An autonomous taxi fleet operation management decision-making method considering demand space-time uncertainty

Abstract

The application provides a kind of automatic driving taxi fleet operation management decision method considering demand space-time uncertainty, belongs to traffic control technical field, including the following steps: 1, information input;2, model building;3, model solution;4, scheme output;The application aims to maximize order profit and minimize vehicle mileage as the goal, while making overall decision optimization to vehicle order task, charging task, regional deployment task and parking task.On this basis, through random chance constraint programming, the randomness of user travel demand time and space distribution is considered to interfere with scheduling decisions, to improve the robustness of fleet operation management decision, to realize the dynamic and efficient matching of user travel demand and fleet resource space-time distribution, to improve the user order satisfaction rate, vehicle utilization rate and system operation efficiency of shared electric travel service.

Description

Technical Field

[0001] This invention belongs to the field of traffic control technology, and in particular relates to a decision-making method for the operation and management of an autonomous taxi fleet that takes into account the spatiotemporal uncertainty of demand. Background Technology

[0002] Electrification, automation, connectivity, and sharing technologies are flourishing in urban transportation systems. Against this backdrop, autonomous taxis—a natural product of the close integration of these four technologies—have emerged. Compared to traditional taxis, autonomous driving eliminates the need for passengers to drive or search for parking spaces, and it also offers greater flexibility in choosing pick-up locations. Furthermore, in a shared model, autonomous taxis can meet the travel needs of more passengers with a smaller fleet size, significantly improving vehicle utilization and reducing labor costs.

[0003] However, since autonomous taxis operate without human drivers and lack decision-making capabilities, the back-end dispatch system needs to provide each vehicle with 24 / 7 dispatch instructions during fleet operation. In other words, the dispatch center needs to comprehensively consider real-time user travel demand, vehicle charging needs, and regional supply and demand balance to optimize the overall decision-making for multiple dispatch tasks, including fleet order acceptance, charging, parking, and regional allocation.

[0004] The diverse scheduling tasks make the decision-making optimization problem of autonomous taxis more complex than that of ride-hailing, traditional taxis, and car-sharing. Furthermore, user travel demand during the scheduling process exhibits spatiotemporal randomness. Therefore, achieving coordinated decision-making optimization for multiple scheduling tasks of autonomous taxi fleets under scenarios with uncertain user travel demand is of great significance for further improving the user's intelligent travel experience, reducing enterprise operating costs, and promoting the sustainable development of autonomous electric shared mobility service models.

[0005] A search revealed that the article "Robust Optimization Model for Empty Car Scheduling in Car-Sharing 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-demand imbalance problem in current car-sharing systems. This model gives the uncertain passenger demand as an interval number and restricts the sum of passenger demand over time to reduce the conservatism of the model's solution and enhance the model's adaptability to changes in passenger demand.

[0006] Disadvantages: While this method addresses the uncertainty of user demand through robust optimization, effectively handling such demands, its output solution is relatively conservative. This results in poor economic efficiency for the fleet dispatching scheme, requiring more investment in empty vehicle dispatching to mitigate potential supply-demand imbalances. Furthermore, the proposed method does not consider fleet order acceptance and charging dispatching tasks, making it unsuitable for direct application in the operational management optimization of autonomous taxi fleets.

[0007] The article "Optimal Scheduling of Driverless Taxis Based on Uncertain Demand," published in Volume 28, Issue 11 of Computer Integrated Manufacturing Systems in 2022, proposes a model-free deep reinforcement learning framework to reduce the difficulty of hailing 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 the customer's waiting cost, and uses a policy-based deep reinforcement learning algorithm—the dual-delay deep deterministic policy gradient algorithm—to schedule driverless taxis. Experiments show that the algorithm can effectively cope with uncertain demand situations.

[0008] Disadvantages: This algorithm requires massive amounts of real-world travel log data for model pre-training, and collecting large amounts of real and valid data for model training typically requires high data acquisition costs and computing power resources. Furthermore, this model only considers the empty vehicle dispatching problem for driverless taxis, and does not consider fleet order acceptance and charging tasks.

[0009] The article "Modeling and Optimization of Multiaction Dynamic Dispatching Problem for Shared Autonomous Electric Vehicles," published in the *Journal of Advanced Transportation*, 2021 (Issue 2021, No. 1368286), addresses the problem of order acceptance, parking, charging, and regional dispatching for autonomous shared electric vehicles. First, it models the problem using Markov decision processes. Then, it establishes two optimization models based on combinatorial optimization theory: short-sighted and long-sighted models. Next, it designs a combination of a deep Q-learning algorithm and the Kuhn-Munkres algorithm to solve the global optimization problem of the model. Simulation results demonstrate that combining combinatorial optimization theory with reinforcement learning theory is an effective solution for optimizing the operation and management of autonomous shared electric vehicle fleets.

