on motion consistency, compare it with the time difference Δt of the GPS point and the maximum speed v max of the vehicle, then there is where λ is an adjustment parameter;

[0115] d. Determine the initial probability P(r1,i) of the first GPS point o1 directly through the observation probability P(o t |r t,i ). Calculate the maximum - probability path of each candidate road segment at each moment according to the recurrence formula, expressed as:

[0116] δ t (j)=max i[δ t-1 (i)·P(r t,j |r t-1,i )]·P(o t |r t,j ), where δ t (j) represents the maximum cumulative probability at road segment j at time t. By tracing back from the maximum probability state at the final time, the optimal road segment sequence is obtained to achieve map matching;

[0117] 2) Spatiotemporal grid processing: Divide the city into a 500m×500m grid structure, aggregate dynamic data and static data and import them into the grid structure for distribution;

[0118] 3) Define nodes and edges according to static data, assign weights to edges using dynamic data, construct a transfer relationship graph, and combine static topology and dynamic features using a graph neural network to predict the propagation of network-wide demand;

[0119] 4) Use static data as the system state and dynamic data as the observed values, update the state estimate through Kalman filtering, establish a state space model, capture spatial topology and time series simultaneously, predict transfer demand within a future time period based on a spatiotemporal graph convolutional network, and integrate the static road network and dynamic data into a virtual simulation environment in combination with a digital twin platform to real-time simulate multi-modal traffic flow.

[0120] Specifically, by establishing the spatiotemporal mapping relationship between dynamic data and the static road network, realizing the virtual-real interaction closed-loop of the traffic system, integrating the probability model (HMM), graph computing (GNN), state estimation (Kalman filtering) and deep learning (STGCN), it can break through the limitations of a single model, and the double probability constraints of high-precision map matching can improve the trajectory matching accuracy, which is significantly better than traditional geometric matching, and combine spatiotemporal separation processing technology to reduce the model complexity.

[0121] Furthermore, extract travel demand characteristics and model multi-modal demand, including:

[0122] 1) Static demand modeling: Based on historical data analysis of periodic demands such as commuting and going to school, extract station-level OD distribution, peak hours, and transfer hot spot data, and identify similar travel patterns based on cluster analysis;

[0123] 2) Dynamic demand modeling: Based on the LSTM model, predict the passenger flow at stations in the next 30 minutes, perform real-time passenger flow prediction, and combine event-driven demand to display real-time data on sudden surges in temporary demand. Capture trends and periodicities based on time series analysis, and analyze the correlation between weather, events and demand changes;

[0124] 3) Demand decomposition and association: Distinguish the contributions of different transportation modes for mode splitting, establish a transfer selection probability model, consider walking time, waiting time, and comfort factors, and dynamically adjust the weights;

[0125] 4) Network-level demand modeling: Model the transportation network as a graph structure, with dynamic demand as node features, predict the propagation of network-wide demand, construct an objective function based on minimizing the network-wide travel time and transfer waiting time, and set constraints according to vehicle capacity, schedule coordination, and resource scheduling upper limits to construct a collaborative optimization model.

[0126] Specifically, by integrating static laws and dynamic changes, achieve accurate characterization and prediction of multi-modal transportation demand. The static layer captures long-term laws, and the dynamic layer responds to real-time changes, forming a complementary modeling framework that simultaneously captures time dependence and external driving factors, effectively reducing prediction errors. Introduce psychological perception variables into the transfer selection model, which can effectively improve the goodness of fit (R 2 ) compared with the traditional Logit model, and achieve network-level global optimization.

[0127] Furthermore, the steps of the multi-modal transfer network topology design specifically include:

[0128] 1) Transfer node division and hierarchy construction:

[0129] a. Node definition: Set the stations of a single transportation mode as physical nodes, set the multi-modal intersection points as transfer hubs, and virtually process multiple stations within a short distance after merging them into logical nodes;

[0130] b. Hierarchy division: Process by layer according to transportation mode and function;

[0131] 2) Transfer edge connection and weight calculation:

[0132] a. Edge type design: Set the connections within the same transportation mode as intra-mode edges, set the transfer connections between different transportation modes as inter-mode edges, and set the walking path as a virtual transfer edge;

