Automatic driving taxi team multi-task matching dynamic scheduling distributed robust optimization decision-making method
Through the iterative multi-task matching decision-making method combining KM algorithm, Gurobi solver and greedy strategy, the binary graph and network flow theory modeling is used, combined with distributed robust opportunity constraints, the orders and charging in the autonomous driving taxi fleet are solved. The optimization problems of regional allocation and parking tasks have improved operational efficiency and user experience, and adapted to the time and space uncertainty of users' travel needs.
Patent Information
- Application Number
- CN202411141757.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-19
- Publication Date
- 2024-11-26
AI Technical Summary
The existing autonomous driving taxi fleet scheduling methods cannot effectively handle multi-tasks such as order reception, charging, regional allocation and parking, and are difficult to adapt to the time and space uncertainty of users' travel needs, resulting in insufficient operational efficiency and profitability and poor user experience. .
Using a multi-task matching decision-making method combining iterative KM algorithm, Gurobi solver and greedy strategy, it optimizes vehicle order reception, charging, regional configuration and Parking tasks improve the robustness and efficiency of scheduling decisions.
Achieve multi-task optimization in the operation of autonomous taxi fleets, improve operational efficiency and profitability, reduce resource waste, improve user satisfaction and order satisfaction rate, and ensure stable operation in the case of large demand fluctuations. .
Smart Images

Figure CN119026861A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a vehicle scheduling method, and in particular to a distributed robust optimization decision-making method for multi-task matching dynamic scheduling of an autonomous driving taxi fleet. Background Art
[0002] As a key component of intelligent transportation systems, autonomous driving technology is developing rapidly around the world. Electric and shared autonomous taxis are seen as a revolutionary advancement in future urban transportation, and have attracted much attention for their potential advantages in reducing the number of private cars, improving accessibility for vulnerable groups, reducing travel costs and improving safety. Currently, mobile travel service providers such as Waymo, Baidu Apollo and DiDi are actively promoting the pilot operation of autonomous taxi services. Current research on autonomous vehicles has covered core technical fields such as perception, decision-making, planning and control, providing basic support for the realization of autonomous taxi fleets. However, it is necessary to point out that in terms of the realization of autonomous shared taxis, in addition to complete and reliable autonomous driving technology as the basic support, a fully automated fleet dynamic scheduling management decision-making system is also an indispensable key guarantee for its large-scale commercial application.
[0003] In order to solve the above challenges, this paper takes the autonomous taxi fleet as the research object and proposes a dynamic dispatch decision method for SAEV fleets that is compatible with uncertain travel scenarios, while taking into account multiple tasks such as order acceptance, charging, regional deployment and parking. This invention can improve the operational efficiency and profitability of the SAEV fleet, improve the user's travel experience, and promote the sustainable development of autonomous taxi services.
[0004] After searching, it was found that the article "Robust Optimization Model for Empty Car Scheduling in Shared Car Systems" published in Volume 38, Issue 2 of "Control and Decision" in 2023 proposed a robust optimization model for empty car scheduling to address the supply and demand imbalance problem in the current shared car system. This model gives the uncertain passenger demand as an interval number, and at the same time limits the sum of passenger demand in time to reduce the conservatism of the model's solution and enhance the model's adaptability to changes in passenger demand.
[0005] Disadvantages: This method considers the uncertainty of user demand through a robust optimization method, which can better cope with uncertain user demand, but the output solution is relatively conservative, making the fleet scheduling solution less economical, that is, more costs need to be invested in empty vehicle scheduling to alleviate the possible imbalance between supply and demand. In addition, the proposed method does not consider the fleet order acceptance and charging scheduling tasks, and cannot be directly applied to the operation management optimization of the autonomous taxi fleet.
[0006] The article "Optimal Scheduling of Autonomous Taxis Based on Uncertain Demand" published in Volume 28, Issue 11 of "Computer Integrated Manufacturing Systems" in 2022 proposed a model-free deep reinforcement learning framework to reduce the difficulty for passengers to get a taxi during peak hours and the situation of empty taxis. This framework uses a Markov decision model for modeling, comprehensively considers the operator's revenue and customer waiting costs, and uses a policy-based deep reinforcement learning algorithm-the double-delay deep deterministic policy gradient algorithm to dispatch autonomous taxis. Experiments have proved that the algorithm can effectively cope with situations of uncertain demand.
[0007] Disadvantages: The algorithm requires a large amount of real travel log data for model pre-training, and collecting a large amount of real and effective data and training the model usually requires high data collection costs and computer computing resources. In addition, the model only considers the empty car dispatching problem of driverless taxis, and does not consider the fleet's order acceptance and charging tasks.
[0008] The article "Modeling and Optimization of Multiaction Dynamic Dispatching Problem for Shared Autonomous Electric Vehicles", published in the 2021 issue of "Journal of Advanced Transportation" with the number 1368286, aims at the order-taking, parking, charging and regional dispatching and dispatching problems of autonomous shared electric vehicles. It first models the model according to the Markov decision process, and then establishes two optimization models based on combinatorial optimization theory: short-sighted and long-sighted. Secondly, a deep Q-learning algorithm is designed to be combined with the Kuhn-Munkres algorithm to solve the global optimization of the model. The simulation results of the example show that combining combinatorial optimization theory with reinforcement learning theory is an effective solution to the optimization of the operation and management of autonomous shared electric vehicle fleets.