[0010] Disadvantages: This method considers multiple scheduling tasks in the actual operation of an autonomous taxi fleet, including vehicle order acceptance, charging, parking, and regional dispatch. However, it does not consider the spatiotemporal randomness of user travel demand. Therefore, the proposed method falls into the category of deterministic optimization. Although the algorithm outputs good results in numerical simulation environments, its performance under actual operating conditions cannot be guaranteed.

[0011] One of the main tasks of autonomous taxis is to respond to each user's travel request, picking them up and transporting them to their destination. Additionally, since the vehicles are powered by batteries, timely charging when the battery is low is another critical task. Furthermore, due to the frequent fluctuations in vehicle supply and demand across different regions and time periods, imbalances in supply and demand occur frequently, necessitating the execution of inter-regional vehicle dispatching tasks to rebalance the regional distribution of vehicles.

[0012] On the one hand, existing autonomous taxi fleet dispatch strategies only offer targeted solutions for single tasks or different combinations of tasks, lacking comprehensive decision-making optimization for multiple dispatch tasks such as vehicle order acceptance, charging, parking, and regional allocation. As a result, these methods cannot provide fleets with multi-task dispatch instructions around the clock, making it difficult to economically and efficiently meet the operational management decision-making needs of autonomous taxi fleets. On the other hand, existing autonomous taxi fleet dispatch strategies are almost all conducted in deterministic environments, that is, they do not consider the interference and impact of the randomness of the time and spatial distribution of user travel demand on dispatch decisions, resulting in the dispatch effectiveness of these methods in real-world scenarios being unreliable. Summary of the Invention

[0013] The purpose of this invention is to provide a decision-making method for the operation and management of an autonomous taxi fleet that takes into account the spatiotemporal uncertainty of demand, characterized by the following steps:

[0014] S1: Input model information, including passenger requests, vehicle status, station distribution, and regional supply and demand.

[0015] S2: The vehicle charging task and the vehicle order receiving task are modeled using bipartite graph maximum weight matching; the vehicle area allocation task is modeled using stochastic programming with chance constraints; and the vehicle parking task is modeled using linear mathematical programming.

[0016] S3: Solve the model constructed in S2;

[0017] S4: Based on model solving, output real-time scheduling instructions for autonomous taxis to accept orders, charge, allocate areas, and park, and complete the operation and management decisions of autonomous taxi fleets that take into account the spatiotemporal uncertainty of demand.

[0018] Furthermore, in S1, the passenger request is specifically defined as follows: the user travel needs between each operating sub-region are set to be independent of each other, and if the passenger is not assigned an available vehicle by the system within 15 minutes after placing an order, the passenger cancels the order, that is, the travel request is rejected by the platform.

[0019] The specific vehicle status is as follows: each user's travel request is served by one vehicle, and one vehicle serves only one user's travel request at a time; one charging gun charges only one vehicle at a time.

[0020] The specific regional supply and demand situation is as follows: the tasks of fleet charging, order taking and parking dispatch are set to be carried out independently in each sub-region, while the task of vehicle regional allocation is carried out between sub-regions.

[0021] Furthermore, in S2, the vehicle charging task and the vehicle order-taking task are modeled as a bipartite graph maximum weight matching problem;

[0022] For the vehicle charging task: The optimization objective is to allocate low-battery vehicles to nearby charging stations within each time period, and to minimize the total charging mileage of all vehicles. This is expressed by Formula 1:

[0023]

[0024] in, For vehicles The vehicle will arrive at the charging station at time t. The distance required for charging; For vehicles Do you want to go to a charging station? Charge;

[0025] To ensure that each charging station serves at most one taxi at a time, and that each taxi is allocated only one charging station, this is expressed by Formulas 2 and 3. Formula 2 is expressed as follows:

[0026]

[0027] Formula 3 is expressed as:

[0028]

[0029] The constraint on the values ​​of the decision variables is expressed by Formula 4 as follows:

[0030] For vehicle order-taking tasks, when allocating available vehicles for order-taking tasks, maximizing total order profit is set as the optimization objective, which is expressed by Formula 5:

[0031]

[0032] in, For vehicles The order is completed at time t. The revenue generated from the orders; Available vehicles Is it related to travel orders? Matching; The calculation formula is expressed as Formula Six, which is as follows:

[0033]

[0034] in, For order revenue; The order-receiving cost, specifically the vehicle's electricity cost, is set by a coefficient ε relating the vehicle's electricity cost to its mileage. Linear correlation, expressed by Formula 7, is as follows:

[0035]

[0036] Where ε represents the power consumption per unit kilometer of autonomous taxis;

[0037] Formulas 8 and 9 ensure that in vehicle order-taking tasks, each vehicle serves at most one user request at a time, and each user's travel request is served by only one vehicle. Formula 8 is expressed as:

[0038]

[0039] Formula 9 is expressed as:

[0040]

[0041] The decision variables are ensured to be binary variables by using constraints, which are expressed by Equation 10:

[0042] Furthermore, in S2, during the vehicle regional dispatch task, to meet the travel requests of users in each sub-region in future time periods, the optimization objective is to minimize the total mileage of all dispatched vehicles, as expressed by Formula 11:

[0043]

[0044] in, Whether to allocate available vehicles from region k to region g at time t; Let be the number of available vehicles that can be transferred from region k to region g at time t; Let t be the distance the vehicle needs to travel from region k to region g at time t.