[0133] b. Weight modeling: The time weight calculation is expressed as transfer time = walking time + waiting time. Set the transfer cost as the economic weight, and calculate the comprehensive weight by fusing multiple indicators through the entropy weight method. The formula is expressed as:

[0134] W ij = αT ij + βC ij + γ(1 - S ij ),

[0135] where T ij is time, C ij is cost, and S ijis the comfort level (0-1), α, β, γ are weight coefficients;

[0136] 3) Network topology model construction: establish a hypergraph model, make a topology table based on the adjacency table, calculate indicators such as degree distribution, clustering coefficient, average path length, and analyze complex network characteristics;

[0137] 4) Transfer optimization and dynamic adjustment:

[0138] a. Static optimization: Solve the optimal transfer path through the Dijkstra algorithm, analyze the critical path, and simulate the redundant path design under node failure to enhance robustness;

[0139] b. Dynamic adjustment: Dynamically adjust the weight of the edge according to the passenger flow data, predict future passenger flow changes based on the LSTM network, dynamically add or delete nodes or edges, and realize topological evolution.

[0140] Specifically, the type classification covers the connection within the same mode of transportation, between different modes of transportation, and walking paths. The weight modeling comprehensively considers multiple important factors such as time, cost, and comfort. The entropy weight method is used to integrate multiple indicators to calculate the comprehensive weight, which can more comprehensively measure the actual situation of transfers and provide a more accurate basis for subsequent path planning and analysis.

[0141] Establishing a hypergraph model and calculating a variety of complex network characteristic indicators can help us gain a deeper understanding of the structure and properties of the network, analyze the network topology from multiple perspectives, and provide a scientific basis for optimizing the network;

[0142] Static optimal path solving and critical path analysis are performed, and edge weights, additions and deletions of nodes or edges are dynamically adjusted according to passenger flow data to achieve topological evolution. This can adapt to different needs and changes, improve the adaptability and robustness of the network, and make the multi-modal transfer network more efficient and reliable in actual operation.

[0143] Furthermore, the steps of collaborative optimization of the multi-mode timetable specifically include:

[0144] 1) Build a collaborative optimization model:

[0145] a. Design the objective function based on minimizing the average transfer waiting time of the entire network, and balance time efficiency and operating costs through Pareto optimization to make multi-objective trade-offs, expressed as:

[0146]

[0147] in is the transfer waiting time, P ij To account for transfer traffic, constrain the fluctuation range of departure intervals to maximize the robustness of the timetable;

[0148] b. Set vehicle turnover time ≤ tmin The driver's working hours are less than the upper limit of the regulation as the physical constraint condition, and the effective transfer time window is set as the transfer constraint condition, and it meets the following conditions:

[0149] Where Δt tolerance Tolerating waiting times for passengers;

[0150] 3) Algorithm optimization:

[0151] a. Static collaborative optimization: discretize the timetable into time slots based on integer programming, use the departure time as the decision variable, and solve the global optimal solution;

[0152] b. Dynamic real-time adjustment: Define the state space, action space, and reward function, and train the agent to dynamically adjust the schedule based on the PPO algorithm.

[0153] c. Rolling time domain optimization: Resolve the timetable for the next hour every 15 minutes and update it in combination with real-time data;

[0154] 4) Transfer time window synchronization:

[0155] a. Identification of key transfer points: Identify high-traffic transfer hubs through complex network centrality indicators (such as betweenness centrality), and use the DBSCAN algorithm to find high-frequency transfer OD pairs and time periods.

[0156] b. Schedule shift strategy: adjust the departure time of secondary transportation modes so that their arrival time matches the arrival time of primary transportation modes;

[0157] c. Buffer time insertion: reserve time buffer at key transfer nodes to cope with delays, expressed as:

[0158] t buffer =μ delay +kσ delay , where μ delay is the average delay, and k is the confidence coefficient.

