Heterogeneous sharing automatic driving car sharing scheduling method and system with perspectiveness

Abstract

The invention discloses a foresight heterogeneous sharing automatic driving car sharing scheduling method and system, and relates to the technical field of intelligent traffic. The method comprises the following steps: generating candidate matching pairs of a shared autonomous vehicle state and a request; constructing a regional value function according to the state variable by using a Markov decision method, and evaluating the regional value function; constructing a dynamic ride-sharing matching algorithm according to interaction of the riding request and the shared automatic driving vehicle in the scheduling area; and based on the evaluation result of the regional value function, selecting the optimal matching of the shared autonomous vehicle and the riding request according to the candidate matching pair by using a dynamic car-sharing matching algorithm, and carrying out car-sharing scheduling according to the optimal matching result. According to the invention, a carpooling scheduling framework is designed, and the potential travel demand and the overall carpooling efficiency are considered, so that different passenger carrying capacities in the operation process of a heterogeneous SAV fleet are optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent transportation, and particularly to a forward-looking heterogeneous shared autonomous vehicle carpooling scheduling method and system. Background Art

[0002] The statements in this section merely provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] Carpooling systems have emerged as a promising solution to alleviate urban traffic congestion, enabling high travel demands to be efficiently met with fewer vehicles. The introduction of Shared Autonomous Vehicles (SAVs) has led to extensive research attention on these systems. SAVs are not affected by human limiting factors such as fatigue or distraction when processing real-time requests and adjusting trip paths, thus demonstrating excellent safety and operational flexibility. In addition, their ability to operate continuously and strict adherence to precise scheduling significantly enhance the efficiency of carpooling services. Research has shown that a fleet of SAVs can match the service capacity of a traditional taxi system while using only half the number of vehicles. Depending on specific traffic network characteristics, each SAV may replace 7 to 11 private cars, thus significantly reducing the ownership of private cars. SAVs combine flexible demand management and strict scheduling execution, bringing significant economic, environmental, and social advantages to urban transportation systems.

[0004] To address this trend, scholars worldwide have conducted extensive research on various aspects of ridesharing. These studies cover multiple topics, including trip request allocation, travel time estimation, pick-up point recommendation, route planning, and pricing strategies. In the early stage, static ridesharing scheduling was the main approach. This method requires all ridesharing requests to be known in advance, enabling optimal matching and maximizing resource utilization. Optimization objectives include minimizing the total travel distance, reducing passenger waiting time, and maximizing taxi utilization rate, etc. Most static optimization studies focus on deeply understanding the characteristics of ridesharing and its impacts. Some researchers developed a taxi ridesharing method aiming to reduce waiting and travel time during peak periods. Their research results show that ridesharing can reduce the distance of each taxi trip by 2 - 3 kilometers. However, their research was limited to requests with close origin-destination (OD) distances between two passengers. Some other researchers proposed a multi-modal autonomous vehicle ridesharing user equilibrium model (MARUE) in a static network equilibrium framework. Based on MARUE, they analyzed the optimal supply decisions of SAVs in terms of fleet size, fare, route planning, and allocation. Generally speaking, static optimization provides a theoretical framework that helps understand the impacts of ridesharing services on the transportation system and how these services can be optimized. Although static optimization provides valuable insights, its practicality is largely affected by the accuracy of demand prediction. In addition, these methods have limited adaptability in dealing with real-time demand and traffic condition fluctuations.

[0005] Compared with the static method, dynamic scheduling enables the system to continuously receive and process new ridesharing requests and perform real-time matching according to the current traffic conditions and vehicle locations. This adaptability enhances the system's ability to handle changing ridesharing requests and fluctuating traffic situations. Current research on dynamic taxi ridesharing mainly focuses on developing efficient algorithms because the inherent characteristics of dynamic ridesharing require quick matching of thousands of trip requests. For example, some researchers proposed a dynamic ridesharing method that can pair a large number of trip requests with taxis in real-time scenarios. Similarly, some other researchers introduced a mixed integer programming model to analyze the ridesharing problem and proposed a Lagrangian decomposition method and two heuristic methods to effectively solve the model. Some other researchers proposed a method that only allows passengers from specific taxi hotspots to share trips with passengers going to similar destinations. Their research shows that 48% of the ridesharing requests from hotspots can be shared, thus reducing the driving distance by 1.2 kilometers per shared trip. The ridesharing matching problem is an important technical challenge for the SAV system because this problem is crucial for optimizing the response to overall travel demand in terms of spatio-temporal dimensions. This adaptability enhances the system's ability to handle changing ridesharing requests and fluctuating traffic situations.

