A spatiotemporal prediction fused network car supply and demand matching and vehicle scheduling system and method

Abstract

The application relates to the technical field of dispatch management, and discloses a network car-hailing supply-demand matching and vehicle dispatching system and method fusing space-time prediction, which constructs a space-time demand-driver strategic behavior joint prediction model, synchronously incorporates strategic behavior features such as driver order rejection and hotspot gathering into a space-time prediction framework, corrects the prediction results from the supply and demand sides, and reduces the deviation between the predicted situation and the actual operation state from the root cause. Through real-time sensing and quantification of the dispatch compliance degree and order acceptance willingness of the driver, the supply-demand matching weight is dynamically corrected, and a differentiated transport capacity guiding strategy is generated, the individual income appeal of the driver is compatible to the global dispatching target of the network car-hailing dispatching platform, and the optimization effect decay caused by the game is reduced. Meanwhile, the differentiated dispatching guidance can effectively relieve the driver's herd effect and hotspot clustering phenomenon, and realize the game balance between the operation efficiency of the network car-hailing dispatching platform and the individual income of the driver.

Description

Technical Field

[0001] This invention relates to the technical field of dispatch management, specifically to a ride-hailing supply and demand matching and vehicle dispatching system and method that integrates spatiotemporal prediction. Background Technology

[0002] With the rapid development of the ride-sharing industry, ride-hailing dispatch technology has evolved from the traditional nearest-location dispatch model to an intelligent predictive dispatch model. However, the existing ride-hailing dispatch system only considers passenger travel demand, failing to incorporate driver-side decision-making into the prediction dimension. This results in a systematic discrepancy between the predicted supply and demand and the actual capacity distribution in operation, leading to adaptability defects in dispatch plans from the generation stage. Drivers, as individuals with independent decision-making capabilities, commonly exhibit strategic behaviors such as selectively rejecting orders, concentrating on high-premium hotspot areas, and selectively executing capacity allocation instructions to maximize individual profits. These strategic behaviors of drivers and... The inherent game theory inherent in the platform's global operational optimization goals directly leads to a low actual execution rate of traditional globally optimal scheduling schemes, with the theoretical optimization effect continuously diminishing during the game process. Simultaneously, homogeneous scheduling strategies easily exacerbate the herding effect among drivers, causing a tidal supply-demand imbalance characterized by "excess capacity in hot spots and insufficient supply in cold spots." Furthermore, drivers' strategic behavior is dynamically evolving and exhibits group concealment. The current ride-hailing scheduling system, based on static rule settings, lacks real-time perception of driver behavior, game theory adaptation, and closed-loop iteration mechanisms. This results in insufficient robustness of the algorithm in long-term operation, ultimately hindering improvements in platform operational efficiency, passenger travel experience, and driver income stability.

[0003] Chinese invention patent application CN120471661A, published on August 12, 2025, discloses a method for automatically adjusting the dispatch radius of ride-hailing services in real time by analyzing supply and demand. This method utilizes the GPS functions of mobile passenger and driver terminals to obtain the current latitude and longitude information of passengers in real time and updates the current latitude and longitude information of drivers. It also uses a tenant SaaS platform to divide cities, business districts, and specific fences to configure the minimum and maximum allowable dispatch radius within each business district. Furthermore, it uses the GPS functions of mobile passenger and driver terminals to obtain the latitude and longitude information of passengers and drivers in real time, allowing tenants to... The city is divided into different business districts based on its latitude, longitude, area, and transportation conditions. A regular hexagonal honeycomb pattern is used for regional division to effectively cover the entire area. A dynamic adjustment mechanism based on the supply-demand ratio parameter effectively balances the supply and demand relationship. The Havesing formula is used to calculate the distance between the two parties to accurately determine whether the passenger is within the current dispatch radius, reducing the time and distance wasted between passengers and drivers. However, the above technical solution does not take into account the strategic behaviors of ride-hailing drivers, such as picking and refusing orders, gathering in high-premium hotspot areas, and selectively executing capacity allocation instructions, which reduces the efficiency and quality of ride-hailing dispatch. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing ride-hailing dispatch systems, which only consider passenger demand and fail to incorporate driver decision-making into the forecasting process, resulting in a systematic discrepancy between predicted supply and demand and actual capacity distribution, dispatching schemes suffer from adaptability defects from the generation stage. Drivers, as individuals with independent decision-making capabilities, commonly exhibit strategic behaviors such as selectively rejecting orders, clustering in high-premium areas, and selectively executing capacity allocation instructions to maximize individual profits. This strategic behavior of drivers inherently interacts with the platform's overall operational optimization goals, directly leading to a low actual execution rate of traditional globally optimal dispatching schemes, and the theoretical optimization effect continuously diminishes during the game. Furthermore, homogeneous dispatching strategies easily exacerbate the herding effect among drivers. This leads to a tidal imbalance between supply and demand, characterized by "excess capacity in hot spots and insufficient supply in cold spots." Simultaneously, drivers' strategic behavior is dynamically evolving and exhibits group concealment. Current ride-hailing dispatch systems, based on static rules, lack real-time perception of driver behavior, game-theoretic adaptation, and closed-loop iteration mechanisms. This results in insufficient robustness of the algorithm over the long term, ultimately hindering improvements in platform operational efficiency, passenger travel experience, and driver income stability. The solution aims to correct situational deviations through joint supply and demand forecasting, improve instruction execution rates by quantifying driver behavior and employing differentiated game-theoretic dispatching, and effectively mitigate supply-demand imbalances and reduce vehicle empty-running rates through closed-loop iterative adaptive strategy evolution. This approach balances platform operational efficiency and individual driver income, ultimately enhancing the long-term robustness of the ride-hailing dispatch system.