[0009] Disadvantages: This method takes into account the multiple dispatching tasks of vehicle order acceptance, charging, parking and regional deployment in the actual operation process of the autonomous taxi fleet, but does not consider the temporal and spatial random characteristics of user-side travel demand. Therefore, the proposed method belongs to the category of deterministic optimization. Although the algorithm output results in the numerical simulation environment perform well, the performance under actual operating conditions cannot be guaranteed.
[0010] One of the main tasks of self-driving taxis is to respond to each user's travel request, pick up the user and transport him to the destination. In addition, since the vehicle is powered by a power battery, timely charging is another key task when the vehicle power is low. In addition, due to the changes in vehicle supply and demand in different regions and time periods, supply and demand imbalance problems frequently occur, so it is necessary to perform inter-regional vehicle allocation tasks to achieve rebalancing of vehicle regional distribution.
[0011] On the one hand, the existing dispatching strategies for driverless taxi fleets only propose solutions for a single task or different task combinations, and lack the overall decision-making optimization for multiple dispatching tasks such as vehicle order acceptance, charging, parking and regional deployment. As a result, these methods are unable to provide the fleet with multiple dispatching task instructions around the clock, and it is difficult to economically and efficiently meet the operational management decision-making needs of driverless taxi fleets; on the other hand, the existing dispatching strategies for driverless taxi fleets are mostly carried out in a deterministic environment, without considering the interference of the randomness of the time and spatial distribution of user travel needs on dispatching decisions, resulting in the dispatching effect of these methods in actual scenarios cannot be guaranteed. Summary of the invention
[0012] Purpose of the invention: The purpose of the present invention is to provide a distributed robust optimization decision-making method for multi-task matching dynamic scheduling of an autonomous taxi fleet.
[0013] The present invention aims to solve the dynamic scheduling problem of autonomous taxi fleets facing uncertain user travel needs. In view of the fact that existing technologies cannot efficiently handle multiple tasks such as order acceptance, charging, regional deployment and parking of autonomous vehicles, and are difficult to adapt to the spatiotemporal uncertainty of user needs, the method of the present invention can effectively hedge the uncertainty in the fleet operation process, optimize the scheduling cost and efficiency, and improve the user travel experience through a solution strategy combining the iterative KM (IKM) algorithm, the Gurobi solver and the greedy strategy.
[0014] In view of the two shortcomings of the prior art summarized above, the present invention proposes a multi-task matching operation management decision method for autonomous driving taxi fleets, including order acceptance, charging, regional allocation and parking, which takes into account the uncertainty of user travel demand. The method aims to maximize the profit of order acceptance and minimize the vehicle mileage, and at the same time, coordinate and optimize the vehicle order acceptance task, charging task, regional allocation task and parking task. On this basis, the interference caused by the randomness of the temporal and spatial distribution of user travel demand on the scheduling decision is taken into account through distributed robust opportunity constrained programming, so as to improve the robustness of the fleet operation management decision, realize the dynamic and efficient matching of user travel demand and the temporal and spatial distribution of fleet capacity resources, and improve the user order satisfaction rate, vehicle utilization rate and system operation efficiency.
[0015] Technical solution: The distributed robust optimization decision-making method for multi-task matching dynamic scheduling of an autonomous taxi fleet described in the present invention comprises the following steps:
[0016] 1. Mathematical modeling
[0017] The operation process of the shared autonomous electric vehicles (SAEV) fleet does not involve human drivers, so the dispatch system needs to formulate real-time mission instructions for each vehicle throughout the day. Specifically, the most basic task of SAEV is to meet the travel needs of users, that is, it needs to go to the starting point of the user's trip to pick up passengers and then take them to the destination; at the same time, SAEV is driven by an electric motor, so the vehicle charging process must be considered to achieve the long-term availability of the fleet; in addition, the tidal and asymmetric nature of user travel demand has caused a serious mismatch between the fleet's capacity resources and the distribution of user travel demand, which has always been one of the main challenges for the efficient operation of mobile travel platforms. Regional deployment has been regarded as the most effective means to solve this problem by rebalancing the distribution of fleet capacity resources to meet user needs. However, the early re-arrangement of fleet capacity resources requires consideration of user travel needs at future times, which requires consideration of the uncertainty of travel demand.
[0018] In view of the practical problems faced by the SAEV fleet in the operation process mentioned above, the present invention takes the vehicle operation status classification as the entry point, takes the improvement of system operation benefits and user travel experience as the optimization orientation, and proposes the idea of using bipartite graph and network flow theory to jointly optimize the SAEV fleet's order acceptance, charging, regional deployment and parking tasks based on the matching decision characteristics of vehicles performing different tasks. Figure 1 shown.