[0159] Specifically, the objective function not only considers minimizing the average transfer waiting time of the entire network and improving the travel experience of passengers, but also balances time efficiency and operating costs through Pareto optimization, taking into account the interests of the operator, while constraining the fluctuation range of departure intervals and enhancing the robustness of the timetable;

[0160] Static collaborative optimization uses integer programming to solve for the global optimal solution, ensuring the optimality of the timetable under certain conditions; dynamic real-time adjustment trains an intelligent agent based on the PPO algorithm, which can flexibly adjust the timetable according to the actual situation to adapt to the dynamically changing environment; rolling horizon optimization combines real-time data to re-solve the timetable for future periods at regular intervals, further improving the timeliness and adaptability of the timetable. Different algorithms cooperate with each other to meet the optimization requirements in different scenarios;

[0161] By using complex network metrics and clustering algorithms, key transfer points and high-frequency transfer periods are accurately identified, providing a clear direction for subsequent timetable adjustment; the timetable offset strategy reasonably adjusts the departure times of secondary transportation modes to better connect different transportation modes; buffer time insertion takes into account delays, reserves space for response, and improves the reliability and stability of transfers.

[0162] Furthermore, the transfer service system includes: a knowledge graph module that integrates multi-source data by constructing a knowledge graph;

[0163] a dynamic path recommendation module that dynamically generates an optimal transfer plan based on reinforcement learning;

[0164] a user profile module that provides differentiated recommendations according to different user situations;

[0165] and an integrated service platform;

[0166] The dynamic matching of facility capacity includes:

[0167] 1) Construction of a dynamic demand-resource matching model: Set the objective function to minimize the passenger waiting time and the facility congestion cost, and set the physical facility capacity limit and the feasibility of capacity scheduling as constraints, expressed as:

[0168] , where is the waiting time at transfer point i, is the congestion penalty;

[0169] 2) Classification of matching strategies: Schedule logistics vehicles to replenish vehicles in real time according to the outbound passenger flow, dynamically adjust the placement of shared bicycles, and increase flexible bus schedules during peak transfer periods to achieve short-term response. Long-term optimization is achieved by reallocating parking lot resources and adjusting subway formations;

[0170] 3) Design of dynamic matching algorithms:

[0171] a. Optimization of capacity scheduling: For dynamic vehicle routing (DVRP) according to real-time demand, use reinforcement learning (such as DQN) or heuristic algorithms for variants of the vehicle routing problem, and adjust the bus departure interval based on queuing theory to achieve coordinated adjustment of schedules, expressed as: (M / M / 1 model), where λ is the arrival rate and μ is the service rate;

[0172] b. Dynamic allocation of facility resources: Use mixed integer programming (MIP) to optimize the layout of transfer node facilities and achieve spatial reconfiguration, expressed as: max∑ i,j x ij ·u ij s.t.∑x ij ≤S i where x ij is the allocation quantity of facility type j at node i, and u ij is the utility value;

[0173] 4) Edge computing and quick response:

[0174] a. Deployment of edge nodes: Deploy edge servers at transfer hubs to process local data in real time;

[0175] b. Control strategy: Trigger actions based on a rule engine and use fuzzy logic to handle uncertainties;

[0176] 5) Closed-loop feedback optimization:

[0177] a. Digital twin simulation: Rehearse scheduling strategies in a virtual environment and implement them specifically after evaluating the effects;

[0178] b. Online training of reinforcement learning: Feed actual operation data back to the model to continuously optimize the scheduling strategy.

[0179] Specifically, with the objective of minimizing the passenger waiting time and facility congestion cost, considering the physical facility capacity limit and the feasibility of transport capacity scheduling, it balances the constraints of the passenger experience and actual operation, making the model more practical and reasonable;

[0180] The classification of short-term response and long-term optimization strategies makes reasonable arrangements for the needs of different time scales. It can not only quickly respond to real-time passenger flow changes but also rationally allocate and adjust resources from a long-term perspective, making the matching of facility transport capacity more scientific and comprehensive;

[0181] Deploying edge servers at transfer hubs to process local data in real time and performing control based on a rule engine and fuzzy logic can quickly respond to on-site situations, adjust strategies in a timely manner, and improve the response speed and processing capacity of the system.

[0182] The beneficial effects of the present invention are as follows: The transfer connection optimization method in the multi-mode transportation integration planning constructs a multi-source data acquisition framework combining dynamic and static data based on a multi-dimensional data fusion system, which can improve data accuracy and realize real-time network perception. Based on the construction of a multi-agent simulation system, it realizes the correlation modeling of weather events and demand fluctuations. Based on the hypergraph model, a dynamic weight mechanism is introduced to improve the robustness of the algorithm. A hierarchical optimization framework is designed: integer programming in the upper layer is used to handle global optimization, and deep reinforcement learning in the lower layer is used to achieve dynamic adjustment. The intelligent insertion strategy of buffer time can improve the late arrival absorption ability, and a facility utility function model is established to improve the transfer efficiency.