[0045] After the regional allocation task is completed, the probability that the number of available vehicles in all regions is greater than or equal to the number of user travel requests is greater than or equal to the following, can be expressed by Formula Twelve:

[0046]

[0047] in, This can be expressed by Formula Thirteen as follows:

[0048]

[0049] in, The number of potentially available vehicles in region m at time t is expressed by Formula 14:

[0050]

[0051] in, Let m be the total number of potentially available vehicles in region m at time t; Let be the probability that a vehicle in region m at time t is currently performing an order-taking task will become a usable vehicle in the next time period; Let m be the number of vehicles performing order-taking tasks in region m at time t. Let be the probability that a vehicle performing a charging task in region m at time t will become a usable vehicle in the next time period; Let m be the number of vehicles charging in region m at time t. Let be the number of vehicles allocated to region m at time t, which can be expressed by formula 15:

[0052]

[0053] Based on formulas 12 to 15, the constraints for vehicle area allocation are expressed by formula 16:

[0054]

[0055] Decision variable constraints and These are expressed using Formula 17 and Formula 18 respectively. Formula 17 is expressed as follows: Formula 18 is expressed as:

[0056] Furthermore, in S2, the vehicle parking task is as follows: after completing the decision-making process for vehicle order acceptance, vehicle charging, and vehicle area allocation, all remaining vehicles that have not been assigned a task in this round of scheduling will autonomously move to the nearest parking space to wait for the next round of task allocation.

[0057] Furthermore, S3 is as follows: For the fleet charging and order-taking scheduling task based on bipartite graph maximum weight matching, the Kuhn-Munkres algorithm is used to solve the optimal matching scheme between low-battery vehicles and charging piles, as well as the optimal matching scheme between available vehicle groups and passenger travel orders; For the stochastic chance constrained programming model of vehicle regional allocation, after transforming it into a deterministic model, the Gurobi solver is used to obtain the optimal regional allocation scheduling scheme. The derivation of the equivalent transformation specifically includes the following steps:

[0058] S31: Lemma:

[0059] The number of available vehicles in region m after the regional dispatching task ends at time t is: Furthermore, the user demand in region m during time period t is a random variable. Its probability distribution function is Then probability measure It is valid if and only if in

[0060] S32: Proof:

[0061] Based on S31 settings This can be expressed by Formula Nineteen:

[0062]

[0063] For a given confidence level There must exist a number For Equation 20 to hold true, Equation 20 is expressed as: At the same time, formula 21 always holds true, and equation 21 can be expressed as:

[0064]

[0065] Therefore, based on Formula 20 and Formula 21, we have Formula 22, which is expressed as:

[0066]

[0067] in, for The inverse function; at the same time, the solution to Formula 22 is not unique. When the solution is not unique, the smallest solution is chosen, as shown in Formula 23, which is expressed as:

[0068]

[0069] Then the definite equivalent form of formula nineteen is: Q.E.D. of the lemma;

[0070] Based on the lemma, the chance constraint equation, i.e., Equation 16, is equivalently expressed through Equation 24 as follows:

[0071]

[0072] Compared with the prior art, the beneficial effects of the present invention are mainly reflected in:

[0073] 1. This invention comprehensively considers the fleet's order acceptance, charging, parking, and regional dispatch tasks for decision optimization, making the proposed fleet operation management decision-making method more widely applicable. Through adaptive adjustments, the dispatch strategy proposed in this invention can cover supply and demand optimization matching scenarios such as ride-hailing, car-sharing, and food delivery.

[0074] 2. This invention characterizes the randomness of user travel demand through the Poisson probability distribution function and constructs opportunity constraints to flexibly address the interference caused by the uncertainty of user travel demand in the next period on the optimization of fleet regional dispatching tasks. It then constructs a fleet multi-task matching dynamic scheduling stochastic programming opportunity constraint model and further proves the existence of its deterministic equivalent form through rigorous mathematical derivation. This enables fast, efficient, and stable solutions to multi-task scheduling tasks of autonomous taxi fleets, including order taking, charging, regional dispatching, and parking, effectively improving the robustness of fleet operation management decisions. Attached Figure Description

[0075] Figure 1 The present invention provides the implementation steps of an autonomous taxi fleet operation management decision-making method that takes into account the spatiotemporal uncertainty of demand.