[0019] In addition, the following assumptions are prerequisites for mathematical modeling:
[0020] (1) Assume that the dispatch system can obtain information such as the vehicle's current status (order acceptance / charging / regional deployment / parking), parking lot availability, and parking space availability in real time;
[0021] (2) Each travel order can only be served by one vehicle, and one vehicle can only serve one order at a time;
[0022] (3) The electric vehicles are of the same model, and their battery capacity, driving range and other vehicle configuration parameters are the same;
[0023] (4) The charging piles installed in the charging station are fast charging piles, and the fast charging models are unified.
[0024] Vehicle order task
[0025] Mathematically, the fleet order problem can be viewed as finding the optimal matching in a weighted bipartite graph, such as Figure 2 As shown in the figure. The entities involved in the allocation can be divided into two categories: the set of available vehicles V, and the set of orders to be matched O, which respectively represent two independent point subsets X and Y in the complete bipartite graph G = (X, Y, E, W). The connection between vehicles and orders represents the cost of different vehicles taking different orders, which is reflected in the bipartite graph as the edge set E between X and Y with weight W.
[0026] Considering that during the operation of the SAEV system, while trying to meet the travel needs of users, the system operating income should be increased as much as possible. When the number of travel requests is fixed, reducing the order cost means more operating profit. Therefore, the decision variables is set as whether vehicle i accepts order o at time t (binary variable), and minimizing the total order-taking cost in each round of matching is set as the optimization goal, as shown in formula (1), where: is the cost required for vehicle i to accept order o at time t.
[0027]
[0028] In addition, it should be ensured that each SAEV serves at most one user request at a time, and each travel request can only be satisfied by one SAEV, as shown in equations (2) and (3). Formula (4) ensures that the decision variables are binary variables.
[0029]
[0030]
[0031]
[0032] If the order waiting time exceeds T max , then the order is considered rejected, as shown in formula (5).
[0033]
[0034] Formula (6) is the battery power constraint. If the travel mileage D of order o is o Greater than the current remaining driving range of vehicle i The match is canceled.
[0035]
[0036] The revenue R brought to the system by completing each travel order o Based on mileage D o The piecewise function representation of is shown in formula (7). In the formula, base feeis a constant and α is a price factor.
[0037]
[0038] Vehicle charging tasks
[0039] Compared with the vehicle order-taking task, the vehicle charging process does not involve users. Therefore, the charging task mainly considers the vehicle power consumption cost of the fleet going to the charging station. The decision variables is set as whether vehicle i goes to charging station j for charging at time t, and the minimization of the charging cost of all vehicles in a single round of scheduling is set as the optimization goal, as shown in formula (8), where: It is the cost required for vehicle i to go to charging station j to perform charging task at time t.
[0040]
[0041] At the same time, each charging station should be limited to serve at most one SAEV at a time, and each SAEV to be charged can only be assigned to one charging station, as shown in equations (9) and (10). Formula (11) ensures that the decision variables are binary.
[0042]
[0043]
[0044]
[0045] Vehicle regional deployment tasks
[0046] The vehicle regional deployment and scheduling process can be regarded as the flow of fleet capacity resources between different regions over time, such as Figure 3 As shown, therefore, the present invention models and optimizes the vehicle regional deployment task based on network flow theory. The regional deployment decision variables are set as represents the number of available vehicles transferred from area k to area g at time t. At the same time, in order to reduce the system operation cost as much as possible, the cost minimization of each round of scheduling is set as the optimization goal, as shown in formula (12), where is the cost required to transfer a vehicle from area k to area g at time t.
[0047]
[0048] The user demand in area m at time t is set as a random variable The chance constraint form is adopted to ensure that the constraint is established at least under a certain confidence level. Specifically, after the vehicle regional allocation task is completed, the probability that the number of available vehicles in each sub-region is greater than or equal to the number of user demands should be greater than or equal to As shown in formula (13), the number of available vehicles in area m is It is given by formula (14).
[0049]
[0050]
[0051] in, is the number of available vehicles in area m at time t, is the total number of vehicles transferred into area m, as shown in formula (15) and formula (16).
[0052]
[0053]
[0054] in, is the number of available vehicles in area m at time t, and is the number of potentially available vehicles.
[0055] Combining formulas 13-16, the chance constraint (13) can be re-expressed as formula (17).
[0056]
[0057] The travel demand of users in area m is known The mean and variance of , and the second-order moment matrix is ∑ m,t , and assume that ∑ m,t is a positive definite matrix, that is, ∑ m,t > 0. Assuming that the probability distribution p of user travel demand belongs to the set Γ, the distributed robust chance constraint of vehicle regional deployment can be expressed as formula (18).
[0058]
[0059] In addition, it is considered that the vehicle regional allocation process cannot be sustained for too long to avoid the "lag effect", as shown in formula (19).
[0060]
[0061] Finally, the decision variable constraints are shown in formula (20).
[0062]
[0063] According to the strong duality property, the distributed robust chance constraint (18) is equivalent to (21).
[0064]
[0065] Vehicle parking tasks
[0066] Considering that the vehicle parking task is relatively simple and has high execution frequency and timeliness requirements, after completing the previous task, the vehicle goes directly to the nearest parking space to wait for the next round of scheduling task instructions.