[0183] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A transfer connection optimization method in multi - mode transportation integration planning, characterized in that, It includes the following steps: S1. Data collection and demand analysis: Build the dynamic data foundation of the multimodal transportation network and conduct demand modeling and analysis; S2. Topological design of the multimodal transfer network: Optimize the physical layout and functional coordination of transfer nodes; S3. Coordinated optimization of multimodal timetables: Reduce the waiting time for cross-modal transfers and improve the robustness of timetables; S4. Development of the transfer service system: Provide real-time transfer decision support; S5. Dynamic matching of facility capacity: Ensure that the capacity of transfer facilities is always adapted to the passenger flow demand in real time.

2. The transfer connection optimization method in a multi-mode transportation integration planning according to claim 1, characterized in that, Obtain multi-source data based on the transportation network. The multi-source data includes: 1) Static data: a. Transportation network topology: The line planning distribution, station location distribution, and transfer node location of different transportation modes; b. Timetable operation: The starting and ending times of departure of shifts, the location distribution of stops, and the headway duration; c. Facility attributes: Station capacity, parking lot capacity, and the location distribution of shared vehicle drop-off points; d. Transfer rules: The transfer distance and transfer time threshold of different transportation modes; 2) Dynamic data: a. Real-time traffic conditions: The traffic flow, driving speed and direction of different lines, and road congestion conditions; b. Passenger flow data: The passenger flow, population distribution and flow direction of different stations, and travel purposes; c. External factors: Weather conditions and traffic emergencies; d. User individual data: Travel trajectory preferences and transfer tendencies; The specific steps for obtaining multi-source data include: 1) Obtaining static data: a. Enter the data open platform of the urban traffic management department, download the GTFS dataset to obtain the timetables and line and station information under different transportation modes; b. Obtain the road network distribution, station coordinates, and spatial data of different transfer channels through OpenStreetMap; c. Obtain the location distribution of drop-off points and service areas of shared travel enterprises; 2) Obtaining dynamic data: a. Obtain the real-time positions of different means of transportation based on the vehicle-mounted positioning system; b. Obtain the passenger flow density data based on the cameras and WIFI probe modules set up on the platforms; c. Obtain the real-time traffic conditions based on the data interface of the transportation system and third-party map services; d. Record the user transfer query data based on the user terminal and analyze the starting and ending points and route distribution of user travel.

3. The transfer connection optimization method in a multi-modal transportation integration plan according to claim 2, characterized in that, Perform data processing based on the obtained static data and dynamic data, including: 1) Data cleaning: a. Eliminate outliers in the dynamic data based on GPS drift points. Import the pandas and matplotlib databases for data processing and visualization display; Load the GPS data through a CSV file and remove the invalid data containing null value rows; Calculate based on the distance and time difference, and set the distance threshold and time threshold to determine the drift point standard; Remove the drift points exceeding the set threshold range and conduct visual comparison; b. Calculate the average value of sensor data and fill in the missing sensor data according to the average interpolation method; c. Start the Arc workstation environment, enter the editing mode by inputting the ArcEdit command, load data using the ec+ command and the Coverage file path, display hanging lines based on the dearcnodedangle command, and repair errors according to nodecdangle and draw; 2) Data standardization: a. Standardize coordinate data in different maps based on a unified coordinate system; b. Standardize timetable data of different traffic modes based on timestamp alignment; 3) Feature extraction: a. Extract the station topological relationship and the length of transfer channels from static data; b. Calculate the real-time passenger flow density and vehicle delay duration from dynamic data.