[0005] (II) Technical Solution This invention is achieved through the following technical solution: a method for ride-hailing supply and demand matching and vehicle dispatching that integrates spatiotemporal prediction, the method comprising the following: S1: Construct a joint prediction model of spatiotemporal demand and driver strategic behavior. Based on historical order data, road network status data and driver historical behavior data, it synchronously outputs the spatiotemporal distribution prediction results of passenger demand and driver strategic behavior within the target scheduling area. S2: Based on the predicted spatiotemporal distribution of passenger demand and the predicted spatiotemporal distribution of driver strategic behavior, generate an initial global supply and demand matching scheme and an idle capacity rebalancing scheduling scheme. S3: Real-time collection of driver terminal operation status data, identification of drivers' strategic behaviors such as picking orders, hot spot clustering, and selective execution of dispatch instructions, and quantitative generation of dispatch compliance coefficient and order acceptance willingness coefficient for each driver; S4: Based on the scheduling compliance coefficient and the order acceptance willingness coefficient, dynamically adjust the matching weight factor of the initial global supply and demand matching scheme and the guidance strategy of the idle capacity rebalancing scheduling scheme to generate a game equilibrium scheduling scheme that takes into account both the overall operational efficiency of the platform and the individual income of drivers. S5: The game equilibrium scheduling scheme is sent to the corresponding driver terminal and passenger terminal, and the parameters and scheduling strategy threshold of the spatiotemporal demand-driver strategic behavior joint prediction model are updated in reverse based on the actual order execution results, forming a prediction-scheduling-feedback closed-loop optimization mechanism.

[0006] Preferably, the specific process of constructing the spatiotemporal demand-driver strategic behavior joint prediction model is as follows: Using the spatiotemporal grid of the target scheduling area as the basic unit, the order volume, passenger waiting time, and driver empty-run rate in historical time periods are extracted as demand-side features, while the driver rejection rate, hotspot area deviation, dispatch instruction response delay, and order-acceptance distance preference are extracted as driver strategic behavior features. Combined with external environmental features such as weather, holidays, and large-scale events, a spatiotemporal graph convolutional network model is trained to simultaneously output the passenger demand and driver strategic behavior tendency values ​​of each spatiotemporal grid in the future preset time period.

[0007] Preferably, the specific process of generating the initial global supply and demand matching scheme and the idle capacity rebalancing scheduling scheme is as follows: Based on the predicted spatiotemporal distribution of passenger demand and the current online capacity distribution, a driver-passenger bipartite graph matching model is constructed with the goal of minimizing the average passenger waiting time and vehicle empty mileage, and the initial global supply and demand matching scheme is obtained by solving the model. At the same time, based on the predicted supply and demand gap values ​​of each spatiotemporal grid, an initial rebalancing scheduling scheme for allocating idle vehicles to the gap area is generated.

[0008] Preferably, the specific process of collecting real-time driver terminal operating status data, identifying drivers' strategic behaviors such as picking and choosing orders, hotspot clustering, and selectively executing dispatch instructions, and quantifying and generating dispatch compliance coefficients and order acceptance willingness coefficients for each driver is as follows: Real-time data collection includes driver order rejection records, driving trajectory deviation data, dispatch instruction response time, and order acceptance destination preference data; calculation of the order acceptance willingness coefficient based on the order rejection frequency within a preset statistical period, calculation of the dispatch compliance coefficient based on the execution ratio and response time of dispatch instructions, and calculation of the hotspot clustering tendency value based on the overlap between vehicle trajectory and hotspot areas.

[0009] Preferably, the specific process of dynamically correcting the matching weight factor of the initial global supply and demand matching scheme is as follows: For drivers whose willingness to accept orders is below a preset threshold, increase the matching weight of high-yield orders and short-distance driving orders in the matching queue; for drivers whose dispatch compliance coefficient is above a preset threshold, give priority to matching high-quality orders and increase the weight of dispatch incentive subsidies; for drivers with excessively high hotspot clustering tendency, reduce their order matching priority in hotspot areas and guide transportation capacity to be dispersed to colder areas.

[0010] Preferably, the specific process of generating a game-theoretic equilibrium scheduling scheme that balances the overall operational efficiency of the platform with the individual benefits of drivers is as follows: Based on the idle capacity rebalancing scheduling scheme, differentiated scheduling instructions are generated for drivers with different strategic behavior characteristics: direct allocation instructions are issued to drivers with high compliance, guided scheduling instructions with income subsidies are issued to drivers with medium compliance, and indirect guided instructions based on order push are issued to drivers with low compliance. By using differentiated instructions, capacity clustering is avoided, and a game steady state under global supply and demand equilibrium is achieved.