[0067] 2. Algorithm Design
[0068] Iterative KM algorithm
[0069] In order to quickly solve the fleet order matching solution, the present invention proposes an iterative KM (IKM) algorithm, such as Figure 4 As shown. The algorithm iteration process is as follows: first, according to Δt2 and the waiting time of each order, the current set of orders to be matched is divided into k subsets (k = 0, 1, 2...K), and the waiting time range of the orders in the kth set is k·Δt2≤T<(k+1)·Δt2, as shown in formula (22). Then, the matching priority of all order subsets is determined, and the priority sorting principle is: the longer the waiting time, the higher the matching priority; finally, based on the KM algorithm, the subsets are solved one by one according to the matching priority to finally achieve matching optimization for all travel requests.
[0070]
[0071] Where, T max is the maximum waiting time that the user can tolerate; Δt2(0<Δt2≤T max ) is the time control factor of the “first come, first served” matching strategy. The smaller its value is, the more it can reflect the “first come, first served” matching principle.
[0072] In addition, the KM algorithm is used to solve the vehicle charging task. For the vehicle regional allocation task, considering that the model obtained after the equivalent conversion is an integer programming model, the Gurobi solver is used to solve it. For the vehicle parking task, the distance-based greedy algorithm is used to solve it.
[0073] Compared with the prior art, the present invention has the following beneficial effects:
[0074] 1. The multi-task dynamic scheduling decision method of the present invention effectively solves the limitations of traditional scheduling systems when dealing with multiple tasks in the operation of autonomous driving fleets. First, the method can respond to and optimize tasks such as order acceptance, charging, regional deployment and parking in real time, improving the operational efficiency and profitability of the fleet. Due to the use of bipartite graph maximum weight matching and network flow theory for modeling, the present invention can more accurately match vehicles and tasks, reduce resource waste, and at the same time reduce user waiting time and improve user satisfaction.
[0075] 2. The distributed robust optimization strategy of the present invention provides an effective solution to the uncertainty of the spatiotemporal distribution of user travel demand. Through distributed robust opportunity constraints, the present invention not only enhances the adaptability of the dispatching system to uncertain factors, but also ensures that efficient fleet operations can be maintained even when demand fluctuates greatly. Combining the iterative KM algorithm, Gurobi solver and greedy algorithm, the present invention enables dispatching decisions to quickly respond to real-time changes in travel demand, thereby optimizing the user experience while ensuring operational stability.
[0076] 3. Multi-task dynamic scheduling decision method: One of the core innovations of this invention is to propose a dynamic scheduling decision method that integrates multiple tasks such as order acceptance, charging, regional deployment and parking. This method uses bipartite graph maximum weight matching and network flow theory for modeling, effectively integrating different task requirements in fleet operations and achieving coordinated optimization of multiple tasks.
[0077] 4. Distributed robust optimization strategy: Another key point of the present invention is the introduction of a distributed robust optimization strategy, which hedges the uncertainty of the spatiotemporal distribution of user travel demand through distributed robust opportunity constraints. This strategy not only improves the robustness of the scheduling system and ensures stable operation under uncertain environments, but also achieves fast solutions through a customized solution strategy that combines the IKM algorithm with the Gurobi solver and the greedy strategy, meeting the needs of real-time scheduling. BRIEF DESCRIPTION OF THE DRAWINGS
[0078] Figure 1 Modeling multi-task joint optimization for SAEV fleets;
[0079] Figure 2 It is a vehicle-order matching based on a weighted complete bipartite graph;
[0080] Figure 3 To dispatch tasks for vehicle regions based on network flows;
[0081] Figure 4 Optimize the order-taking task for fleets based on the IKM algorithm;
[0082] Figure 5 The order completion status of each sub-region;
[0083] Figure 6 The full-day operating income for different areas. DETAILED DESCRIPTION
[0084] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be further described below.
[0085] The distributed robust optimization decision-making method for multi-task matching dynamic scheduling of an autonomous taxi fleet in this embodiment includes the following steps:
[0086] 1. Mathematical modeling
[0087] The operation process of the shared autonomous electric vehicles (SAEV) fleet does not involve human drivers, so the dispatch system needs to formulate real-time mission instructions for each vehicle throughout the day. Specifically, the most basic task of SAEV is to meet the travel needs of users, that is, it needs to go to the starting point of the user's trip to pick up passengers and then take them to the destination; at the same time, SAEV is driven by an electric motor, so the vehicle charging process must be considered to achieve the long-term availability of the fleet; in addition, the tidal and asymmetric nature of user travel demand has caused a serious mismatch between the fleet's capacity resources and the distribution of user travel demand, which has always been one of the main challenges for the efficient operation of mobile travel platforms. Regional deployment has been regarded as the most effective means to solve this problem by rebalancing the distribution of fleet capacity resources to meet user needs. However, the early re-arrangement of fleet capacity resources requires consideration of user travel needs at future times, which requires consideration of the uncertainty of travel demand.