[0076] Figure 2 This invention provides a decision-making framework for autonomous taxi fleet operation management that takes into account the spatiotemporal uncertainty of demand.

[0077] Figure 3 This is a schematic diagram showing the location of the study area and the results of the area division in the embodiment.

[0078] Figure 4 This is a bar chart comparing the daily order fulfillment rate and operating revenue of the seven sub-regions in the example. Detailed Implementation

[0079] The following will describe in more detail, with reference to the schematic diagram, a method for decision-making in the operation and management of an autonomous taxi fleet that takes into account the spatiotemporal uncertainty of demand. The diagram illustrates a preferred embodiment of the invention. It should be understood that those skilled in the art can modify the invention described herein while still achieving the beneficial effects of the invention. Therefore, the following description should be understood as being of general knowledge to those skilled in the art and is not intended to limit the invention.

[0080] In this invention, the parameters of each formula are explained as follows:

[0081] parameter:

[0082] vehicle The vehicle will arrive at the charging station at time t. The distance (km) required for charging;

[0083] vehicle The order is completed at time t. Revenue generated from orders (in RMB);

[0084] vehicle The order is completed at time t. The revenue generated from these orders (in RMB);

[0085] vehicle The order is completed at time t. Mileage traveled (km);

[0086] ε: Electricity consumption per kilometer for autonomous taxis (kWh / km);

[0087] The number of user travel demands in region m at time t (random variable);

[0088] The confidence level that the vehicle regional allocation opportunity constraint in region m at time t is true;

[0089] The number of available vehicles (in vehicles) in region m at time t after the completion of the vehicle regional allocation task; The total number of available vehicles (vehicles) in region m at time t;

[0090] The total number of vehicles (units) allocated to area m at time t;

[0091] The total number of potentially available vehicles (vehicles) in region m at time t;

[0092] The number of vehicles (units) executing order-taking tasks in region m at time t;

[0093] The number of vehicles charging in region m at time t;

[0094] The probability that a vehicle in region m at time t is performing an order-taking task will become a usable vehicle in the next time period;

[0095] The probability that a vehicle performing a charging task in region m at time t will become a usable vehicle in the next time period;

[0096] The probability distribution function of travel demand of users in region m at time t;

[0097] The maximum travel demand of users in region m at time t;

[0098] gather:

[0099] M = {1,…m,…M}: Subregion numbering for the study;

[0100] The set of available vehicles in region m at time t;

[0101] The set of vehicles waiting to be charged in region m at time t;

[0102] The set of available charging stations in region m at time t;

[0103] The set of travel orders in region m at time t;

[0104] The set of vehicles executing order-taking tasks in region m at time t;

[0105] The set of surplus vehicles in region m at time t;

[0106] Decision variables:

[0107] vehicle Do you want to go to a charging station? Charge;

[0108] Available vehicles Is it related to travel orders? Matching;

[0109] Whether available vehicles can be transferred from region k to region g at time t;

[0110] The number of available vehicles transferred from region k to region g at time t.

[0111] like Figure 1 As shown, an autonomous taxi fleet operation management decision-making method that takes into account the spatiotemporal uncertainty of demand mainly includes four steps: information input, model building, model solving, and solution output.

[0112] Step 1: Model Assumptions

[0113] 1) Assume that user travel demand is independent of each operating sub-region;

[0114] 2) Each user's travel request can be served by one vehicle, and one vehicle can only serve one user's travel request at a time. Similarly, one charging gun can only serve one vehicle at a time.

[0115] 3) If a user is not assigned an available vehicle by the system within 15 minutes after placing an order, the user cancels the order, which means that the travel request is rejected by the platform.

[0116] 4) The tasks of fleet charging, order taking and parking dispatch are carried out independently in each sub-region, while the task of vehicle regional dispatch is carried out between sub-regions.

[0117] Step 2: Model Building

[0118] Assume the entire operating area of ​​the autonomous taxi fleet is divided into M sub-regions. The fleet dispatching process in each time period mainly involves two decision-making processes: one is assigning charging tasks to vehicles with insufficient battery power, and the other is assigning order acceptance, regional dispatch, and parking tasks to vehicles with sufficient battery power (available vehicles). The vehicle charging and order acceptance tasks have a one-to-one matching decision characteristic, and are therefore modeled as a bipartite graph maximum weight matching problem. The decision for vehicle regional dispatch needs to be based on the current spatial distribution of the fleet and the spatial distribution of user travel in the next time period. Considering the uncertainty of user travel requests in the next time period, this invention uses a stochastic programming chance constraint method to model the vehicle regional dispatch task to address the randomness of user travel requests. The specific modeling process is as follows:

[0119] 1): Vehicle charging task

[0120] The vehicle charging task is modeled as a bipartite graph maximum weight matching problem. In each time period, low-battery vehicles are assigned to nearby charging stations, and the optimization objective is to minimize the total charging mileage of all vehicles, as shown in Equation (1).