[0006] However, the ridesharing platform still faces critical operational efficiency challenges, mainly manifested as excessive passenger waiting times and excessive vehicle idle times. These inefficiencies not only affect service quality but also result in higher energy consumption and environmental concerns. Research shows that approximately 80% of Manhattan taxi trips can be shared by two passengers, highlighting the inefficiency of the current taxi operation model. The advancement of autonomous driving technology has given rise to the emergence of compact two-seater SAVs (such as Cybercab) in the market. Although a large amount of research has been dedicated to designing more efficient ridesharing strategies, the vehicle type allocation problem for heterogeneous SAV fleets remains a crucial yet understudied area that needs to be optimized according to specific travel characteristics. This factor plays a core role in determining the overall ridesharing system performance. In addition, traditional scheduling methods have obvious limitations. These methods are too short-sighted in the decision-making process and lack forward-looking planning for future ridesharing opportunities, thus limiting the optimal scheduling results. It is worth noting that many studies assume that complete travel information is available. However, this assumption fundamentally mismatches the actual operating situation. In reality, real-time order changes significantly affect vehicle route planning and dynamic matching decisions. Therefore, the existing research still lacks an understanding of the complex interrelationships among trip characteristics, potential travel demands, and the overall ridesharing system efficiency. Summary of the Invention

[0007] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide a forward-looking heterogeneous shared autonomous vehicle ridesharing scheduling method and system. By designing a ridesharing scheduling framework, considering potential travel demands and overall ridesharing efficiency, it optimizes the different passenger-carrying capacities during the operation of heterogeneous SAV fleets.

[0008] To achieve the above purpose, the present invention is realized through the following technical solutions: The first aspect of the present invention provides a forward-looking heterogeneous shared autonomous vehicle ridesharing scheduling method, including the following steps: Obtain ride requests, shared autonomous vehicle status, and road network information within the scheduling area; Generate candidate matching pairs of shared autonomous vehicle status and requests based on ride requests, shared autonomous vehicle status, and combined road network information; Determine the state variables of the regional value function according to the historical order information of the scheduling area, construct the regional value function using the Markov decision method based on the state variables, and evaluate the regional value function; Construct a dynamic ridesharing matching algorithm based on the interaction between ride requests and shared autonomous vehicles within the scheduling area; Based on the evaluation results of the regional value function, the optimal matching of shared autonomous vehicles and ride requests is selected according to candidate matching pairs using the dynamic carpool matching algorithm, and carpool scheduling is performed according to the optimal matching results.

[0009] Further, the specific steps for generating candidate matching pairs of shared autonomous vehicle status and requests by combining ride requests and shared autonomous vehicle status with road network information are as follows: Set request acceptance conditions according to the relationship among ride requests, shared autonomous vehicle status, and road network information; Based on the satisfaction of ride requests with the request acceptance conditions, determine the boarding and alighting order based on the shared autonomous vehicle status; Determine the shortest path of the shared autonomous vehicle according to the boarding and alighting order, and determine candidate matching pairs of shared autonomous vehicle status and requests.

[0010] Further, the specific steps for constructing a regional value function according to state variables using the Markov decision method and evaluating the regional value function are as follows: Divide the scheduling area into several small areas; Use the Markov decision method to model and estimate the regional value function for the state variables of small areas. Among them, introduce the temporal difference learning method to estimate the regional value function, and update the parameters based on the Bellman equation.

[0011] Furthermore, the regional value function represents the long-term reward that a shared autonomous vehicle can obtain after entering the current area within a preset time period.

[0012] Furthermore, the state variables of the regional value function include the number of orders starting from the current area, the number of orders with the destination being the current area, the number of shared autonomous vehicles in the current area, and the average order waiting time in the current area.

[0013] Further, the specific steps for constructing a dynamic carpool matching algorithm according to the interaction between ride requests and shared autonomous vehicles in the scheduling area are as follows: Use the bipartite graph matching algorithm to transform the real-time carpool scheduling problem; According to the matching degree and compatibility between shared autonomous vehicles and ride requests in the scheduling area, consider multiple factors to determine the advantage function and set constraint conditions.

[0014] Furthermore, the multiple factors to be considered include: Satisfy the most ride requests; Reduce the passenger waiting time and detour distance; Improve the passenger capacity rate and service success rate of shared autonomous vehicles; Incorporate the consideration of long-term benefits into the scheduling decision.

[0015] The second aspect of the present invention provides a forward-looking heterogeneous shared autonomous vehicle carpooling scheduling system, including: A data acquisition module configured to acquire ride requests, shared autonomous vehicle status, and road network information within the scheduling area; A candidate matching module configured to generate candidate matching pairs of shared autonomous vehicle status and requests based on ride requests, shared autonomous vehicle status, and in combination with road network information; A regional value evaluation module configured to determine state variables of a regional value function based on historical order information of the scheduling area, construct a regional value function according to the state variables using a Markov decision method, and evaluate the regional value function; A matching algorithm module configured to construct a dynamic carpooling matching algorithm based on the interaction between ride requests and shared autonomous vehicles within the scheduling area; A carpooling scheduling module configured to, based on the evaluation result of the regional value function, use the dynamic carpooling matching algorithm to select an optimal match between shared autonomous vehicles and ride requests according to the candidate matching pairs, and perform carpooling scheduling according to the optimal matching result.

[0016] The third aspect of the present invention provides a medium having a program stored thereon, and when the program is executed by a processor, it implements the steps in the forward-looking heterogeneous shared autonomous vehicle carpooling scheduling method as described in the first aspect of the present invention.

[0017] The fourth aspect of the present invention provides a device including a memory, a processor, and a program stored on the memory and executable on the processor, and when the processor executes the program, it implements the steps in the forward-looking heterogeneous shared autonomous vehicle carpooling scheduling method as described in the first aspect of the present invention.