[0088] In view of the practical problems faced by the SAEV fleet in the operation process mentioned above, the present invention takes the vehicle operation status classification as the entry point, takes the improvement of system operation benefits and user travel experience as the optimization orientation, and proposes the idea of using bipartite graph and network flow theory to jointly optimize the SAEV fleet's order acceptance, charging, regional deployment and parking tasks based on the matching decision characteristics of vehicles performing different tasks. Figure 1 For each sub-area, the allocation process of vehicle charging, order acceptance, regional deployment and parking tasks in each time period t includes: First, all idle vehicles are grouped into a set I t , and then filter the vehicle groups with power less than Λ into a set And assign these vehicles to J t Then, all travel orders are combined into a set O t , and allocate these orders to the remaining Vehicles, that is, the set of available vehicles for performing order-taking tasks After completing the above scheduling decision in time period t, the remaining available vehicles in each sub-area are the set Then, according to the supply and demand status of each sub-region (S t ) makes vehicle regional deployment and scheduling decisions between all sub-regions. Finally, all vehicles that have not been assigned tasks in this round of scheduling, that is, the set will be assigned to the nearest parking space (P t )Wait for the next round of task assignment.
[0089] In addition, the following assumptions are prerequisites for mathematical modeling:
[0090] (1) Assume that the dispatch system can obtain information such as the vehicle's current status (order acceptance / charging / regional deployment / parking), parking lot availability, and parking space availability in real time;
[0091] (2) Each travel order can only be served by one vehicle, and one vehicle can only serve one order at a time;
[0092] (3) The electric vehicles are of the same model, and their battery capacity, driving range and other vehicle configuration parameters are the same;
[0093] (4) The charging piles installed in the charging station are fast charging piles, and the fast charging models are unified.
[0094] Vehicle order task
[0095] Mathematically, the fleet order problem can be viewed as finding the optimal matching in a weighted bipartite graph, such as Figure 2 As shown in the figure. The entities involved in the allocation can be divided into two categories: the set of available vehicles V, and the set of orders to be matched O, which respectively represent two independent point subsets X and Y in the complete bipartite graph G = (X, Y, E, W). The connection between vehicles and orders represents the cost of different vehicles taking different orders, which is reflected in the bipartite graph as the edge set E between X and Y with weight W.
[0096] Considering that during the operation of the SAEV system, while trying to meet the travel needs of users, the system operating income should be increased as much as possible. When the number of travel requests is fixed, reducing the order cost means more operating profit. Therefore, the decision variables is set as whether vehicle i accepts order o at time t (binary variable), and minimizing the total order-taking cost in each round of matching is set as the optimization goal, as shown in formula (1), where: is the cost required for vehicle i to accept order o at time t.
[0097]
[0098] In addition, it should be ensured that each SAEV serves at most one user request at a time, and each travel request can only be satisfied by one SAEV, as shown in equations (2) and (3). Formula (4) ensures that the decision variables are binary variables.
[0099]
[0100]
[0101]
[0102] If the order waiting time exceeds T max , then the order is considered rejected, as shown in formula (5).
[0103]
[0104] Formula (6) is the battery power constraint. If the travel mileage D of order o is o Greater than the current remaining driving range of vehicle i The match is canceled.
[0105]
[0106] The revenue R brought to the system by completing each travel order o Based on mileage D o The piecewise function representation of is shown in formula (7). In the formula, base fe e is a constant and α is a price factor.
[0107]
[0108] Vehicle charging tasks
[0109] Compared with the vehicle order-taking task, the vehicle charging process does not involve users. Therefore, the charging task mainly considers the vehicle power consumption cost of the fleet going to the charging station. The decision variables is set as whether vehicle i goes to charging station j for charging at time t, and the minimization of the charging cost of all vehicles in a single round of scheduling is set as the optimization goal, as shown in formula (8), where: It is the cost required for vehicle i to go to charging station j to perform charging task at time t.
[0110]
[0111] At the same time, each charging station should be limited to serve at most one SAEV at a time, and each SAEV to be charged can only be assigned to one charging station, as shown in equations (9) and (10). Formula (11) ensures that the decision variables are binary.
[0112]
[0113]
[0114]
[0115] Vehicle regional deployment tasks
[0116] The vehicle regional deployment and scheduling process can be regarded as the flow of fleet capacity resources between different regions over time, such as Figure 3As shown, therefore, the present invention models and optimizes the vehicle regional deployment task based on network flow theory. The regional deployment decision variables are set as represents the number of available vehicles transferred from area k to area g at time t. At the same time, in order to reduce the system operation cost as much as possible, the cost minimization of each round of scheduling is set as the optimization goal, as shown in formula (12), where is the cost required to transfer a vehicle from area k to area g at time t.
[0117]
[0118] The user demand in area m at time t is set as a random variable The chance constraint form is adopted to ensure that the constraint is established at least under a certain confidence level. Specifically, after the vehicle regional allocation task is completed, the probability that the number of available vehicles in each sub-region is greater than or equal to the number of user demands should be greater than or equal to As shown in formula (13), the number of available vehicles in area m is It is given by formula (14).
[0119]
[0120]
[0121] in, is the number of available vehicles in area m at time t, is the total number of vehicles transferred into area m, as shown in formula (15) and formula (16).
[0122]
[0123]
[0124] in, is the number of available vehicles in area m at time t, and is the number of potentially available vehicles.
[0125] Combining formulas 13-16, the chance constraint (13) can be re-expressed as formula (17).