[0121]

[0122] Meanwhile, it should be ensured that each charging pile serves at most one taxi at a time, and each taxi is assigned only one charging pile, as shown in formulas (2) and (3). Formula (4) is the constraint on the value of the decision variable.

[0123]

[0124]

[0125]

[0126] 2): Vehicle order taking task

[0127] The vehicle order-taking task is modeled as a bipartite graph maximum weight matching problem. When allocating available vehicles for order-taking tasks, maximizing the total order profit is set as the optimization objective, as shown in formula (5).

[0128]

[0129] In the formula, Revenue from orders is equal to the revenue from receiving orders. Subtract order processing costs As shown in formula (6).

[0130]

[0131] In the formula, the order acceptance cost The main cost is the vehicle's electricity consumption, and it is assumed that its pass coefficient ε is related to the vehicle's mileage. Linear correlation, as shown in formula (7).

[0132]

[0133] Furthermore, the order-taking task should ensure that each vehicle serves at most one user request at a time, and that each user's travel request is served by only one vehicle, as shown in formulas (8) and (9). Constraint (10) ensures that the decision variables are binary variables.

[0134]

[0135]

[0136]

[0137] 3): Vehicle regional dispatch task

[0138] The frequent occurrence of regional supply and demand imbalances due to variations in user travel requests across different regions and time periods necessitates inter-regional vehicle allocation to rebalance vehicle distribution and better meet user travel requests in each sub-region during future periods. The optimization objective is set as minimizing the total mileage of all dispatched vehicles, as shown in formula (11).

[0139]

[0140] Since vehicle allocation within a region needs to be based on future travel demand, the uncertainty of user demand must be considered. The user travel demand in region m during the next time period is set as a random variable. Furthermore, opportunity constraints are used to ensure that the constraints hold at least at a certain confidence level. Specifically, after the regional allocation and scheduling task is completed, the probability that the number of available vehicles in all regions is greater than or equal to the number of user travel requests should be greater than or equal to... As shown in formula (12).

[0141]

[0142]

[0143] in, It is the number of potential available vehicles in region m at time t, as shown in formula (14). It is the number of vehicles allocated to region m at time t, as shown in formula (15).

[0144]

[0145]

[0146] Combining formulas (12) to (15), the constraints for vehicle area allocation can be obtained as follows:

[0147]

[0148] Finally, the decision variable constraints are shown in equations (17) and (18).

[0149]

[0150]

[0151] 4): Parking task

[0152] After completing the above order acceptance, charging, and regional allocation and scheduling decision-making process, all remaining vehicles that have not been assigned a task in this round of scheduling will autonomously move to the nearest parking space to wait for the next round of task allocation.

[0153] Step 3: Algorithm Design

[0154] To formulate optimal task instructions for charging, order acceptance, regional dispatching, and parking for autonomous taxis in real time and efficiently for each time period, based on the optimization model constructed above, the following solution strategy is designed:

[0155] For fleet charging and order-taking scheduling tasks based on bipartite graph maximum weight matching, the Kuhn-Munkres (KM) algorithm is used to solve the optimal matching scheme between low-battery vehicles and charging stations, and the optimal matching scheme between available vehicle groups and passenger travel orders. For the stochastic chance-constrained programming model of vehicle regional allocation, after equivalent transformation to obtain a deterministic model, the Gurobi solver is used to obtain the optimal regional allocation scheduling scheme, such as... Figure 2 As shown.

[0156] Considering that the stochastic chance-constrained programming model constructed in Chapter 4.2.3 is difficult to solve directly, this invention combines the characteristics of the chance constraint formula (16), and after transforming it to obtain a deterministic equivalent form (integer linear programming problem), it uses Gurobi to solve it. The specific derivation process of the transformation is as follows:

[0157] Lemma 1

[0158] Assume that after the regional dispatching task ends at time t, the number of available vehicles in region m is . Furthermore, the user demand in region m during time period t is a random variable. Its probability distribution function is Then probability measure It is valid if and only if in

[0159] prove:

[0160] Based on the above assumptions, It can be represented as formula (19).

[0161]

[0162] Obviously, for each given confidence level There must exist a number (Possibly multiple or ∞) make the following formula true.

[0163]

[0164] If a larger number is used instead probability It will increase, therefore If and only if

[0165] At the same time, note that the following equation always holds true:

[0166]

[0167] Therefore, combining formula (20) and formula (21) we have:

[0168]

[0169] in yes The inverse function of .