[0018] The above one or more technical solutions have the following beneficial effects: The present invention discloses a forward-looking heterogeneous shared autonomous vehicle carpooling scheduling method and system, and proposes a forward-looking carpooling scheduling framework for heterogeneous shared autonomous vehicles to improve the overall operation efficiency of heterogeneous SAV fleets in urban environments. The framework combines a rule-based method with a Markov decision process (MDP), and through the collaborative work of three major modules: carpool matching, path planning, and hot spot guidance, realizes the intelligent scheduling of SAVs and the optimal allocation of resources. The key feature of the framework lies in its consideration of the heterogeneous characteristics of SAVs, that is, differential scheduling decisions are made for vehicles with different capacities, thereby improving the overall system benefits while ensuring passenger satisfaction. In addition, by constructing a regional value function and combining a temporal difference learning method, the system can real-time sense the changes in urban travel demands and guide idle SAVs to high-potential areas to improve the order acceptance rate and resource utilization rate.

[0019] Advantages of additional aspects of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings forming a part of this specification are used to provide a further understanding of the present invention. The schematic embodiments and descriptions thereof of the present invention are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0021] Figure 1 It is a flowchart of a forward-looking heterogeneous shared autonomous vehicle carpooling scheduling method in Embodiment 1 of the present invention; Figure 2 It is a schematic diagram showing the change of service success rate of SAVs at different times of a day in Embodiment 1 of the present invention; Figure 3 It is a comparison chart of the energy consumption changes of three methods in a day in Embodiment 1 of the present invention; Figure 4 It is a trend chart of the utilization rate of SAVs in a day in Embodiment 1 of the present invention; Figure 5 It is a trend chart of the seat occupancy rate of SAVs in a day in Embodiment 1 of the present invention; Figure 6 It is a comparison chart of the distribution ratios of the passenger-carrying states of SAVs under three methods in Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0023] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprising" and / or "including" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof; Embodiment 1: Embodiment 1 of the present invention provides a forward-looking heterogeneous shared autonomous vehicle carpooling scheduling method, as Figure 1 shown, including the following steps: Step 1: Obtain ride requests, shared autonomous vehicle states, and road network information within the scheduling area.

[0024] (1) Ride requests.

[0025] In this embodiment, the set of all passenger requests is defined as . The research period is divided into n time periods of equal duration, and the subset of ride requests within each time period t is denoted as , where: .

[0026] Each ride request is denoted as , and is represented by a tuple composed of travel characteristics. Among them, is the submission time of the request, and are the pick-up location and drop-off location respectively, is the number of passengers in the request. indicates whether the passenger accepts carpooling, and its value remains unchanged throughout the planning period. If the passenger accepts carpooling, ; otherwise, there is . indicates whether the passenger has been served. If served within time period t, then , and this request will be removed from the matching pool; if not served, then there is , and this request will enter the matching pool of the next time period for continued matching. indicates the number of time periods that this request has not been satisfied. If the request is not satisfied within multiple consecutive time periods, the passenger will cancel the request and choose other means of transportation, which is regarded as a failure of carpooling. Without loss of generality, the time limit in this embodiment is set to 8 time periods.

[0027] (2) SAV state.

[0028] For the SAV, the state of the jth SAV within time period t is denoted as a tuple . Among them, is the position of the SAV at time period t; is the number of passengers in the vehicle currently; is the passenger capacity of the SAV; The present invention considers two vehicle types: low capacity , high capacity ; is a 0-1 variable indicating whether the SAV is in the service state currently; If it is in service, ; Otherwise, ; represents the number of consecutive idle time periods; If the SAV is idle for multiple consecutive time periods, it will be guided to high-demand areas. Without loss of generality, this embodiment is set to 8 time periods. represents the current driving path of the SAV; This path consists of a time-sequence series of a set of nodes, including the shortest path node set from the current position to the boarding point , and the shortest path node set from the boarding point to the alighting point . This path is updated dynamically over time.

[0029] To represent the dynamic evolution of the path, the symbol "-" is used to indicate the movement between positions. For example, if the SAV is serving a request , its current path is . If a new carpooling request is received at this time and the journey is successfully shared with it, the feasible path at this time can be represented as .

[0030] (3) Road network.

[0031] The road network is denoted as a tuple R, represented as . Among them, is the set of road nodes, is the set of road edges, is the set of weights of road edges, used to calculate the shortest path and energy consumption between two positions.

[0032] In most studies on vehicle scheduling problems, graph theory is often used for modeling, while ignoring the specific structure of the road network. Usually, the distance between adjacent boarding points is calculated using the Euclidean distance formula. However, in the dynamic carpooling scheduling problem, it is crucial to incorporate the road network as a key component in problem-solving, which is consistent with the real situation. Therefore, in the scheduling decision-making process of this embodiment, the distance between any two positions is determined through the adjacency matrix of the road network, and the Dijkstra algorithm is used to calculate the shortest path, which can effectively evaluate the shortest driving distance and energy consumption of the SAV.

[0033] After obtaining the above information, as shown in the appendix Figure 1As shown, the dynamic carpooling framework proposed in this embodiment specifically includes three steps: regional value learning, candidate matching pair generation, and real-time scheduling of carpooling.

