An automatic driving taxi operation strategy hierarchical cooperative optimization method

Abstract

The application relates to the technical field of vehicle dispatching operation, and discloses an automatic driving taxi operation strategy hierarchical cooperative optimization method, which comprises the following steps: acquiring traffic network data, vehicle fleet state data, charging facility data and user travel demand data of a target operation area; constructing a multi-scale hierarchical cooperative optimization strategy according to the acquired data; inputting an order distribution and vehicle path planning strategy, a vehicle cross-region dispatching and in-region cruising strategy and a dynamic pricing and charging and discharging threshold strategy into an automatic driving taxi operation simulation model to perform full-cycle simulation operation, and outputting comprehensive operation benefit indexes; taking the comprehensive operation benefit indexes as optimization targets, iteratively optimizing parameters of the dynamic pricing and charging and discharging threshold strategy by using a full-cycle optimization method, and updating the dynamic pricing and charging and discharging threshold strategy in the next long time interval by using the optimized parameters. The application can improve the adaptability and robustness of the operation strategy.

Description

Technical Field

[0001] This invention relates to the field of vehicle dispatching and operation technology, and specifically to a hierarchical collaborative optimization method for autonomous taxi operation strategies. Background Technology

[0002] The development of operational plans for autonomous taxis is still in its early stages. Autonomous electric vehicles offer high controllability, require no downtime, and are centrally managed by a dispatch center. Therefore, effective centralized control has a lower impact on the overall road network and power grid compared to existing shared vehicles (including ride-hailing and one-way rental car-sharing) and traditional taxis controlled by individual drivers. Good operational control can significantly improve operational efficiency. However, due to the overall scale, the requirements for centralized system control are higher, and the challenges are greater.

[0003] Existing research on autonomous taxi operation strategies often focuses on a single type of strategy, with insufficient consideration given to the coordinated implementation of multiple operational methods, including order matching, vehicle dispatching, charging and discharging strategies, and pricing schemes. This leads to difficulties in achieving overall coordination and unity in the actual operation process, as focusing on improving a single dispatching strategy can hinder this process. Furthermore, the operation of autonomous taxis impacts the road network and influences user travel choices. Therefore, coordinating various operational strategies while considering their impact on the road network and users, and issuing real-time operational commands to autonomous taxis, is crucial for improving the overall operational efficiency of autonomous taxis. Summary of the Invention

[0004] To address the aforementioned shortcomings in existing technologies, this invention provides a hierarchical collaborative optimization method for autonomous taxi operation strategies. This method constructs a multi-scale strategy encompassing short-term order allocation and route selection, mid-term cross-regional scheduling and intra-regional vehicle cruising and movement, and daily dynamic pricing with a charging / discharging strategy based on inventory theory. Based on this, a physical logic simulation model of autonomous taxis under an urban road network is established to verify, evaluate, and update the operation strategy. Specifically, short-term order allocation and mid-term cross-regional scheduling are calculated in real-time using linear programming models. Considering the impact of long-term operation strategies on the overall system, an optimization method integrating the optimal foreground region stochastic search convergence optimization method, synchronous perturbation stochastic approximation method, and coordinate search method is used to establish a simulation optimization framework, achieving system optimization for daily operation.

[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0006] A hierarchical collaborative optimization method for autonomous taxi operation strategies includes the following steps:

[0007] Acquire traffic network data, fleet status data, charging facility data, and user travel demand data for the target operating area;

[0008] Based on the acquired data, a multi-scale hierarchical collaborative optimization strategy is constructed. Order allocation and vehicle route planning strategies are executed within a predefined short time interval, cross-regional vehicle dispatching and intra-regional cruising strategies are executed within a predefined medium time interval, and dynamic pricing and charging / discharging threshold strategies are applied within a predefined long time interval.

[0009] The order allocation and vehicle route planning strategy, the cross-regional vehicle dispatching and intra-regional cruising strategy, and the dynamic pricing and charging / discharging threshold strategy are input into the autonomous driving taxi operation simulation model for full-cycle simulation operation, and the comprehensive operation efficiency index is output.