[0170] At the same time, sometimes the solution to equation (22) is not unique, that is, the function It may have multiple values. In this case, choose the smallest solution, as shown in formula (23).

[0171]

[0172] The definitive equivalent form of this pattern (19) is: The lemma is now complete.

[0173] Based on Lemma 1, the chance constraint formula (16) can be equivalently expressed as follows:

[0174]

[0175] Step 4: Solution Output

[0176] Based on the model solution, the system outputs real-time multi-task dispatch instructions for autonomous taxis, including order acceptance, charging, regional allocation, and parking, thus enabling autonomous taxi fleet operation and management decisions that take into account the spatiotemporal uncertainties of demand.

[0177] The method of the present invention is further explained below through specific embodiments.

[0178] Example

[0179] The case study area is located in Hangzhou, Zhejiang Province, China, such as... Figure 3 As shown. The dataset used consists of user travel order logs generated by an electric taxi operating company from January 1, 2019 to August 22, 2019, totaling 219,611 records. The vehicle operating area includes Xihu District, Gongshu District, Shangcheng District, Xiaoshan District, and Binjiang District.

[0180] To accommodate the multi-task scheduling requirements of vehicles, the case study area was divided into seven regular hexagons with a radius of 10km, such as... Figure 3 As shown. In addition, the main parameters used in the case analysis process and their values ​​are as follows:

[0181] 1) At the initial moment, there were 126 autonomous taxis in the entire study area, randomly and evenly distributed in 7 sub-regions;

[0182] 2) The vehicle's battery capacity and maximum range are set at 50kWh and 300km respectively, and the charging power is 50kW;

[0183] 3) The time interval between fleet order acceptance and charging tasks is set to 5 minutes; the time interval between regional dispatch and scheduling tasks is set to 120 minutes.

[0184] 4) The remaining battery power threshold for determining whether a vehicle needs charging is set at 30% to ensure that the vehicle has enough remaining driving range to reach a charging station for charging.

[0185] Spatiotemporal distribution of travel:

[0186] Based on the starting location of historical travel orders, all orders were divided into 7 sub-regions. Simultaneously, the system's daily operating time was divided into 12 time intervals (each interval being 2 hours), and the average (λ) and maximum (K) number of user trips in each sub-region within each time interval were calculated. The results are shown in Table 1.

[0187]

[0188]

[0189] Table 1

[0190] Assume that the number of user travel requests in each time period follows a Poisson distribution, as shown in formula (25).

[0191]

[0192] Based on the statistical results of the number of user travel demands in each sub-region, user travel order data for each sub-region in a day is generated by random sampling. This data is then used as input data. The model and algorithm proposed in this invention are used to optimize the overall planning of fleet charging, order acceptance, parking, and regional dispatching tasks. The specific results are as follows.

[0193] Vehicle dispatching scheme

[0194] The algorithm output shows that a total of 3,218 user orders were generated throughout the entire study area, and 242 user orders were rejected. The order satisfaction rate was 92.48%, and the total revenue generated from the orders was 184,176.55 yuan.

[0195] A total of 2 vehicle regional dispatching tasks were carried out throughout the day, with a total of 19 vehicles dispatched (16 vehicles were dispatched in the 5th round and 3 vehicles were dispatched in the 7th round), generating a total vehicle dispatching cost of 215.43 yuan.

[0196] Order fulfillment status and operating revenue in the 7 sub-regions are as follows: Figure 4As shown, sub-region 7 has the highest total number of orders and generates the most operating revenue, making it a clear hot spot for demand. However, its order fulfillment rate is relatively low (88%) compared to other sub-regions, indicating a supply shortage. Furthermore, it should be noted that sub-region 3 has the fewest orders throughout the day, thus it is considered a cold spot for demand; however, its order fulfillment rate is also the lowest (87%), therefore it is also considered a supply shortage area.

[0197] The order fulfillment rates in regions 4, 5, and 6 are relatively high, and therefore are considered to be in a state of supply and demand balance. The order fulfillment rates in regions 1 and 2 are significantly higher than the other five regions, because they are considered to be in a state of supply exceeding demand.

[0198] Horizontal comparison verification

[0199] To fully verify the effectiveness of the strategy proposed in this invention, four cases were implemented simultaneously. Case 1 only executed vehicle order acceptance, charging, and parking tasks, without considering vehicle regional dispatch tasks between sub-regions; Cases 2, 3, and 4 all considered fleet order acceptance, charging, regional dispatch, and parking tasks simultaneously. The difference is that Case 2 optimized the regional dispatch scheduling task according to the principle of even distribution of fleet capacity resources. Case 3 implemented the formulation of fleet regional dispatch scheduling tasks through robust optimization, thereby ensuring that the travel needs of users in each sub-region can still be met even in the worst case. Case 4 implemented the dispatch strategy proposed in this invention, and the comparison results are shown in Table 2.