[0034] The generation of candidate matching pairs aims to provide effective options for the real-time scheduling of carpooling, and uses the regional value function and other metric results to achieve efficient carpooling services. For each ride request , a search is performed to identify the set of SAV candidates that can satisfy this request . This can form an initial set of SAV-order pairs , where , satisfying , . When a new ride request is added to the current schedule of a certain SAV, it may be necessary to detour to reach the pick-up location of this request. To optimize the travel efficiency of the SAV, its path is determined by the shortest path algorithm. In addition, the detour ratio between the existing request and the new request is calculated to evaluate the feasibility of this carpooling method. Finally, this evaluation process will determine the set of effective SAV-ride request pairs. Through the generation, screening, and evaluation of candidate pairs, high-quality and feasible matching pairs can be carefully selected, providing a solid foundation for real-time carpooling scheduling decisions.

[0035] The evaluation of the regional value function plays an important role in estimating the travel demand and passenger value of each region and time period. This process is achieved by establishing a value function model. First, the research area is divided into multiple different sub-regions. Subsequently, using historical taxi travel demand data, the value function of each region is estimated through the Bellman Equation and Temporal Difference (TD) method. This method combines the update rule of the Bellman Equation with the TD incremental learning method to evaluate the travel demand and value of each region. Through continuous iteration and learning, the estimation accuracy of the regional value function can be improved, thus providing a more reliable basis for candidate matching pair generation and carpooling scheduling decisions.

[0036] The main goal of real-time carpooling scheduling is to determine the optimal carpooling plan and scheduling plan based on candidate matching pairs. In a real-world scenario, an SAV usually receives multiple ride requests, and each request may also form multiple effective matching pairs with multiple SAVs. After the candidate matches for each ride request are generated, we construct a matching graph to show the relationships between different candidate matching pairs. To evaluate the quality of these matches, we introduce a weight function as the edge weight in the matching graph (i.e., ). This weight function comprehensively considers multiple key factors, including: the energy consumption of the SAV, the detour distance, the waiting time, and the revenue of the SAV. These factors together provide a comprehensive evaluation of each candidate matching pair. Finally, a method based on the Kuhn - Munkres algorithm (also known as the Hungarian algorithm) is used to select the SAV - ride request matching pair that can maximize the overall benefit. If a SAV is idle for 8 consecutive time periods, the system will evaluate its current time period and location. Based on this evaluation, the system will select a high - demand area as the guiding destination for the SAV. After receiving the guiding instruction, the SAV will travel to the area with a higher ride request generation rate according to the path information provided by the system.

[0037] Step 2: Generate candidate matching pairs of shared autonomous vehicle (SAV) states and requests by combining ride requests, SAV states, and road network information.

[0038] The generation of candidate matching pairs is a pre - processing process aimed at narrowing the solution space by identifying and eliminating infeasible SAV - ride request matching pairs.

[0039] Step 2.1: Set request acceptance conditions based on the relationship among ride requests, SAV states, and road network information.

[0040] Consider a SAV with the current location , which has accepted a ride request and can reach the pick - up point of this request at time period t 1 . At this time, if this SAV receives a new ride request at time period t, it will accept this request under the following conditions: Condition 1: The remaining seat number of the SAV is greater than or equal to the number of passengers of the request .

[0041] (1).

[0042] Condition 2: The actual distance between the current SAV location and the pick - up point of the new request does not exceed the threshold .

[0043] (2).

[0044] Condition 3: The deviation degree of the driving path between two locations is defined as the ratio of the shortest driving distance to the actual driving distance. The deviation degree of the driving path of the SAV from the current location via the pick - up and drop - off points of the new request does not exceed the threshold , as shown in Equation (3). Among them, represents the shortest driving distance from node x to node y, Indicates the actual driving distance from node x to node y.

[0045] (3).

[0046] Condition 4: If the SAV has not reached , the path offset of the first request needs to meet the threshold condition (4).

[0047] , if (4).

[0048] Condition 5: When the SAV has reached , the path offset of the first request needs to meet the threshold condition (5).

[0049] , if (5).

[0050] Among them, Indicates the moment when the vehicle receives the first request, Indicates the current moment At and before that moment, Indicates the current moment After .

[0051] Step 2.2: Determine the boarding and alighting order based on the state of the shared autonomous vehicle according to the satisfaction of the request acceptance conditions for the ride request.

[0052] If the request Meets Conditions 1 and 2, the possible boarding and alighting orders can be obtained. Specifically, the boarding and alighting orders for the two requests need to be planned according to the current state of the SAV. However, different boarding and alighting orders will have different path offset situations. Therefore, these possible boarding and alighting orders also need to be additionally checked according to Conditions 3-5.

[0053] In a specific implementation manner, when the SAV receives the request , if it has not reached , there are four possible boarding and alighting orders: , , , .

[0054] When the SAV has reached , there are two possible boarding and alighting orders: , .

[0055] Step 2.3: Determine the shortest path of the shared autonomous vehicle according to the boarding and alighting order, and determine the candidate matching pairs of the shared autonomous vehicle status and the request.

[0056] According to the above limited arrangement scheme, this embodiment uses the enumeration method to determine the shortest path of the SAV. Once the path is selected, the SAV will strictly execute according to this path.

[0057] During time period t, the initial SAV-request pair set can be obtained through Conditions 1 and 2 , where . To ensure feasibility, additional checks are also required using Conditions 3-5 to obtain the effective SAV-request pair set , .