[0011] Preferably, the specific process for forming the prediction-scheduling-feedback closed-loop optimization mechanism is as follows: The system statistically analyzes the actual order completion rate, driver execution rate, passenger waiting time, and empty-run rate for each round of scheduling. The deviation between the actual execution data and the predicted data is used as a loss term to iteratively update the weight parameters of the spatiotemporal demand-driver strategic behavior joint prediction model. Simultaneously, based on the evolution trend of driver group strategic behavior, the matching weight threshold and incentive subsidy coefficient are dynamically adjusted to achieve adaptive evolution of the scheduling strategy.

[0012] Preferably, the method is also adaptable to multi-category heterogeneous transportation capacity scenarios: For different categories of transportation services such as express cars, premium cars, and carpooling, corresponding strategic behavior quantification systems and matching weight rules are constructed. In the carpooling scenario, the driver's willingness to carpool is incorporated into the route planning and order splicing constraints to improve the acceptance rate of carpooling orders.

[0013] A ride-hailing supply-demand matching and vehicle dispatching system integrating spatiotemporal prediction is provided for executing the aforementioned ride-hailing supply-demand matching and vehicle dispatching method integrating spatiotemporal prediction. The system includes: Spatiotemporal demand-driver strategic behavior joint prediction module: Constructs a spatiotemporal demand-driver strategic behavior joint prediction model, based on historical order data, road network status data and driver historical behavior data, and synchronously outputs the spatiotemporal distribution prediction results of passenger demand and driver strategic behavior within the target scheduling area. Initial supply and demand matching and capacity scheduling generation module: Based on the predicted spatiotemporal distribution of passenger demand and the predicted spatiotemporal distribution of driver strategic behavior, it generates an initial global supply and demand matching scheme and an idle capacity rebalancing scheduling scheme. Driver strategic behavior perception and quantification module: collects driver terminal operation status data in real time, identifies drivers' strategic behaviors such as picking orders, hot spot clustering, and selectively executing dispatch instructions, and quantifies and generates dispatch compliance coefficient and order acceptance willingness coefficient for each driver; Supply and demand game equilibrium scheduling module: Based on the scheduling compliance coefficient and the order acceptance willingness coefficient, dynamically adjust the matching weight factor of the initial global supply and demand matching scheme and the guidance strategy of the idle capacity rebalancing scheduling scheme to generate a game equilibrium scheduling scheme that takes into account both the overall operational efficiency of the platform and the individual income of drivers. The scheduling closed-loop optimization module distributes the game equilibrium scheduling scheme to the corresponding driver and passenger terminals, and updates the parameters and scheduling strategy thresholds of the spatiotemporal demand-driver strategic behavior joint prediction model based on the actual order execution results, forming a prediction-scheduling-feedback full closed-loop optimization mechanism.

[0014] (III) Beneficial Effects This invention provides a ride-hailing supply-demand matching and vehicle dispatching system and method that integrates spatiotemporal prediction. It has the following beneficial effects: First, by constructing a spatiotemporal demand-driver strategic behavior joint prediction model, the strategic behavior characteristics of drivers, such as order rejection and hotspot clustering, are simultaneously incorporated into the spatiotemporal prediction framework along with passenger demand characteristics. This corrects the prediction results from both the supply and demand sides, fundamentally reducing the deviation between the predicted situation and the actual operating status. It solves the problem of inaccurate traditional single-dimensional demand prediction, provides a more realistic situational basis for subsequent dispatch decisions, and effectively improves the basic feasibility of ride-hailing dispatch solutions.

[0015] Second, by real-time sensing and quantification of drivers' dispatch compliance and willingness to accept orders, the supply and demand matching weights are dynamically adjusted and differentiated capacity guidance strategies are generated. This integrates individual driver income demands into the overall dispatch objectives of the ride-hailing dispatch platform, significantly improving the actual execution rate of ride-hailing dispatch instructions and reducing the attenuation of optimization effects caused by game theory. At the same time, differentiated dispatch guidance can effectively alleviate the herding effect and hot spot clustering phenomenon among drivers, smooth out the tidal imbalance of supply and demand, and achieve a game equilibrium between the operational efficiency of the ride-hailing dispatch platform and the individual income of drivers.