[0010] Using the comprehensive operational efficiency index as the optimization target, the parameters of the dynamic pricing and charge / discharge threshold strategy are iteratively optimized using a full-cycle optimization method, and the optimized parameters are used to update the dynamic pricing and charge / discharge threshold strategy for the next long time interval.

[0011] The present invention has the following beneficial effects:

[0012] This invention decouples and collaboratively optimizes short-, medium-, and long-term operational strategies through a hierarchical optimization framework, solving the overall efficiency bottleneck caused by single-strategy optimization. This allows the strategies to cooperate and form a global optimum. Furthermore, addressing the challenge of mixed decision variables in long-term strategy optimization, a full-cycle nested optimization framework is adopted to obtain a high-quality solution close to the global optimum within an acceptable timeframe. The entire optimization process is based on a high-fidelity simulation model and fully considers the real-world complexities such as nonlinear vehicle charging, dynamic road network congestion, random user demand, and electricity price fluctuations, making the optimized strategy more adaptable and robust in actual operation. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of a hierarchical collaborative optimization method for autonomous taxi operation strategy according to the present invention;

[0014] Figure 2 This is a schematic diagram of the autonomous taxi operation scenario of the present invention;

[0015] Figure 3 This is a schematic diagram illustrating the multi-stage rolling optimization logic for the operation of the autonomous driving shared car travel service of the present invention.

[0016] Figure 4 This is a schematic diagram of the simulation optimization framework for the long-term operation strategy of the present invention;

[0017] Figure 5 This is a simulation logic diagram of the autonomous taxi operation process of the present invention. Detailed Implementation

[0018] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0019] like Figures 1 to 4 As shown in the figure, an embodiment of the present invention provides a hierarchical collaborative optimization method for autonomous taxi operation strategies, comprising the following steps S1 to S4:

[0020] S1. Obtain traffic network data, fleet status data, charging facility data, and user travel demand data for the target operating area;

[0021] In an optional embodiment of the present invention, step S1 divides the space of the autonomous taxi in the city into different areas based on the road network and location of the city where the autonomous taxi operates. Z represents the set of regions, with a total of N regions. Each region is equipped with a dedicated charging and discharging node. When a vehicle is idle, it prioritizes parking at a node for maintenance and charging / discharging. If the dedicated parking spaces are saturated, the vehicle will park in nearby public parking spaces or cruise at low speed. The vehicle's operational tasks are directly dispatched by the autonomous taxi operation system, covering user order response, cross-regional vehicle dispatch, and charging and cruising within the region. To balance city-level computational efficiency with simulation evaluation longitude, the travel distance is simplified to the topological connectivity between key regional nodes, as shown in Figure 2. Furthermore, the vehicle's speed is state-dependent, characterized by a basic traffic flow graph considering both endogenous and exogenous congestion. That is, the fleet's operating status actively affects the road congestion level, which in turn affects the vehicle speed. Therefore, this embodiment acquires real-time traffic network data, fleet status data, charging facility data, and user travel demand data for the target operating area based on the operational scenario of the autonomous taxi service.

[0022] S2. Construct a multi-scale hierarchical collaborative optimization strategy based on the acquired data. Execute order allocation and vehicle route planning strategies within a predefined short time interval, execute cross-regional vehicle dispatching and intra-regional cruising strategies within a predefined medium time interval, and apply dynamic pricing and charging / discharging threshold strategies within a predefined long time interval.

[0023] In an optional embodiment of the present invention, step S2, executing the order allocation and vehicle routing strategy, includes:

[0024] An integer programming model for order allocation is constructed within each short time interval with the optimization objective of maximizing short-term operating profit and is solved in real time. The constraints of the integer programming model for order allocation include the constraint of maximizing user travel demand for order allocation and the constraint of the number of vehicles to be dispatched for order allocation.