[0058] Step 3: Determine the state variables of the regional value function according to the historical order information of the dispatching area, construct the regional value function according to the state variables using the Markov decision method, and evaluate the regional value function.

[0059] Step 3.1: Determine the state variables of the regional value function according to the historical order information of the dispatching area.

[0060] In this embodiment, the regional value function represents the long-term return that the shared autonomous vehicle can obtain after entering the current area during the preset time period. The state variables of the regional value function include the number of orders departing from the current area, the number of orders with the current area as the destination, the number of shared autonomous vehicles in the current area, and the average order waiting time in the current area.

[0061] Step 3.2: Construct the regional value function according to the state variables using the Markov decision method, and evaluate the regional value function.

[0062] Step 3.2.1: Divide the dispatching area into several small areas.

[0063] In a specific implementation manner, in order to realize the guidance of the SAV to the high-demand area, this embodiment introduces a regional value function. This function can quantitatively evaluate the area according to the historical order information of each area. Divide the concerned area into m small areas, representing the set of all areas. At each time period t, each area has a corresponding state representation, including the vector composed of the following state variables: (6).

[0064] Among them, is the number of orders departing from this area within time period t, is the number of orders with the destination being this area within time period t, is the number of SAVs in this area within time period t, is the average order waiting time in this area within time period t.

[0065] Step 3.2.2: Use the Markov decision method to model and estimate the regional value function of the state variables of the small area.

[0066] Among them, the temporal difference learning method is introduced to estimate the regional value function, and the parameters are updated based on the Bellman equation.

[0067] In a specific implementation, the state variables are input into a Markov decision process (MDP) to model and estimate the value of each area The regional value function is defined as , indicating the long-term return that the SAV can obtain after entering this area in time period t. The initial value of each area is set to 0 and is updated as the data is traversed, and finally converges as the regional value.

[0068] To estimate the regional value function, this embodiment introduces the temporal difference (TD) learning method, which is implemented by introducing the temporal sequence s, and the parameters are updated based on the Bellman equation. The Bellman update equation is as follows: (7).

[0069] Among them, is the learning rate; is the immediate return at the current time period t; is the discount factor, which determines the influence of future events on the Markov process. In the present invention, it is set to 0.9; is the estimated value of the next state.

[0070] Immediate return includes multiple factors such as the number of orders completed by the SAV in the current area and the number of passengers served, as well as the average waiting time of passengers. Through continuous multi-round training, the system can accurately learn the value change trend of each area, and then optimize the guidance direction when the SAV is in the idle state.

[0071] Step 4: Construct a dynamic carpool matching algorithm according to the ride request and the interaction between the shared autonomous vehicle in the scheduling area.

[0072] Step 4.1: Use the bipartite graph matching algorithm to transform the real-time carpool scheduling problem.

[0073] Step 4.2: Determine the advantage function and set the constraint conditions by considering multiple factors according to the matching degree and compatibility between the shared autonomous vehicles and ride requests within the scheduling area.

[0074] The multiple factors to be considered include: satisfying the largest number of ride requests; reducing the passenger waiting time and detour distance; increasing the occupancy rate and service success rate of the shared autonomous vehicles; and incorporating the consideration of long-term benefits into the scheduling decision.

[0075] The main objective of the carpooling scheduling problem proposed in this embodiment is to achieve a coordinated and optimized overall benefit by finding the optimal match between shared autonomous vehicles (SAVs) and ride requests within each time period. This optimization aims to maximize the utilization rate of SAVs while reducing their energy consumption.

[0076] In a specific implementation manner, to solve this problem, the real-time carpooling scheduling problem is transformed into an online bipartite graph matching problem. By using the bipartite graph matching algorithm, an optimal matching solution can be determined for this problem.

[0077] In this embodiment, the bipartite graph used consists of two types of nodes: SAV nodes and ride request nodes. The SAV nodes represent the currently available SAVs, while the ride request nodes represent the carpooling requests waiting for service. The bipartite graph connects these two types of nodes through an edge set These edges are assigned weights, representing the matching degree or compatibility between the SAVs and ride requests. These weights consider multiple factors, including the energy consumption of the SAVs, the boarding and alighting distances, the current and future revenues of the SAVs, etc. The carpooling scheduling problem for each time period can be expressed as: (8).

[0078] The constraint conditions are as follows: (9), (10).

[0079] Where, represents whether the matching is successful. The advantage function represents the matching evaluation between the SAV j serving the ride request k and the carpooling request i; the value of the 0-1 variable is 1 indicating successful matching and 0 indicating unsuccessful matching. Equation (9) means that each SAV can serve at most one carpooling request; Equation (10) means that each carpooling request can be served by at most one SAV.

[0080] The occurrence of carpooling depends on the interaction between the ride request and the SAV within the scheduling area. The matching result is reflected in the advantage function in the carpooling model. The carpooling model designed in this embodiment comprehensively considers the following aspects: 1) satisfying the most ride requests; 2) reducing the passenger waiting time and detour distance, improving the travel efficiency and user experience; 3) increasing the passenger-carrying rate and service success rate of the SAV, achieving efficient utilization of resources and reduction of energy consumption; 4) taking into account the long-term benefits in the scheduling decision.

[0081] Considering the above aspects, the following advantage function is proposed: (11).