[0126]
[0127] The travel demand of users in area m is known The mean and variance of , and the second-order moment matrix is ∑ m,t , and assume that ∑ m,t is a positive definite matrix, that is, ∑ m,t> 0. Assuming that the probability distribution p of user travel demand belongs to the set Γ, the distributed robust chance constraint of vehicle regional deployment can be expressed as formula (18).
[0128]
[0129] In addition, it is considered that the vehicle regional allocation process cannot be sustained for too long to avoid the "lag effect", as shown in formula (19).
[0130]
[0131] Finally, the decision variable constraints are shown in formula (20).
[0132]
[0133] According to the strong duality property, the distributed robust chance constraint (18) is equivalent to (21).
[0134]
[0135] Vehicle parking tasks
[0136] Considering that the vehicle parking task is relatively simple and has high execution frequency and timeliness requirements, after completing the previous task, the vehicle goes directly to the nearest parking space to wait for the next round of scheduling task instructions.
[0137] 2. Algorithm Design
[0138] Iterative KM algorithm
[0139] In order to quickly solve the fleet order matching solution, the present invention proposes an iterative KM (IKM) algorithm, such as Figure 4 As shown. The algorithm iteration process is as follows: first, according to Δt2 and the waiting time of each order, the current set of orders to be matched is divided into k subsets (k = 0, 1, 2...K), and the waiting time range of the orders in the kth set is k·Δt2≤T<(k+1)·Δt2, as shown in formula (22). Then, the matching priority of all order subsets is determined, and the priority sorting principle is: the longer the waiting time, the higher the matching priority; finally, based on the KM algorithm, the subsets are solved one by one according to the matching priority to finally achieve matching optimization for all travel requests.
[0140]
[0141] Where, T max is the maximum waiting time that the user can tolerate; Δt2(0<Δt2≤T max ) is the time control factor of the “first come, first served” matching strategy. The smaller its value is, the more it can reflect the “first come, first served” matching principle.
[0142] In addition, the KM algorithm is used to solve the vehicle charging task. For the vehicle regional allocation task, considering that the model obtained after the equivalent conversion is an integer programming model, the Gurobi solver is used to solve it. For the vehicle parking task, the distance-based greedy algorithm is used to solve it.
[0143] Specific cases and results:
[0144] The data set used in this case analysis is the operation trajectory data of 1,000 new energy taxis operating in 16 districts of Shanghai from March 1, 2021 to 31, 2022, with a data collection interval of 10 seconds. After cleaning the original data set, the vehicle trip data is extracted to obtain the key information of the user's single trip, including the starting and ending locations of a single trip, the start time of the trip, the end time of the trip, and the mileage of a single trip.
[0145] 1. Fleet dispatch plan
[0146] Table 1 Simulation parameter settings
[0147] Order acceptance time interval 5minute Battery capacity 60kWh Charging interval 5minute Charging power 60kW α 3.5 yuan / km Vehicle Range 300km <![CDATA[base fee ]]> 16 yuan User waiting time 15minute soc threshold 30% Regional deployment time interval 120minute
[0148] The main parameter settings for the case analysis are shown in Table 1. The calculation results show that the total number of orders generated in 16 areas of Shanghai throughout the day was 17,245, the number of completed orders was 16,174, the number of rejected orders was 1,071, and the order satisfaction rate was 93.79%. The operating income brought by meeting user travel needs was 2,483,097 yuan. In addition, the number of dispatched vehicles brought by the execution of vehicle regional deployment tasks throughout the day was 607 times, generating a vehicle dispatch cost of 22,778 yuan. Figure 5 .
[0149] like Figure 5 As shown, in terms of the number of orders, the distribution of each sub-region is extremely uneven. The number of orders in Minhang District and Pudong District is significantly higher than that in other areas. Minhang District has the least number of orders, with no more than 100 orders throughout the day. In terms of order fulfillment rate, 13 sub-regions are above 90%, of which Hongkou District is as high as 99.26%. It is worth noting that although the number of orders in Minhang District and Pudong District is very high, travel demand is still met at a high level. These two areas are characterized by being located in or close to the center of Shanghai, with a large number of orders throughout the day and concentrated distribution. The order fulfillment rates in Fengxian District and Qingpu District are both below 80%. These two areas are geographically far away from the city center, with a small number of orders and a non-concentrated distribution.
[0150] like Figure 6As shown in the figure, the operating income of each sub-region varies greatly. Minhang District and Pudong District have the highest operating income, and the total of the two accounts for more than 50%. Among the 16 regions, only Minhang, Pudong, Baoshan, Jiading and Songjiang have an income share of more than 5%, and the remaining 11 sub-regions have a low income share, accounting for only 22% of the total operating income. This also reflects the significant differences in the spatial distribution of travel demand to a certain extent.
[0151] 2. Horizontal comparison
[0152] In order to further comprehensively verify the effectiveness of the distributed robust optimization based on hedging user travel uncertainty proposed in the present invention, the following comparative experiments were designed: Case 1, does not perform regional deployment tasks, and only performs vehicle order acceptance, charging and parking tasks. Case 2, adopts the Reactive repositioning algorithm, which was proposed by Alonso-Mora et al. The core idea is to reposition idle vehicles to areas with a larger number of rejected travel requests. Case 3, according to the idea of robust optimization, the worst-case travel demand can still be fully met. Case 4, assuming that user demand obeys Poisson distribution, and then optimizes vehicle regional deployment according to random planning chance constraints. Case 5 implements the strategy proposed by the present invention.