[0016] Third, by establishing a closed-loop optimization mechanism of prediction-scheduling-feedback, the prediction model parameters and scheduling strategy thresholds are updated iteratively based on actual order execution data. This enables the algorithm to adapt to the long-term evolution of the strategic behavior of the driver group and continuously improve the robustness and operational accuracy of the scheduling algorithm. At the same time, the solution supports the adaptation of multi-category heterogeneous transportation capacity scenarios, covering diverse operational needs such as express cars, premium cars, and carpooling. It has strong engineering applicability and can effectively reduce the operating costs of ride-hailing dispatch platforms and improve the quality of passenger travel services. Attached Figure Description

[0017] Figure 1 A flowchart of a ride-hailing supply and demand matching and vehicle scheduling method that integrates spatiotemporal prediction provided by the present invention; Figure 2This invention provides a schematic diagram of a ride-hailing supply and demand matching and vehicle dispatching system that integrates spatiotemporal prediction. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] An embodiment of a ride-hailing supply-demand matching and vehicle dispatching system and method that integrates spatiotemporal prediction is as follows: Example 1: Please see Figure 1 A method for ride-hailing supply and demand matching and vehicle dispatching that integrates spatiotemporal prediction, the method comprising the following: S1: Construct a joint prediction model of spatiotemporal demand and driver strategic behavior. Based on historical order data, road network status data and driver historical behavior data, it synchronously outputs the spatiotemporal distribution prediction results of passenger demand and driver strategic behavior within the target scheduling area. The specific process for constructing the spatiotemporal demand-driver strategic behavior joint prediction model is as follows: Using the spatiotemporal grid of the target scheduling area as the basic unit, the order volume, passenger waiting time, and driver empty-run rate in historical time periods are extracted as demand-side features, while the driver rejection rate, hotspot area deviation, dispatch instruction response delay, and order-acceptance distance preference are extracted as driver strategic behavior features. Combined with external environmental features such as weather, holidays, and large-scale events, a spatiotemporal graph convolutional network model is trained to simultaneously output the passenger demand and driver strategic behavior tendency values ​​of each spatiotemporal grid in the future preset time period.

[0020] The hardware and software used in this step include: the data acquisition end connects to the platform's historical order database, third-party map road network service interface, meteorological data service interface, and large-scale event information database; the computing end uses a cloud-based GPU server cluster to deploy a spatiotemporal graph convolutional network training and inference engine based on the PyTorch framework.

[0021] The specific processing steps are as follows: The target city for dispatch is divided into 500m×500m spatial grid units, and a spatiotemporal grid sequence is constructed with a time granularity of 5 minutes. The total number of orders, average passenger waiting time, and average driver empty-run rate of each spatiotemporal grid over the past 90 days are extracted as the demand-side feature set. The average driver rejection rate, percentage of stay in hotspot areas, average response time of dispatch instructions, and average order-acceptance distance threshold of each grid within the corresponding time period are extracted as the driver strategic behavior feature set. At the same time, external features such as weather level, holiday attributes, and surrounding large-scale events are matched as auxiliary inputs. The above multi-source features are input into the spatiotemporal graph convolutional network model, and end-to-end training is performed using a multi-task loss function (mean squared error of demand prediction + cross-entropy loss of strategic behavior prediction) until the loss function converges.

[0022] The output of this step is as follows: every 5 minutes, the predicted passenger demand, average driver rejection tendency, and dispatch compliance rate for each spatiotemporal grid will be output for the next 30 minutes, forming a complete spatiotemporal distribution prediction result for both supply and demand.

[0023] S2: Based on the predicted spatiotemporal distribution of passenger demand and the predicted spatiotemporal distribution of driver strategic behavior, generate an initial global supply and demand matching scheme and an idle capacity rebalancing scheduling scheme. The specific process for generating the initial global supply and demand matching scheme and the idle capacity rebalancing scheduling scheme is as follows: Based on the predicted spatiotemporal distribution of passenger demand and the current online capacity distribution, a driver-passenger bipartite graph matching model is constructed with the goal of minimizing the average passenger waiting time and vehicle empty mileage, and the initial global supply and demand matching scheme is obtained by solving the model. At the same time, based on the predicted supply and demand gap values ​​of each spatiotemporal grid, an initial rebalancing scheduling scheme for allocating idle vehicles to the gap area is generated.

[0024] The hardware and software used in this step include: a real-time order pool and an online capacity status database, a cloud-based high-performance computing server, and the deployment of an improved Hungarian algorithm solver and a capacity route planning engine.

[0025] The specific processing steps are as follows: First, the current list of pending orders and the list of online available vehicles are retrieved in real time. A driver-passenger bipartite graph is constructed based on the pick-up distance, estimated waiting time, and estimated order revenue as the foundational edge weights. Then, combining the driver strategic behavior tendency values ​​for each region output from step S1, a revenue correction coefficient is added to the matching edge weights of regions with high order rejection tendency. An improved Hungarian algorithm is used to solve for the minimum weight perfect matching in the bipartite graph, generating an initial global supply-demand matching scheme. Simultaneously, based on the predicted supply-demand gap (demand - estimated available capacity) for each spatiotemporal grid over the next 30 minutes, high-demand cold spots and overcapacity hot spots are identified. Combined with the estimated driver compliance in each region, the available capacity level is calculated, generating an initial idle capacity rebalancing scheduling scheme that includes the target allocation area and suggested arrival time.

[0026] The output of this step includes: the initial supply and demand matching scheme for each order's corresponding vehicle, and the initial capacity rebalancing scheduling scheme including the target area and the range of vehicles to be allocated.