[0025] Based on the order allocation results obtained from the solution, the shortest travel path from the origin to the destination of the vehicle is calculated using traffic network data.

[0026] Step S2, executing the vehicle cross-regional dispatch strategy, includes:

[0027] In each intermediate time interval, a multi-objective vehicle scheduling integer programming model is constructed with the primary optimization objective of maximizing user travel demand and the secondary optimization objective of minimizing vehicle travel time, and the model is solved in real time.

[0028] The constraints of the multi-objective vehicle scheduling integer programming model include constraints on the number of vehicles to be scheduled and constraints on meeting user travel demand for vehicles.

[0029] Step S2 involves implementing a patrol strategy within the designated area, including:

[0030] Based on the comparison between the predicted vehicle supply and the predicted user travel demand in the next intermediate time interval within the target operating area, different vehicle parking space transfer logics are executed.

[0031] When the predicted number of vehicles supplied is greater than or equal to the number of users' travel needs, execute the vehicle parking space transfer logic with the goal of reducing parking fees.

[0032] When the predicted number of vehicles is less than the number of users' travel needs, execute either a vehicle parking space transfer logic or a vehicle low-speed cruising logic aimed at increasing the number of available vehicles.

[0033] The dynamic pricing strategy applied in step S2 correlates user travel demand with an elastic demand function, where the elastic demand function represents that user travel demand decreases exponentially as the price increases within the pricing parameter range.

[0034] The charging and discharging threshold strategy applied in step S2 includes an upper charging threshold and a lower discharging threshold. When the vehicle's battery level is higher than the lower discharging threshold, the vehicle is controlled to discharge to the grid. When the vehicle's battery level is lower than the upper charging threshold, the vehicle is controlled to charge from the grid.

[0035] This embodiment adopts a centralized management and control model to coordinate fleet operations, and the operational behaviors are divided into the following four categories:

[0036] (1) Order allocation and route allocation: The vehicle executes the orders allocated by the system to meet the user's travel needs, and at the same time generates a route allocation plan based on the road network congestion status to complete the passenger pick-up and drop-off task across regions;

[0037] (2) Cruise movement within the area: The vehicle moves between the dedicated charging station and the public parking space or at low speed to optimize the utilization rate of the parking facilities;

[0038] (3) Cross-regional dispatch: Autonomous taxis travel empty between different regions to alleviate spatial supply and demand imbalance;

[0039] (4) Vehicle-to-grid interaction charging and discharging: Based on the preset battery threshold and the real-time charge status of the vehicle, the autonomous taxi carries out charging and discharging operations to achieve the optimal balance between service efficiency and charging and discharging revenue.

[0040] This embodiment constructs a hierarchical collaborative optimization framework for autonomous taxi operation services, such as... Figure 3 As shown, this framework optimizes operational strategies from three time dimensions based on daily configuration scheme optimization: short-term user order allocation, medium-term cross-regional vehicle dispatching and intra-regional movement and cruising strategies, and long-term dynamic pricing and vehicle charging / discharging threshold strategies. Specifically, for short-term operational strategies, user demand is generated every short time interval ∆, triggering the order allocation strategy; for medium-term operational strategies, cross-regional vehicle rescheduling and intra-regional relocation strategies are triggered every medium time interval t; dynamic pricing and charging / discharging thresholds are updated every long time interval T. Daily configuration schemes, long-term pricing strategies, and charging / discharging thresholds have a significant impact on overall operational performance, and this embodiment uses simulation optimization algorithms to optimize them. In contrast, short-term and medium-term operational strategies are constructed as integer programming models and embedded in the simulation environment, and solved in real time by calling a commercial solver. Let Γ be the set of operational time periods, where ∆, t, and T all belong to Γ.