[0153] Table 2 Comparison of scheduling strategies
[0154]
[0155] Note: * is the optimal result of the row, and the decimal result is rounded to two decimal places.
[0156] As can be seen from Table 2, compared with case 1, the order fulfillment rates of case 2, 3, 4, and 5 have all increased to varying degrees, which shows that regional dispatching tasks can effectively improve system service capabilities and user experience. At the same time, the increase in overall operating revenue and net operating revenue for the whole day shows that although regional dispatching tasks require costs, they can still bring about an increase in net revenue. In addition, the reduction in average order mileage and the average proportion of idle vehicles, as well as the increase in the average proportion of order vehicles, all indicate that the execution of regional dispatching tasks can effectively improve supply matching and increase vehicle utilization. In general, the improvement of case 2, 3, 4, and 5 in different indicators compared to case 1 shows the potential for improvement in vehicle regional dispatching tasks in terms of user travel experience, system profitability, and fleet operating efficiency.
[0157] Compared with the deterministic optimization method case 2, the uncertain optimization method case 5 has brought greater improvements to the user's travel experience and system operation, increasing the order fulfillment rate and all-day operating income by 2.26% and 2.75% respectively, which shows the advantage of considering user uncertainty in the process of optimizing vehicle regional allocation tasks. For case 2, its poor effect is mainly attributed to the lag of the strategy. Specifically, under the setting of case 2, vehicles tend to be dispatched to areas with lower current order fulfillment rates, but at this time the order rejection event has already occurred. At the same time, the orders that are not currently satisfied may exceed the user's waiting time by the time the vehicle is dispatched and cannot be satisfied.
[0158] For the three uncertain optimization methods (case 3, 4 and 5), case 3 has the lowest order fulfillment rate of 92.86%, the lowest all-day operating revenue, and the lowest number of vehicles and vehicle scheduling costs. This shows that the random programming strategy has better economy and limited uncertainty hedging performance. Case 4 has higher order fulfillment rate and all-day operating revenue, but it generates the largest number of dispatched vehicles (610) and the largest vehicle scheduling cost (23006.24). Therefore, this strategy has good performance in hedging uncertainty, but it needs to pay the price of economy. For case 5, it brings the highest order fulfillment rate and all-day operating revenue. At the same time, it dispatches fewer vehicles than case 4, thus bringing less vehicle scheduling costs. From the final effect, case 5 generates the highest all-day operating net revenue compared with case 4 and case 3. This shows that the proposed distributed robust optimization strategy has a good balance between economy and robustness.
[0159] The above is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any technician in the relevant technical field, without departing from the scope of the technical solution of the present invention, makes any form of equivalent replacement or modification to the technical solution and technical content disclosed in the present invention, which does not depart from the content of the technical solution of the present invention and still falls within the protection scope of the present invention.
Claims
1. A distributed robust optimization decision-making method for multi-task matching dynamic scheduling of an autonomous taxi fleet, characterized in that: Decision optimization is achieved through a solution strategy that combines iterative KM algorithm, Gurobi solver and greedy strategy.
2. The distributed robust optimization decision-making method for multi-task matching dynamic scheduling of an autonomous taxi fleet according to claim 1, characterized in that: The steps include: (1) Mathematical modeling: A. Vehicle order-taking task modeling; B. Vehicle charging task modeling; C. Vehicle regional deployment task modeling; D. Vehicle parking tasks; (2) Algorithm optimization: iterative KM algorithm.
3. The distributed robust optimization decision-making method for multi-task matching dynamic scheduling of an autonomous taxi fleet according to claim 2 is characterized in that: The mathematical modeling method is as follows: For each sub-area, the allocation process of vehicle charging, order acceptance, regional deployment and parking tasks in each time period t includes: First, all idle vehicles are grouped into a set I t , and then filter the vehicle groups with power less than Λ into a set And assign these vehicles to J t available charging piles, and then combine all travel orders into a set O t , and allocate these orders to the remaining Vehicles, that is, the set of available vehicles for performing order-taking tasks After completing the above scheduling decision in time period t, the remaining available vehicles in each sub-area are the set Then, according to the supply and demand status of each sub-region (S t ) makes vehicle regional deployment and scheduling decisions between all sub-regions. Finally, all vehicles that have not been assigned tasks in this round of scheduling, i.e., the set will be assigned to the nearest parking space (P t )Wait for the next round of task assignment.