4. The transfer connection optimization method in a multi - mode transportation integration planning according to claim 3, characterized in that, Based on the characteristics of the extracted static and dynamic data, accurately map the dynamic data to the static road network. By establishing the causal relationship and collaborative rules between dynamic and static data, a joint analysis framework is constructed, and a dynamic-static data fusion model is established to fuse and process static and dynamic data, specifically including: 1) Map matching: Match vehicle GPS trajectory points to the nearest road or track according to the Hidden Markov Model, specifically: a. For each GPS point o t Set to search for all road segments {r within a surrounding radius R t,i}, project the GPS point onto the nearest point of the candidate road segment, and record the coordinates of the projection point and the distance d to the road segment t,i ; b. Set the actual road segment r where the vehicle is located t,i In the hidden state, according to the GPS point o t Generated by the road segment r t,i Set the observation probability P(o t |r t,i ), and construct an observation model, expressed as: where σ is the standard deviation of the GPS error; c. Set the transition probability P(r t-1,i → r t,j ) according to the possibility of transferring from the previous road segment r t,j to the current road segment r t-1,i . Based on the path connectivity judgment, if r t,j is directly connected to r t-1,i , the probability is relatively high. Calculate the distance ΔS between the two projection points based on the motion consistency, compare it with the time difference Δt of the GPS points and the maximum vehicle speed v max , then there is where λ is an adjustment parameter; d. Determine the initial probability P(r1,i) of the first GPS point o1 directly through the observation probability P(o t |r t,i ), and calculate the maximum probability path of each candidate road segment at each moment according to the recurrence formula, expressed as: δ t (j) = max i [δ t-1 (i)·P(r t,j |r t-1,i )]·P(o t |r t,j ), where δ t (j) represents the maximum cumulative probability at time t on road segment j. By backtracking from the maximum probability state at the final time, the optimal road segment sequence is obtained to achieve map matching; 2) Spatiotemporal grid processing: Divide the city into a 500m×500m grid structure, aggregate dynamic and static data and import them into the grid structure for distribution; 3) Define nodes and edges based on static data, assign weights to edges using dynamic data, construct a transfer relationship graph, and combine static topology and dynamic features using a graph neural network to predict the propagation of network-wide demand; 4) Use static data as the system state and dynamic data as the observed value, update the state estimate through Kalman filtering, establish a state space model, capture both spatial topology and time series simultaneously, predict transfer demand within a future time period based on a spatio-temporal graph convolutional network, and integrate the static road network and dynamic data into a virtual simulation environment in combination with a digital twin platform to real-time simulate multi-modal traffic flow.

5. The transfer connection optimization method in a multi - mode transportation integration plan according to claim 1, characterized in that, Extract travel demand characteristics and model multi-modal demands, including: 1) Static demand modeling: Analyze periodic demands such as commuting and going to school based on historical data, extract station-level OD distribution, peak hours, and transfer hotspot data, and identify similar travel patterns based on cluster analysis; 2) Dynamic demand modeling: Predict the passenger flow at stations in the next 30 minutes based on the LSTM model for real-time passenger flow prediction, and combine event-driven demands to display temporarily increased demand data in real-time. Capture trends and periodicities based on time series analysis, and analyze the correlation between weather, events, and demand changes; 3) Demand decomposition and correlation: Distinguish the contributions of different traffic modes for mode splitting, establish a transfer selection probability model, consider factors such as walking time, waiting time, and comfort, and dynamically adjust weights; 4) Network-level demand modeling: Model the traffic network as a graph structure, use dynamic demand as node features to predict the propagation of network-wide demand, construct an objective function based on minimizing the network-wide travel time and transfer waiting time, and set constraint conditions according to vehicle capacity, timetable coordination, and resource scheduling upper limits to construct a collaborative optimization model.

6. The transfer connection optimization method in a multi-modal transportation integration plan according to claim 1, characterized in that, The steps of the multi-modal transfer network topology design specifically include: 1) Transfer node division and hierarchy construction: a. Node definition: Set the stations of a single mode of transportation as physical nodes, set the multi-mode intersection as the transfer hub, and merge multiple stations within a short distance into logical nodes for virtual processing; b. Hierarchical division: hierarchical processing according to traffic mode and function; 2) Transfer edge connection and weight calculation: a. Edge type design: set line connections within the same transportation mode as intra-mode edges, set transfer connections between different transportation modes as inter-mode edges, and set walking paths as virtual transfer edges;. b. Weight modeling: Time weight calculation is expressed as transfer time = walking time + waiting time. The transfer fee is set as the economic weight. The entropy weight method is used to integrate multiple indicators and calculate the comprehensive weight. The formula is expressed as follows: W ij = αT ij + βC ij + γ(1 - S ij ) where T ij is time, C ij is cost, S ij is comfort level (0 - 1), and α, β, γ are weight coefficients; 3) Network topology model construction: establish a hypergraph model, make a topology table based on the adjacency table, calculate indicators such as degree distribution, clustering coefficient, average path length, and analyze complex network characteristics; 4) Transfer optimization and dynamic adjustment: a. Static optimization: Solve the optimal transfer path through the Dijkstra algorithm, analyze the critical path, and simulate the redundant path design under node failure to enhance robustness; b. Dynamic adjustment: Dynamically adjust the weight of the edge according to the passenger flow data, predict future passenger flow changes based on the LSTM network, dynamically add or delete nodes or edges, and realize topological evolution.