[0027] By constructing a spatiotemporal demand-driver strategic behavior joint prediction model, strategic behavioral characteristics such as driver order rejection and hotspot clustering are simultaneously incorporated into the spatiotemporal prediction framework along with passenger demand characteristics. This corrects the prediction results from both the supply and demand sides, fundamentally reducing the deviation between the predicted situation and the actual operational status. It solves the problem of inaccurate traditional single-dimensional demand prediction, provides a more realistic situational basis for subsequent dispatch decisions, and effectively improves the basic feasibility of ride-hailing dispatch solutions.

[0028] S3: Real-time collection of driver terminal operation status data, identification of drivers' strategic behaviors such as picking orders, hot spot clustering, and selective execution of dispatch instructions, and quantitative generation of dispatch compliance coefficient and order acceptance willingness coefficient for each driver; The specific process of collecting real-time driver terminal operation status data, identifying drivers' strategic behaviors such as picking and choosing orders, hotspot clustering, and selectively executing dispatch instructions, and quantifying and generating dispatch compliance coefficients and order acceptance willingness coefficients for each driver is as follows: Real-time data collection includes driver order rejection records, driving trajectory deviation data, dispatch instruction response time, and order acceptance destination preference data; calculation of order acceptance willingness coefficient based on order rejection frequency within a preset statistical period, calculation of dispatch compliance coefficient based on dispatch instruction execution ratio and response time, and calculation of hotspot clustering tendency value based on vehicle trajectory and hotspot area overlap.

[0029] The hardware and software used in this step include: driver-side smartphone terminal (with built-in GPS positioning module, cellular network communication module, and driver-side operation APP), platform-side Flink real-time streaming computing engine, and driver behavior feature database.

[0030] The specific processing steps are as follows: The driver's mobile app transmits vehicle location data, order acceptance / rejection records, and dispatch instruction reception and response data in real time, with a data transmission frequency of once per second. The platform uses a 1-hour sliding window to stream-count the behavioral data of each driver: The total number of orders dispatched and rejected within the window is counted, the rejection rate is calculated and normalized to a 0-1 range to obtain an order acceptance willingness coefficient (a higher value indicates a stronger willingness to accept orders); the number of dispatch instructions issued and actually executed within the window is counted, weighted by the average response time of the instructions, and normalized to obtain a dispatch compliance coefficient (a higher value indicates stronger compliance); the percentage of time a vehicle spends in platform-marked hotspot areas within the window is counted, combined with the frequency of trajectories actively heading towards hotspots, to calculate a hotspot clustering tendency value (a higher value indicates a stronger tendency to cluster).

[0031] The output of this step is as follows: three quantitative indicators for each online driver: willingness to accept orders coefficient, dispatch compliance coefficient, and hotspot clustering tendency value, which are updated in real time every 5 minutes.

[0032] S4: Based on the scheduling compliance coefficient and the order acceptance willingness coefficient, dynamically adjust the matching weight factor of the initial global supply and demand matching scheme and the guidance strategy of the idle capacity rebalancing scheduling scheme to generate a game equilibrium scheduling scheme that takes into account both the overall operational efficiency of the platform and the individual income of drivers. The specific process of dynamically correcting the matching weight factor of the initial global supply and demand matching scheme is as follows: For drivers whose willingness to accept orders is below a preset threshold, increase the matching weight of high-yield orders and short-distance driving orders in the matching queue; for drivers whose dispatch compliance coefficient is above a preset threshold, give priority to matching high-quality orders and increase the weight of dispatch incentive subsidies; for drivers with excessively high hotspot clustering tendency, reduce their order matching priority in hotspot areas and guide transportation capacity to be dispersed to colder areas.

[0033] The specific process for generating a game-theoretic equilibrium scheduling scheme that balances the overall operational efficiency of the platform with the individual benefits of drivers is as follows: Based on the idle capacity rebalancing scheduling scheme, differentiated scheduling instructions are generated for drivers with different strategic behavior characteristics: direct allocation instructions are issued to drivers with high compliance, guided scheduling instructions with income subsidies are issued to drivers with medium compliance, and indirect guided instructions based on order push are issued to drivers with low compliance. By using differentiated instructions, capacity clustering is avoided, and a game steady state under global supply and demand equilibrium is achieved.

[0034] The method is also suitable for multi-category heterogeneous transportation capacity scenarios: For different categories of transportation services such as express cars, premium cars, and carpooling, corresponding strategic behavior quantification systems and matching weight rules are constructed. In the carpooling scenario, the driver's willingness to carpool is incorporated into the route planning and order splicing constraints to improve the acceptance rate of carpooling orders.

[0035] The hardware and software used in this step include: a cloud-based scheduling decision server, a matching weight rule engine, and a scheduling instruction distribution gateway.

[0036] The specific processing procedure is divided into two parts: matching weight correction and capacity scheduling optimization. P1. Matching Weight Adjustment: The preset thresholds for order acceptance willingness are 0.5, dispatch compliance is 0.8, and hotspot clustering tendency is 0.7. For drivers with an order acceptance willingness coefficient below 0.5, the weight of order revenue in the matching ranking is increased from 30% to 60%, and the weight of pick-up distance is increased from 40% to 50%, prioritizing short-distance, high-revenue orders. For drivers with a dispatch compliance coefficient above 0.8, they are included in the high-quality dispatch whitelist, given priority matching rights for high-quality orders during peak hours, and given an additional 0.1 times the dispatch service score reward weight. For drivers with a hotspot clustering tendency value above 0.7, their order matching priority in the current hotspot area is downgraded by 2 levels, while the push weight of orders in the surrounding 3-kilometer cold spots is increased, achieving natural diversion of transportation capacity through order guidance.