[0082] By solving the advantage function, the most suitable scheduling scheme is found. Among them, the function has a positive impact on the advantage function, meaning that serving ride requests in high-value areas can generate greater benefits. The state represents the actual arrival time and location of SAV j after serving ride requests i and k. The function has a negative impact on the advantage function. SAVs in areas with lower value are more likely to be selected to serve ride requests. represents the time and location where SAV j arrives after serving ride request k. represents the immediate benefit of SAV j serving ride request i and has a positive impact on the advantage function. represents the energy consumption of SAV j serving ride request i and has a negative impact on the advantage function. represents the resource utilization rate of carpooling for two ride requests. The smaller the value, the higher the matching degree. represents that the pick-up and drop-off points of two ride requests are at the same location. and respectively represent the completion time periods of carpooling requests i and k. Usually, there is , but if , it means that the driving route of the SAV does not change after incorporating carpooling request i.

[0083] Step 5: Based on the evaluation results of the regional value function, use the dynamic carpooling matching algorithm to select the optimal match between the shared autonomous vehicle and the ride request according to the candidate match pairs, and perform carpooling scheduling according to the optimal matching result.

[0084] In a specific implementation manner, the energy consumption of the SAV is mainly affected by two factors: the driving distance and the type of the SAV vehicle, and is specifically expressed as: (12).

[0085] In the formula, Denote the energy consumed when the n-type SAV travels from node x to node y; Denote the energy consumption per unit distance of the n-type SAV, with the unit of liters per kilometer (L / km); Denote the shortest distance between nodes x and y.

[0086] Consider a scenario where the position of SAV j when receiving ride request i is . At this time, the unit price of energy is denoted as . If the service distance of the SAV is within two kilometers, the passenger will be charged a fixed total fee ; conversely, if the travel distance exceeds 2 kilometers, the passenger must pay an additional fee, with the unit distance fee being . Obviously, the energy consumption cost during the process of the SAV accepting the ride request and traveling from the current position to the boarding position of the request can be calculated as: (13).

[0087] Similarly, the energy consumption cost for the SAV to complete this ride request can be defined as: (14).

[0088] The total energy consumption cost for the SAV to complete this ride request is: (15).

[0089] The direct income after the SAV completes the ride request is the amount that the passenger needs to pay, as shown in the following formula: (16).

[0090] Although carpooling can improve the utilization rate of SAVs, reduce energy consumption, and improve passenger service efficiency, it may also have a negative impact on service quality, mainly reflected in two aspects: one is that detours may occur during the service process, and the other is that the resource utilization efficiency of carpooling may be relatively low. Therefore, when carpooling occurs, the system will feedback the relevant penalty value to the advantage function to minimize detours and improve carpooling efficiency. The penalty terms are defined as follows: Low carpooling utilization rate penalty: (17).

[0091] Detour distance penalty: (18).

[0092] Total penalty term: (19).

[0093] Among them, The smaller the value, the more beneficial it is for the carpool matching of two ride requests. On the contrary, a larger value indicates a less ideal carpooling result, suggesting a lower possibility of combining the two requests. Through the analysis and planning of carpooling results, an optimal scheduling scheme is obtained, greatly improving the efficiency of the carpooling process and the passenger experience.

[0094] To evaluate the performance of the proposed model, this embodiment conducts an empirical study based on the real taxi operation data in the Manhattan area of New York City. The data comes from the New York City Taxi and Limousine Commission (NYC TLC) and includes the pick-up and drop-off points, travel time, fare, and number of passengers for the whole day of June 1, 2022. The Manhattan area is further divided into 265 small regions to construct a spatial network. This invention uses the travel records during the peak period from 7:00 to 9:00 on this date, including a total of 40,823 ride requests. Based on the actual road structure of New York City, a directed graph containing 4,092 nodes and 9,453 road edges is constructed as the road network in the study.

[0095] In the experiment, two vehicle types are considered: SAV-A with a capacity of 2 people and SAV-B with a capacity of 4 people. The vehicles are initially evenly distributed at different road nodes. The simulation scheduling period is 1 minute, and the entire scheduling process lasts for 120 time periods (i.e., 2 hours). The parameter settings in the generation of candidate matching pairs are as follows: the maximum detour ratio threshold accepted by passengers is 1.3, and the maximum additional detour ratio accepted by SAVs is 1.5.

[0096] To evaluate the effectiveness of the carpooling method proposed in this embodiment, three SAV service methods are set in the experiment, including the SAV carpooling method proposed based on the present invention, a variant of the SAV carpooling method, and a non-carpooling method, which are defined as MSRM, VSRM, and NRM respectively. VSRM is defined as the matching between two ride requests whose distances between the starting point and the destination are within a certain threshold range. For the initial ride request matching process, MSRM and VSRM adopt the same method as NRM. To evaluate the performance of the proposed method in optimizing the carpooling dynamic scheduling problem, the present invention uses: Service Success Rate (SSR), Energy Consumption (EC), Average Energy Consumption (AEC), Utilization Rate of SAV (TUR), Pickup Time (PT), Seat Occupancy Rate (OT), Total Company Net Profit of SAV (TCNP), Ride-hailing Platform Profit (RPP), and Average Passenger Cost (APC) as evaluation indicators. AEC is defined as the total energy consumption divided by the number of passengers served, and TUR is defined as the distance that the SAV transports passengers from the departure place to the destination divided by the total distance traveled by the vehicle. APC is defined as the total fare of passengers divided by the passenger capacity.