[0041] During operation, supply and demand imbalances lead to limited available vehicles in some areas, failing to meet the needs of all users. Therefore, selecting high-quality user orders for service is crucial for improving operating profits. This embodiment constructs an integer programming model for order allocation within each short time interval, aiming to maximize short-term operating profits. This model optimizes the selection of orders within short time intervals. User demand order solution To improve operational efficiency, the integer programming model for order allocation is represented as follows:

[0042] ;

[0043] , , ;

[0044] , , ;

[0045] , ;

[0046] ;

[0047] in, To find the maximum value function, For each short period of time The number of orders allocated. For each short period of time The unit price of orders for vehicles departing from region i. For each short period of time Domestic user travel needs For a short period of time The number of available vehicles in inner region i. For a set of regions, For the collection of operating periods, Let be the set of positive integers. The first equation is the objective function for order allocation, representing maximizing profit by receiving the number of orders; the second equation is the constraint for maximizing user demand satisfaction in order allocation, representing the number of orders allocated. The first equation should be less than or equal to the user travel demand in the short time interval ∆; the second equation is the constraint on the number of vehicles available for order allocation, meaning that the number of vehicles available for allocation in a stage cannot exceed the number of vehicles available in the initial area; the third equation is the constraint on invalid scheduling, meaning that vehicles in the same area will not be allocated to demand; the fifth equation is the domain constraint on the decision variables.

[0048] After order allocation is determined, the Floyd-Warshall shortest path algorithm is used to calculate the shortest route for the user to take an autonomous taxi from the origin to the destination, based on the user's origin and destination and the road network congestion status. This completes the order allocation and route planning.

[0049] To address the imbalance between vehicle supply and user demand across different time periods and regions, this embodiment employs an optimized vehicle scheduling plan to alleviate this imbalance. First, based on arrival probabilities, the number of arriving users in the next scheduling period t is simulated to obtain the predicted number of users traveling to arrival region i. Then iterate through each station to obtain the initial number of available vehicles in region t for the next scheduling phase. At the same time, here we set the estimated number of available vehicles for the next stage as follows: , representing the number of vehicles available in the initial phase In addition, there are vehicles expected to meet charging requirements. The calculation method involves iterating through all unusable vehicles. The minimum usage standard for the next stage is calculated using the charging formula. This is then added to the list of available vehicles expected to reach region i in the next time period. Traverse all paths, find the vehicle whose destination is region i, estimate the vehicle's expected arrival time and expected battery level, and select the available vehicle that meets the time and battery level requirements. Here This is an estimated value, and the actual value may not be reached due to charging space limitations and sudden traffic congestion. Therefore, this embodiment constructs a multi-objective vehicle scheduling integer programming model within each intermediate time interval, with maximizing user travel demand as the first optimization objective and minimizing vehicle travel time as the second optimization objective, expressed as:

[0050] ;

[0051] , , ;

[0052] , , ;

[0053] ;

[0054] ;

[0055] , , ;

[0056] , ;

[0057] in, Let be the number of users served within the intermediate time interval t. To find the minimum value function, Let be the travel time from region i to region j within the intermediate time interval t. Let represent the number of vehicles dispatched from region i to region j within the intermediate time interval t. Let be the number of vehicles dispatched from region j to region j within the intermediate time interval t. The estimated number of available vehicles for the next intermediate time interval t. Let represent the number of vehicles scheduled from region j to region i within the intermediate time interval t. The first equation represents the primary priority objective of maximizing service user demand; the second equation represents the secondary objective of minimizing vehicle travel time; and the third equation represents the maximum constraint on maximizing user demand satisfaction, indicating the next intermediate time interval... The number of services should be less than or equal to the total user travel demand; the fourth equation is the vehicle dispatch quantity constraint, meaning the number of dispatchable vehicles cannot exceed the initial number of available vehicles in the region during the intermediate time interval t. The fifth equation represents an invalid scheduling constraint, meaning that scheduling will not occur at the same station. The sixth equation represents a user demand satisfaction vehicle restriction constraint, indicating that the number of user demands satisfied in region i in the next stage will not exceed the estimated number of vehicles available in region i during the intermediate time interval t. Add the vehicles that arrive by dispatch and subtract the vehicles that leave by dispatch. Since the dispatched vehicles may not be able to arrive within the intermediate time interval t or may not have enough power, we take the less than or equal to sign; Equation 7 is the domain constraint of the decision variable.