[0037] P2. Capacity Dispatch Optimization: Based on the initial rebalancing dispatch scheme, drivers are divided into three levels according to their dispatch compliance coefficient: high compliance (≥0.8), medium compliance (0.5-0.8), and low compliance (<0.5). Drivers with high compliance are directly issued clear dispatch target areas and arrival time limits, plus a basic dispatch subsidy; drivers with medium compliance are issued guided dispatch instructions with tiered subsidies, specifying the guaranteed order revenue after reaching the target area, thus incentivizing them to actively go there; drivers with low compliance are not issued mandatory dispatch instructions, but instead continuously pushed high-return pre-orders to the target area, indirectly guiding capacity through order attraction.

[0038] For multi-category scenarios, differentiated quantitative thresholds and weight benchmarks are set for the three categories of express car, premium car, and carpooling: the premium car category increases the weight of service score and sets the order acceptance willingness threshold at 0.6; the express car category focuses on dispatch efficiency and sets the compliance threshold at 0.7; the carpooling category additionally calculates the driver's carpooling willingness coefficient, and reduces the proportion of multi-order carpooling for drivers with a carpooling willingness below 0.5 to reduce the order cancellation rate.

[0039] The output of this step is: the revised final driver-passenger matching scheme and the hierarchical differentiated capacity rebalancing scheduling instruction together constitute a game-theoretic equilibrium scheduling scheme.

[0040] By real-time sensing and quantification of drivers' dispatch compliance and willingness to accept orders, the system dynamically adjusts the supply and demand matching weights and generates differentiated capacity guidance strategies. This integrates individual driver benefit demands into the overall dispatch objectives of the ride-hailing dispatch platform, significantly improving the actual execution rate of ride-hailing dispatch instructions and reducing the attenuation of optimization effects caused by game theory. At the same time, differentiated dispatch guidance can effectively alleviate the herding effect and hotspot clustering among drivers, smooth out the tidal imbalance between supply and demand, and achieve a game equilibrium between the operational efficiency of the ride-hailing dispatch platform and the individual driver benefits.

[0041] S5: The game equilibrium scheduling scheme is sent to the corresponding driver terminal and passenger terminal, and the parameters and scheduling strategy threshold of the spatiotemporal demand-driver strategic behavior joint prediction model are updated in reverse based on the actual order execution results, forming a prediction-scheduling-feedback closed-loop optimization mechanism.

[0042] The specific process for forming the prediction-scheduling-feedback closed-loop optimization mechanism is as follows: The system statistically analyzes the actual order completion rate, driver execution rate, passenger waiting time, and empty-run rate for each round of scheduling. The deviation between the actual execution data and the predicted data is used as a loss term to iteratively update the weight parameters of the spatiotemporal demand-driver strategic behavior joint prediction model. Simultaneously, based on the evolution trend of driver group strategic behavior, the matching weight threshold and incentive subsidy coefficient are dynamically adjusted to achieve adaptive evolution of the scheduling strategy.

[0043] The hardware and software used in this step include: a scheduling instruction distribution gateway, an order execution status database, a model iteration training server, and a strategy parameter configuration center.

[0044] The specific processing procedure is as follows: The game equilibrium scheduling scheme is distributed to the corresponding driver and passenger apps through the command gateway, and the order lifecycle status is tracked in real time. After each scheduling cycle (30 minutes), key indicators such as order completion rate, driver dispatch command execution rate, average passenger waiting time, and average vehicle empty-run rate are statistically analyzed. The deviation between the actual order volume, driver behavior data and the prediction results of S1 is calculated, and the deviation value is used as an incremental training sample to input into the spatiotemporal demand-driver strategic behavior joint prediction model. The parameters of the model's convolutional and fully connected layers are updated through backpropagation to achieve rolling iterative optimization of the model. At the same time, the command execution effect of drivers under different levels is statistically analyzed. If the execution rate of a certain level is lower than the preset target, the matching weight coefficient and subsidy standard of that level are dynamically adjusted. For example, if the execution rate of drivers with medium compliance is low, the subsidy amount is appropriately increased until the execution rate returns to the target range.

[0045] The output of this step is: the updated joint prediction model parameters and scheduling policy thresholds, realizing the self-optimization closed loop of the scheduling system.

[0046] By establishing a closed-loop optimization mechanism of prediction-scheduling-feedback, and iteratively updating the prediction model parameters and scheduling strategy thresholds based on actual order execution data, the solution can adapt to the long-term evolution of the strategic behavior of the driver group, continuously improving the robustness and operational accuracy of the scheduling algorithm. At the same time, the solution supports adaptation to multiple heterogeneous transportation capacity scenarios, covering diverse operational needs such as express cars, premium cars, and carpooling. It has strong engineering applicability and can effectively reduce the operating costs of ride-hailing dispatch platforms and improve the quality of passenger travel services.