[0097] Figure 2Shows the change in the service success rate of SAV at different times of the day. The service success rate is defined as the percentage of ride requests with completed services out of the total number of requests. Requests that are not responded to within more than 4 minutes of waiting time are considered service failures. It can be seen from the figure that the MSRM method is overall superior to other methods in terms of service success rate. In contrast, the service success rate of NRM fluctuates greatly, while MSRM can maintain a relatively stable service level, indicating that MSRM is more robust and can effectively cope with the dynamic fluctuations of ride demand. During peak hours of ride requests, especially the evening rush hour (18:00–22:00), the advantage of MSRM is more significant, demonstrating its ability to ensure service quality in a high-demand environment. MSRM can better meet passenger needs and improve the overall service level when demand surges. This advantage mainly stems from MSRM optimizing the usage efficiency of SAV and the passenger travel experience from a global and long-term perspective in the scheduling strategy. It achieves a better balance between accepting new requests and controlling the degree of detours, thereby improving the overall utilization rate and service efficiency of SAV. In addition, the service success rate of VSRM is also better than that of NRT, further verifying the effectiveness of the carpooling strategy in improving service coverage ability.

[0098] Figure 3 Shows the change in energy consumption of the three methods during the day. Generally speaking, for each method, the energy consumption of SAV is positively correlated with the number of ride requests completed. During the peak period of ride demand, due to the large number of requests, the energy consumption of SAV increases significantly; on the contrary, during periods of low demand, the energy consumption also decreases accordingly. Compared with VSRM, MSRM completed more service requests with lower energy consumption: the number of service requests increased by 4.0%, but the energy consumption decreased by 1.43%. Compared with NRT, the performance of MSRM is more prominent, with the number of service requests increasing by 11.6% while the energy consumption decreased by 3.67%.

[0099] Figure 4 and Figure 5 Show the change trends of the utilization rate and seat occupancy rate of SAV during the day respectively. It can be clearly seen that MSRM is always superior to VSRM and NRT in terms of the utilization rate and seat occupancy rate of SAV. This finding indicates that the advantage of MSRM lies in its efficient scheduling and utilization of SAV resources, thereby improving the vehicle utilization efficiency. Especially in the medium- and long-distance travel scenarios, the carpooling service shows more significant energy-saving advantages.

[0100] Figure 6 Shows the distribution ratio of the passenger-carrying status of SAV under the three methods. Among them, the proportion of SAV with passengers in the car in the NRT mode is 90.1%, that in VSRM is 70%, while MSRM significantly improves the passenger-carrying capacity of SAV. However, the situation where the number of passengers per vehicle is less than 2 still accounts for 86.5%.

[0101] The present invention belongs to the field of intelligent transportation technology and provides a forward-looking scheduling framework that combines rule-based methods with Markov Decision Process (MDP) to achieve real-time carpooling operations. The framework is constructed around three interrelated components: carpool matching, path planning, and hot spot guidance, aiming to improve operational efficiency while maintaining an optimal balance between passengers' immediate satisfaction and the long-term benefits of SAVs. By leveraging real data from taxi operations in New York City and the urban road network, we conducted a comprehensive case study on the effectiveness of this framework. Experimental results show that compared with traditional methods, this method has achieved significant improvements in multiple aspects, including a 4% increase in service success rate, a 1.43% reduction in energy consumption, a 4.1% increase in the average passenger occupancy rate of vehicles, and a 16.8% increase in the number of passengers served by each SAV. These results strongly demonstrate the transformative potential of heterogeneous SAV services and provide valuable insights for transportation planners to implement more efficient and sustainable urban mobility solutions.

[0102] Example Two: Example Two of the present invention provides a forward-looking heterogeneous shared autonomous vehicle carpooling scheduling system, including: A data acquisition module configured to acquire ride requests, shared autonomous vehicle status, and road network information within the scheduling area; A candidate matching module configured to generate candidate matching pairs of shared autonomous vehicle status and requests based on ride requests and shared autonomous vehicle status in combination with road network information; A regional value evaluation module configured to determine the state variables of the regional value function based on historical order information of the scheduling area, construct a regional value function according to the state variables using Markov decision methods, and evaluate the regional value function; A matching algorithm module configured to construct a dynamic carpool matching algorithm based on the interaction between ride requests and shared autonomous vehicles within the scheduling area; A carpooling scheduling module configured to select the optimal match between shared autonomous vehicles and ride requests based on the evaluation results of the regional value function using the dynamic carpool matching algorithm and perform carpooling scheduling according to the optimal matching results.

[0103] Example Three: Example Three of the present invention provides a medium on which a program is stored, and when the program is executed by a processor, it implements the steps in the forward-looking heterogeneous shared autonomous vehicle carpooling scheduling method as described in Example One of the present invention.

[0104] Example Four: Embodiment 4 of the present invention provides a device, including a memory, a processor, and a program stored on the memory and executable on the processor. When the processor executes the program, the steps in the forward-looking heterogeneous shared autonomous vehicle carpooling scheduling method described in Embodiment 1 of the present invention are implemented.