4. The distributed robust optimization decision-making method for multi-task matching dynamic scheduling of an autonomous taxi fleet according to claim 3 is characterized in that: The order-taking task modeling method in step (1) is as follows: Considering that during the operation of the SAEV system, while trying to meet the travel needs of users, the system operating income should be increased as much as possible. When the number of travel requests is fixed, reducing the order cost means more operating profit. Therefore, the decision variable is set as whether vehicle i accepts order o at time t, a binary variable, and minimizing the total order-taking cost of each round of matching is set as the optimization goal, as shown in formula (1), where: is the cost required for vehicle i to pick up order o at time t, In addition, it should be ensured that each SAEV serves at most one user request at a time, and each travel request can only be satisfied by one SAEV, as shown in equations (2) and (3). Formula (4) ensures that the decision variables are binary variables. If the order waiting time exceeds T max , then the order is considered rejected, as shown in formula (5), Formula (6) is the battery power constraint. If the travel mileage D of order o is o Greater than the vehicle's current remaining range Then cancel the match. The revenue R brought to the system by completing each travel order o Based on mileage D o The piecewise function representation of is shown in formula (7), where base fee is a constant, α is a price factor, 5. The distributed robust optimization decision-making method for multi-task matching dynamic scheduling of an autonomous taxi fleet according to claim 4 is characterized in that: The charging task modeling method in step (1) is as follows: Compared with the vehicle order-taking task, the vehicle charging process does not involve users. Therefore, the charging task mainly considers the vehicle power consumption cost of the fleet going to the charging station. The decision variables is set as whether vehicle i goes to charging station j for charging at time t, and the minimization of the charging cost of all vehicles in a single round of scheduling is set as the optimization goal, as shown in formula (8), where: It is the cost required for vehicle i to go to charging station j to perform charging task at time t. At the same time, each charging station should be limited to serve at most one SAEV at a time, and each SAEV to be charged can only be assigned to one charging station, as shown in equations (9) and (10). Formula (11) ensures that the decision variables are binary.
6. The distributed robust optimization decision-making method for multi-task matching dynamic scheduling of an autonomous taxi fleet according to claim 5, characterized in that: The regional deployment task modeling method in step (1) is as follows: The vehicle regional deployment and scheduling process can be regarded as the flow of fleet capacity resources between different regions over time. Therefore, the present invention models and optimizes the vehicle regional deployment task based on network flow theory, and the regional deployment decision variables are set as represents the number of available vehicles transferred from area k to area g at time t. At the same time, in order to reduce the system operation cost as much as possible, the cost minimization of each round of scheduling is set as the optimization goal, as shown in formula (12), where is the cost of transferring a vehicle from area k to area g at time t, The user demand in area m at time t is set as a random variable The chance constraint form is adopted to ensure that the constraint is established at least under a certain confidence level. Specifically, after the vehicle regional allocation task is completed, the probability that the number of available vehicles in each sub-region is greater than or equal to the number of user demands should be greater than or equal to As shown in formula (13), the number of available vehicles in area m is Given by formula (14), in, is the number of available vehicles in area m at time t, is the total number of vehicles transferred into area m, as shown in formula (15) and formula (16), in, is the number of available vehicles in area m at time t, and is the number of potentially available vehicles, Combining formulas 13-16, the opportunity constraint (13) can be re-expressed as formula (17), The travel demand of users in area m is known The mean and variance of , and the second-order moment matrix is ∑ m,t , and assume that ∑ mt is a positive definite matrix, that is, ∑ m,t > 0, assuming that the probability distribution p of user travel demand belongs to the set Γ, the distributed robust chance constraint of vehicle regional deployment can be expressed as formula (18): In addition, considering that the vehicle regional deployment process cannot be sustained for too long to avoid the hysteresis effect, as shown in formula (19), Finally, the decision variable constraints are shown in formula (20): According to the strong duality property, the distributed robust chance constraint (18) is equivalent to (21), 7. The distributed robust optimization decision-making method for multi-task matching dynamic scheduling of an autonomous taxi fleet according to claim 6, characterized in that: The parking task in step (1): after completing the previous task, the vehicle directly goes to the nearest parking space to wait for the next round of scheduling task instructions.
8. The distributed robust optimization decision-making method for multi-task matching dynamic scheduling of an autonomous taxi fleet according to claim 7, characterized in that: The algorithm optimization method in step (2) is as follows: First, according to △t2 and the waiting time of each order, the current set of orders to be matched is divided into k subsets (k=0, 1, 2...K). The waiting time range of orders in the kth set is k·△t2≤T<(k+1)·△t2, as shown in formula (22). Then, the matching priority of all order subsets is determined. The priority sorting principle is: the longer the waiting time, the higher the matching priority. Finally, based on the KM algorithm, the subsets are solved one by one according to the matching priority, and finally the matching optimization of all travel requests is achieved. Where, T max is the maximum waiting time that the user can tolerate; Δt2(0<Δt2≤T max ) is the time control factor of the "first come, first served" matching strategy. The smaller its value is, the more it can reflect the "first come, first served" matching principle.
Citation Information
Patent Citations
Taxi sharing billing system
CN103778666A
Pricing method in taxi sharing mode
CN111127076A
Online car-hailing order allocation method suitable for passenger waiting time distribution changing along with time
CN112288233A
Optimal scheduling method for multi-vehicle fleet delivery process
CN112686458A
Passenger and taxi matching method based on fuzzy logic controller
CN115099569A
Cited By
Hierarchical collaborative optimization method for operation strategy of automatic driving taxi
CN122198274A