7. The transfer connection optimization method in a multi-mode transportation integration planning according to claim 1, characterized in that, The steps of collaborative optimization of the multi-mode schedule specifically include: 1) Build a collaborative optimization model: a. Design the objective function based on minimizing the average transfer waiting time of the entire network, and balance time efficiency and operating costs through Pareto optimization to make multi-objective trade-offs, expressed as: wherein is the transfer waiting time, and P ij is the transfer flow, and the fluctuation range of the departure interval is constrained to maximize the robustness of the timetable; b. Set the vehicle turnover time ≤ t min With the driver's working time being less than the legal upper limit as the physical constraint condition, set the effective transfer time window as the transfer constraint condition, and satisfy: where Δt tolerance is the waiting time tolerance of passengers; 3) Algorithm optimization: a. Static collaborative optimization: discretize the timetable into time slots based on integer programming, use the departure time as the decision variable, and solve the global optimal solution; b. Dynamic real-time adjustment: Define the state space, action space, and reward function, and train the agent to dynamically adjust the schedule based on the PPO algorithm. c. Rolling time domain optimization: Resolve the timetable for the next hour every 15 minutes and update it in combination with real-time data; 4) Transfer time window synchronization: a. Identification of key transfer points: Identify high-traffic transfer hubs through complex network centrality indicators (such as betweenness centrality), and use the DBSCAN algorithm to find high-frequency transfer OD pairs and time periods. b. Schedule shift strategy: adjust the departure time of secondary transportation modes so that their arrival time matches the arrival time of primary transportation modes; c. Buffer time insertion: reserve time buffer at key transfer nodes to cope with delays, expressed as: t buffer = μ delay + kσ delay , where μ delay is the average delay and k is the confidence coefficient.

8. The transfer connection optimization method in a multi-modal transportation integration plan according to claim 1, characterized in that, The transfer service system includes: a knowledge graph module, which integrates multi-source data by constructing a knowledge graph; Dynamic route recommendation module, which dynamically generates the optimal transfer plan based on reinforcement learning; User portrait module, providing differentiated recommendations based on different user situations; and an integrated service platform; The dynamic matching of facility capacity includes: 1) Construction of dynamic demand-resource matching model: Setting the minimization of passenger waiting time and facility congestion cost as the objective function, setting the physical facility capacity limit and capacity scheduling feasibility as constraints, expressed as: Among them is the waiting time at transfer point i, is the crowding penalty; 2) Matching strategy classification: Schedule logistics vehicle replenishment in real time according to outbound passenger flow, dynamically adjust the placement of shared bicycles, increase flexible bus schedules during peak transfer hours to achieve short-term response, and achieve long-term optimization by reallocating parking lot resources and adjusting subway formation; 3) Design of dynamic matching algorithm: a. Optimization of transport capacity scheduling: Dynamic vehicle scheduling (DVRP) based on real-time demand, using reinforcement learning (such as DQN) or heuristic algorithms for variants of the vehicle routing problem, adjusting the bus departure interval based on queuing theory to achieve coordinated adjustment of shifts, expressed as: where λ is the arrival rate and μ is the service rate; b. Dynamic allocation of facility resources: Use mixed integer programming (MIP) to optimize the layout of transfer node facilities and achieve spatial reconfiguration, expressed as: where x ij is the allocation quantity of facility type j at node i, and u ij is the utility value; 4) Edge computing and rapid response: a. Edge node deployment: Deploy edge servers at transfer hubs to process local data in real time; b. Control strategy: Trigger actions based on a rule engine and use fuzzy logic to handle uncertainties; 5) Closed-loop feedback optimization: a. Digital twin simulation: Rehearse scheduling strategies in a virtual environment and implement them specifically after evaluating the effects; b. Reinforcement learning online training: Feed actual operation data back into the model to continuously optimize the scheduling strategy.

Images