[0200]

[0201] Table 2 Comparison of the effects of four scheduling strategies

[0202] As shown in Table 2, Strategy 1 did not execute vehicle regional dispatching tasks, thus generating no dispatched vehicle quantity or dispatching costs. However, it had the highest number of rejected orders, and the lowest order fulfillment rate and order revenue. The main reason for this is that, under normal circumstances, there is a severe mismatch between user travel demand and the spatiotemporal distribution of fleet capacity resources. Some sub-regions are in a state of oversupply while others are in a state of undersupply. Therefore, long-term implementation of Strategy 1 means a consistently low level of user order fulfillment rate and fleet utilization rate. This is not conducive to cultivating user stickiness, nor is it conducive to the efficient operation of the fleet system. Therefore, Strategy 1 is considered unsustainable.

[0203] Compared to Strategy 1, Strategy 2 performed slightly better, increasing the system's order revenue by 3144.4 with a vehicle dispatching cost of 120.24. However, because the optimization of regional dispatching tasks in Strategy 2 did not consider the travel characteristics of users in different regions, 341 travel orders were still rejected. Therefore, Strategy 2 is considered beneficial but inefficient.

[0204] Strategy 3 further improved order fulfillment rate and total revenue, but at the cost of dispatching the most vehicles and incurring the highest dispatch costs. This is because Strategy 3, based on the conservative nature of robust optimization considering the uncertainty of user travel demand, resulted in some invalid vehicle dispatch routes in its dispatch scheme. Therefore, Strategy 3 is considered to be less economical.

[0205] In contrast, Strategy 4, which flexibly considers users' uncertain needs through opportunity constraints, yielded the highest order revenue (an increase of 7689.3% compared to case 1) and the highest order fulfillment rate (an increase of 3.33% compared to case 1) with fewer vehicles and lower scheduling costs. Therefore, Strategy 4 is considered to be low-cost and efficient.

[0206] In summary, the comparative results demonstrate the wisdom of comprehensively considering multiple tasks in scheduling decisions, which can simultaneously improve order fulfillment rate and platform operating revenue, thereby achieving a win-win situation for both users and operators. On the other hand, the comparative results also show that, compared with robust optimization, the stochastic chance-constrained planning proposed in this invention, which considers the uncertainty of user demand, can bring more order revenue with less scheduling cost, thus possessing both good uncertainty resistance and good economic efficiency.

[0207] The above are merely preferred embodiments of the present invention and do not constitute any limitation on the present invention. Any equivalent substitutions or modifications made by those skilled in the art to the technical solutions and content disclosed in the present invention without departing from the scope of the present invention shall be deemed to have remained within the protection scope of the present invention.

Claims

1. A decision-making method for the operation and management of an autonomous taxi fleet that takes into account the spatiotemporal uncertainty of demand, characterized in that, Includes the following steps: S1: Input model information, including passenger requests, vehicle status, station distribution, and regional supply and demand. S2: The vehicle charging task and the vehicle order receiving task are modeled using bipartite graph maximum weight matching; the vehicle area allocation task is modeled using stochastic programming with chance constraints; and the vehicle parking task is modeled using linear mathematical programming. S3: Solve the model constructed in S2; S4: Based on model solving, output real-time multi-task dispatch instructions for autonomous taxis to accept orders, charge, regional allocation and parking, and complete the operation and management decision of autonomous taxi fleet taking into account the spatiotemporal uncertainty of demand. S3 involves: for fleet charging and order-taking scheduling tasks based on bipartite graph maximum weight matching, the Kuhn-Munkres algorithm is used to solve for the optimal matching scheme between low-battery vehicles and charging piles, as well as the optimal matching scheme between available vehicle groups and passenger travel orders; for the stochastic chance constrained programming model of vehicle regional allocation, after performing an equivalent transformation to obtain a deterministic model, the Gurobi solver is used to obtain the optimal regional allocation scheduling scheme. The equivalent transformation derivation specifically includes the following steps: S31: Lemma: Set time After the regional allocation and dispatch task is completed, the region The number of available vehicles is and the region During the period User demand is a random variable Its probability distribution function is Then the probability measure It is valid if and only if ,in ; S32: Proof: Based on S31 settings This can be expressed by Formula Nineteen as follows: ： ； For a given confidence level There must exist a number To make formula twenty true, formula twenty is expressed as: Meanwhile, Formula 21 always holds true, and Formula 21 is expressed as: ； Therefore, based on Formula 20 and Formula 21, we have Formula 22, which is expressed as: ； in, for The inverse function; meanwhile, the solution to Formula 22 is not unique. When the solution is not unique, the smallest solution is selected, as shown in Formula 23, which is expressed as: ； Then the definite equivalent form of formula nineteen is: The lemma is now proved. Based on the lemma, the chance constraint equation, i.e., Equation 16, is equivalently expressed through Equation 24 as follows: 。 2. The autonomous taxi fleet operation management decision-making method considering the spatiotemporal uncertainty of demand as described in claim 1, characterized in that, In S1, the passenger request is specifically defined as follows: the user travel needs between each operating sub-region are set to be independent of each other, and if the passenger is not assigned an available vehicle by the system within 15 minutes after placing an order, the passenger cancels the order, that is, the travel request is rejected by the platform. The vehicle status is specifically defined as follows: each user's travel request is served by one vehicle, and one vehicle serves only one user's travel request at a time; one charging gun charges only one vehicle at a time. The specific supply and demand situation in the region is as follows: the tasks of fleet charging, order taking and parking dispatch are set to be carried out independently in each sub-region, while the task of vehicle regional allocation is carried out between sub-regions.