[0105] The steps involved in the above Embodiments 2, 3, and 4 correspond to those in Method Embodiment 1. For specific implementation manners, reference may be made to the relevant description part of Embodiment 1.

[0106] Those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computer device. Optionally, they can be implemented by program codes executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be separately fabricated into individual integrated circuit modules, or multiple of them can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0107] Although the specific implementation manners of the present invention have been described above in conjunction with the accompanying drawings, they do not limit the protection scope of the present invention. Those skilled in the art should understand that, based on the technical solutions of the present invention, various modifications or deformations that can be made without creative efforts by those skilled in the art are still within the protection scope of the present invention.

Claims

1. A forward-looking method for scheduling heterogeneous shared autonomous driving cars, characterized in that: The following steps are involved: Obtain ride requests, shared autonomous vehicle status, and road network information within the dispatch area; Generate candidate matching pairs of shared autonomous vehicle status and request based on the ride request and shared autonomous vehicle status combined with road network information; Determine the state variables of the regional value function according to the historical order information of the dispatching area, construct the regional value function according to the state variables using the Markov decision method, and evaluate the regional value function; Build a dynamic ride-sharing matching algorithm based on the interaction between ride requests and shared autonomous vehicles in the dispatch area; Based on the evaluation results of the regional value function, a dynamic carpooling matching algorithm is used to select the optimal match between the shared autonomous driving car and the ride request according to the candidate matching pairs, and carpooling scheduling is performed according to the optimal matching results.

2. The forward-looking heterogeneous shared autonomous driving carpooling scheduling method according to claim 1, characterized in that: The specific steps of generating candidate matching pairs of shared autonomous vehicle status and request based on ride requests and shared autonomous vehicle status combined with road network information are as follows: Setting request acceptance conditions based on the relationship between ride requests, shared autonomous vehicle status, and road network information; Determine the order of boarding and disembarking based on the shared autonomous vehicle status according to whether the ride request meets the request acceptance conditions; The shortest path for the shared autonomous vehicle is determined based on the pick-up and drop-off sequence, and candidate matching pairs of the shared autonomous vehicle status and the request are determined.

3. The forward-looking heterogeneous shared autonomous driving carpooling scheduling method according to claim 1, characterized in that: The specific steps of constructing the regional value function based on the state variables using the Markov decision method and evaluating the regional value function are as follows: Divide the dispatching area into several small areas; The Markov decision method is used to model and estimate the regional value function of the state variables of a small area. The temporal difference learning method is introduced to estimate the regional value function, and the parameters are updated based on the Bellman equation.

4. The forward-looking heterogeneous shared autonomous driving carpooling scheduling method according to claim 3, characterized in that: The regional value function represents the long-term return that a shared autonomous vehicle can obtain after entering the current area within a preset time period.

5. The forward-looking heterogeneous shared autonomous driving carpooling scheduling method according to claim 4, characterized in that: The state variables of the regional value function include the number of orders departing from the current area, the number of orders destined for the current area, the number of shared autonomous vehicles in the current area, and the average order waiting time in the current area.

6. The forward-looking heterogeneous shared autonomous driving carpooling scheduling method according to claim 1, characterized in that: The specific steps to build a dynamic ride-sharing matching algorithm based on the interaction between ride requests and shared autonomous vehicles in the dispatch area are: Transform the real-time ride-sharing scheduling problem using a bipartite graph matching algorithm; Based on the matching degree and compatibility between shared autonomous vehicles and ride requests in the dispatch area, the advantage function is determined by considering multiple factors and constraints are set.

7. The forward-looking heterogeneous shared autonomous driving carpooling scheduling method according to claim 6, characterized in that: Consider a variety of factors including: Satisfy the most ride requests; Reduce passenger waiting time and detour distance; Improve the passenger load and service success rate of shared autonomous vehicles; Incorporate long-term benefit considerations into scheduling decisions.

8. A forward-looking heterogeneous shared autonomous driving carpooling scheduling system, characterized in that: include: a data acquisition module configured to acquire ride requests, shared autonomous vehicle status, and road network information within a dispatch area; a candidate matching module configured to generate a candidate matching pair of a shared autonomous vehicle state and a request based on the ride request and the shared autonomous vehicle state in combination with road network information; A regional value evaluation module is configured to determine state variables of a regional value function according to historical order information of a dispatching area, construct a regional value function according to the state variables using a Markov decision method, and evaluate the regional value function; a matching algorithm module configured to construct a dynamic ride-sharing matching algorithm based on the interaction of ride requests with shared autonomous vehicles in a dispatch area; The carpooling scheduling module is configured to select the best match between the shared autonomous driving car and the ride request based on the candidate matching pairs based on the evaluation results of the regional value function, and perform carpooling scheduling based on the best matching results using a dynamic carpooling matching algorithm.

9. A computer-readable storage medium, characterized in that: Multiple instructions are stored therein, and the instructions are suitable for being loaded by a processor of a terminal device and executed by the forward-looking heterogeneous shared autonomous driving carpooling scheduling method described in any one of claims 1-7.

10. A terminal device, characterized in that: It includes a processor and a computer-readable storage medium, the processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executed by the forward-looking heterogeneous shared autonomous driving carpooling scheduling method described in any one of claims 1-7.

Images