[0047] Example 2: Please see Figure 2 A ride-hailing supply-demand matching and vehicle dispatching system integrating spatiotemporal prediction is provided for executing the aforementioned ride-hailing supply-demand matching and vehicle dispatching method integrating spatiotemporal prediction. The system includes: Spatiotemporal demand-driver strategic behavior joint prediction module: Constructs a spatiotemporal demand-driver strategic behavior joint prediction model, based on historical order data, road network status data and driver historical behavior data, and synchronously outputs the spatiotemporal distribution prediction results of passenger demand and driver strategic behavior within the target scheduling area. Initial supply and demand matching and capacity scheduling generation module: Based on the predicted spatiotemporal distribution of passenger demand and the predicted spatiotemporal distribution of driver strategic behavior, it generates an initial global supply and demand matching scheme and an idle capacity rebalancing scheduling scheme. Driver strategic behavior perception and quantification module: collects driver terminal operation status data in real time, identifies drivers' strategic behaviors such as picking orders, hot spot clustering, and selectively executing dispatch instructions, and quantifies and generates dispatch compliance coefficient and order acceptance willingness coefficient for each driver; Supply and demand game equilibrium scheduling module: Based on the scheduling compliance coefficient and the order acceptance willingness coefficient, dynamically adjust the matching weight factor of the initial global supply and demand matching scheme and the guidance strategy of the idle capacity rebalancing scheduling scheme to generate a game equilibrium scheduling scheme that takes into account both the overall operational efficiency of the platform and the individual income of drivers. The scheduling closed-loop optimization module distributes the game equilibrium scheduling scheme to the corresponding driver and passenger terminals, and updates the parameters and scheduling strategy thresholds of the spatiotemporal demand-driver strategic behavior joint prediction model based on the actual order execution results, forming a prediction-scheduling-feedback full closed-loop optimization mechanism.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for ride-hailing supply and demand matching and vehicle dispatching that integrates spatiotemporal prediction, characterized in that, The method includes the following: S1: Construct a joint prediction model of spatiotemporal demand and driver strategic behavior. Based on historical order data, road network status data and driver historical behavior data, it synchronously outputs the spatiotemporal distribution prediction results of passenger demand and driver strategic behavior within the target scheduling area. S2: Based on the predicted spatiotemporal distribution of passenger demand and the predicted spatiotemporal distribution of driver strategic behavior, generate an initial global supply and demand matching scheme and an idle capacity rebalancing scheduling scheme. S3: Real-time collection of driver terminal operation status data, identification of drivers' strategic behaviors such as picking orders, hot spot clustering, and selective execution of dispatch instructions, and quantitative generation of dispatch compliance coefficient and order acceptance willingness coefficient for each driver; S4: Based on the scheduling compliance coefficient and the order acceptance willingness coefficient, dynamically adjust the matching weight factor of the initial global supply and demand matching scheme and the guidance strategy of the idle capacity rebalancing scheduling scheme to generate a game equilibrium scheduling scheme that takes into account both the overall operational efficiency of the platform and the individual income of drivers. S5: The game equilibrium scheduling scheme is sent to the corresponding driver terminal and passenger terminal, and the parameters and scheduling strategy threshold of the spatiotemporal demand-driver strategic behavior joint prediction model are updated in reverse based on the actual order execution results, forming a prediction-scheduling-feedback closed-loop optimization mechanism.

2. The ride-hailing supply and demand matching and vehicle dispatching method integrating spatiotemporal prediction as described in claim 1, characterized in that, The specific process for constructing the spatiotemporal demand-driver strategic behavior joint prediction model is as follows: Using the spatiotemporal grid of the target scheduling area as the basic unit, we extract order volume, passenger waiting time, and driver empty-run rate in historical time periods as demand-side features, and driver order rejection rate, hotspot area deviation, scheduling instruction response delay, and order acceptance distance preference as driver strategic behavior features. By combining external environmental characteristics such as weather, holidays, and large-scale events, a spatiotemporal graph convolutional network model is trained to simultaneously output the passenger demand and driver strategic behavior tendency values ​​for each spatiotemporal grid within a preset future time period.

3. The ride-hailing supply and demand matching and vehicle dispatching method integrating spatiotemporal prediction as described in claim 1, characterized in that, The specific process for generating the initial global supply and demand matching scheme and the idle capacity rebalancing scheduling scheme is as follows: Based on the predicted spatiotemporal distribution of passenger demand and the current online capacity distribution, a driver-passenger bipartite graph matching model is constructed with the goal of minimizing the average passenger waiting time and vehicle empty mileage, and the initial global supply and demand matching scheme is obtained by solving the model. At the same time, based on the predicted supply and demand gap values ​​of each spatiotemporal grid, an initial rebalancing scheduling scheme for allocating idle vehicles to the gap area is generated.