3. The autonomous taxi fleet operation management decision-making method considering the spatiotemporal uncertainty of demand as described in claim 2, characterized in that, In S2, the vehicle charging task and the vehicle order receiving task are modeled as a bipartite graph maximum weight matching problem. For the vehicle charging task: The optimization objective is to allocate low-battery vehicles to nearby charging stations within each time period, and to minimize the total charging mileage of all vehicles, as expressed by Formula 1. Formula 1 is as follows: ； in, For vehicles At any moment Drive to charging station The distance required for charging; For vehicles Do you want to go to a charging station? Charge; To ensure that each charging station serves at most one taxi at a time, and that each taxi is allocated only one charging station, this is expressed by Formulas 2 and 3, where Formula 2 is expressed as: ； Formula 3 is expressed as follows: ； The constraint on the values ​​of the decision variables is expressed by Formula 4 as follows: ; For vehicle order-taking tasks, when allocating available vehicles for order-taking tasks, maximizing total order profit is set as the optimization objective, which is expressed by Formula 5: ； in, For vehicles At any moment Complete the order The revenue generated from the orders; Available vehicles Is it related to travel orders? Matching; The calculation formula is expressed as Formula Six, which is as follows: ； in, For order revenue; To determine the order acceptance cost, specifically the vehicle's electricity consumption cost, a pass / fail factor for the vehicle's electricity consumption cost is set. Vehicle mileage Linear correlation, expressed by Formula 7, is as follows: ； in, Electricity consumption per unit kilometer for autonomous taxis; Formulas 8 and 9 ensure that in vehicle order-taking tasks, each vehicle serves at most one user request at a time, and each user's travel request is served by only one vehicle. Formula 8 is expressed as follows: ； Formula 9 is expressed as follows: ； The decision variables are ensured to be binary variables by using constraints, which are expressed by Equation 10: .

4. The autonomous taxi fleet operation management decision-making method considering the spatiotemporal uncertainty of demand as described in claim 1, characterized in that, In S2, the vehicle regional dispatch task aims to minimize the total mileage of all dispatched vehicles to meet the travel requests of users in each sub-region in the future. This is expressed by Formula 11: ； in, For a moment From the region Deploy available vehicles to the area ; For a moment From the region Transferred to the region The number of available vehicles; for Vehicles should be moved from the area at all times. Transferred to the region Required mileage; After the regional allocation task is completed, the probability that the number of available vehicles in all regions is greater than or equal to the number of user travel requests is greater than or equal to... This can be expressed by Formula Twelve as follows: ； in, This can be expressed by Formula Thirteen as follows: ； in, For a moment area The potential number of available vehicles is expressed by Formula Fourteen: ； in, For a moment area Total number of potential available vehicles in the country; For a moment area The probability that a vehicle currently performing an order will become a usable vehicle in the next time period; For a moment area The number of vehicles performing order-taking tasks; For a moment area The probability that a vehicle currently performing a charging task will become a usable vehicle in the next time period; For a moment area The number of vehicles currently charging; for Timely allocation to the region The number of vehicles is expressed by Formula Fifteen as follows: ； Based on formulas 12 to 15, the constraints for vehicle area allocation are expressed by formula 16: ； Decision variable constraints and This is expressed by Formula 17 and Formula 18 respectively, with Formula 17 being: Formula 18 is expressed as follows: .

5. The autonomous taxi fleet operation management decision-making method considering the spatiotemporal uncertainty of demand as described in claim 1, characterized in that, In S2, the vehicle parking task specifically involves: after completing the decision-making process for vehicle order taking, vehicle charging, and vehicle area allocation, all remaining vehicles that have not received task allocation in this round of scheduling will automatically move to the nearest parking space to wait for the next round of task allocation.

Images