4. The ride-hailing supply and demand matching and vehicle dispatching method integrating spatiotemporal prediction as described in claim 1, characterized in that, The specific process of collecting real-time driver terminal operation status data, identifying drivers' strategic behaviors such as picking and choosing orders, hot spot clustering, and selectively executing dispatch instructions, and quantifying and generating dispatch compliance coefficients and order acceptance willingness coefficients for each driver is as follows: Real-time data collection includes driver order rejection records, driving trajectory deviation data, dispatch instruction response time, and order acceptance destination preference data; calculation of the order acceptance willingness coefficient based on the order rejection frequency within a preset statistical period, calculation of the dispatch compliance coefficient based on the execution ratio and response time of dispatch instructions, and calculation of the hotspot clustering tendency value based on the overlap between vehicle trajectory and hotspot areas.

5. The ride-hailing supply and demand matching and vehicle dispatching method integrating spatiotemporal prediction according to claim 4, characterized in that, The specific process of dynamically correcting the matching weight factor of the initial global supply and demand matching scheme is as follows: For drivers whose willingness to accept orders is below a preset threshold, increase the matching weight of high-yield orders and short-distance driving orders in the matching queue; for drivers whose dispatch compliance coefficient is above a preset threshold, give priority to matching high-quality orders and increase the weight of dispatch incentive subsidies; for drivers with excessively high hotspot clustering tendency, reduce their order matching priority in hotspot areas and guide transportation capacity to be dispersed to colder areas.

6. The ride-hailing supply and demand matching and vehicle dispatching method integrating spatiotemporal prediction according to claim 1, characterized in that, The specific process for generating a game-theoretic equilibrium scheduling scheme that balances the overall operational efficiency of the platform with the individual benefits of drivers is as follows: Based on the idle capacity rebalancing scheduling scheme, differentiated scheduling instructions are generated for drivers with different strategic behavior characteristics: direct allocation instructions are issued to drivers with high compliance, guided scheduling instructions with income subsidies are issued to drivers with medium compliance, and indirect guided instructions based on order push are issued to drivers with low compliance. By using differentiated instructions, capacity clustering is avoided, and a game steady state under global supply and demand equilibrium is achieved.

7. The ride-hailing supply and demand matching and vehicle dispatching method integrating spatiotemporal prediction according to claim 1, characterized in that, The specific process for forming the prediction-scheduling-feedback closed-loop optimization mechanism is as follows: The system statistically analyzes the actual order completion rate, driver execution rate, passenger waiting time, and empty-run rate for each round of scheduling. The deviation between the actual execution data and the predicted data is used as a loss term to iteratively update the weight parameters of the spatiotemporal demand-driver strategic behavior joint prediction model. Simultaneously, based on the evolution trend of driver group strategic behavior, the matching weight threshold and incentive subsidy coefficient are dynamically adjusted to achieve adaptive evolution of the scheduling strategy.

8. The ride-hailing supply and demand matching and vehicle dispatching method integrating spatiotemporal prediction according to claim 1, characterized in that, The method is also suitable for multi-category heterogeneous transportation capacity scenarios: For different categories of transportation services such as express cars, premium cars, and carpooling, corresponding strategic behavior quantification systems and matching weight rules are constructed. In the carpooling scenario, the driver's willingness to carpool is incorporated into the route planning and order splicing constraints to improve the acceptance rate of carpooling orders.

9. A ride-hailing supply-demand matching and vehicle dispatching system integrating spatiotemporal prediction, used to execute the ride-hailing supply-demand matching and vehicle dispatching method integrating spatiotemporal prediction as described in any one of claims 1-8, characterized in that, The system includes: Spatiotemporal demand-driver strategic behavior joint prediction module: Constructs a spatiotemporal demand-driver strategic behavior joint prediction model, based on historical order data, road network status data and driver historical behavior data, and synchronously outputs the spatiotemporal distribution prediction results of passenger demand and driver strategic behavior within the target scheduling area. Initial supply and demand matching and capacity scheduling generation module: Based on the predicted spatiotemporal distribution of passenger demand and the predicted spatiotemporal distribution of driver strategic behavior, it generates an initial global supply and demand matching scheme and an idle capacity rebalancing scheduling scheme. Driver strategic behavior perception and quantification module: collects driver terminal operation status data in real time, identifies drivers' strategic behaviors such as picking orders, hot spot clustering, and selectively executing dispatch instructions, and quantifies and generates dispatch compliance coefficient and order acceptance willingness coefficient for each driver; Supply and demand game equilibrium scheduling module: Based on the scheduling compliance coefficient and the order acceptance willingness coefficient, dynamically adjust the matching weight factor of the initial global supply and demand matching scheme and the guidance strategy of the idle capacity rebalancing scheduling scheme to generate a game equilibrium scheduling scheme that takes into account both the overall operational efficiency of the platform and the individual income of drivers. The scheduling closed-loop optimization module distributes the game equilibrium scheduling scheme to the corresponding driver and passenger terminals, and updates the parameters and scheduling strategy thresholds of the spatiotemporal demand-driver strategic behavior joint prediction model based on the actual order execution results, forming a prediction-scheduling-feedback full closed-loop optimization mechanism.

Images