[0058] The vehicles within the area must address the following two core trade-offs: (1) The proactive relocation logic of charging resources: When a vehicle in a dedicated parking space is idle, and it is expected that a vehicle urgently needing charging will arrive through scheduling or order in the next phase, the system needs to decide whether the idle vehicle should proactively give up the parking space to ensure that vehicles with insufficient power are given priority for charging, thereby improving the overall availability of the fleet. (2) The logic of parking space switching and cost optimization: When a dedicated parking space is full, it is assessed whether to move a fully charged vehicle to a public parking space and replace it with a vehicle waiting to be charged in the public space. This measure can ensure that vehicles with low power can be charged in a timely manner, and can significantly reduce unnecessary public parking fees.

[0059] Since the dynamic parking space transfer strategy is solved specifically for each area, if an integer programming model is established for solving it, then N integer programming models for the number of areas would need to be solved every t time interval, which would greatly reduce simulation efficiency. Therefore, this embodiment introduces a transfer strategy to determine whether to perform vehicle parking space transfer. Let... ,in Let T be the number of available vehicles generated due to vehicle area movement in the next phase. The number of vehicles that can be used during the intermediate time interval t. The number of vehicles charging in designated parking spaces.

[0060] (1) When At this point, if there are enough vehicles in the area for the next phase of use, the primary operational objective for that area is to reduce parking costs. If , Given the number of available charging stations in area i, and indicating the availability of dedicated parking spaces, then based on parking time in public parking spaces from low to high, move within the area. Vehicles move from public parking spaces into designated parking spaces. If If the designated parking spaces are saturated, then the number of vehicles parked in the public spaces will be ranked from lowest to highest based on the amount of time spent in the public spaces and the number of vehicles available in the designated parking spaces. To exchange.

[0061] (2) When At that time, if there are not enough vehicles in the area for the next phase, the primary operational objective for that area is to increase the number of available vehicles. and This indicates that there are enough unused dedicated parking spaces in the area. Therefore, vehicles in public parking spaces will be moved within the area according to their battery level, from lowest to highest. If there are enough low-priced public parking spaces in the area, vehicles will move from designated parking spaces to public parking spaces; otherwise, if no low-priced public parking spaces are available, the vehicle will enter a low-speed cruising mode and cruise within the area. and and This indicates that there are not enough available parking spaces in the area, but vehicles can be moved by exchanging them with available ones. If If the designated parking spaces are saturated, then the number of vehicles parked in the public spaces will be ranked from lowest to highest based on the amount of time spent in the public spaces and the number of vehicles available in the designated parking spaces. To exchange.

[0062] For dynamic pricing strategies, this embodiment assumes an elastic demand function. With potential demand and price Related, expressed as:

[0063]

[0064] in, This represents the travel demand of users from region i to region j within a long time interval T. This represents the potential user travel demand from region i to region j within a long time interval T. This is the price demand coefficient. This represents the unit price of the order for vehicles originating from region i.

[0065] Since dynamic pricing directly affects travel demand between regions, pricing schemes need to be updated in real time based on potential demand at different times, in order to balance the actual demand in each region and maximize the system's operational efficiency.

[0066] Furthermore, regarding the charging and discharging strategy, the following is set: The upper threshold for charging region i during time period T. The lower threshold for charging is the amount of electricity in vehicle h in region i. Greater than Discharge occurs when Less than Charge it when needed.

[0067] S3. Input the order allocation and vehicle route planning strategy, the vehicle cross-regional dispatch and intra-regional cruising strategy, and the dynamic pricing and charging / discharging threshold strategy into the autonomous driving taxi operation simulation model for full-cycle simulation operation, and output comprehensive operation efficiency indicators.

[0068] In an optional embodiment of the present invention, the autonomous taxi operation simulation model constructed in step S3 includes:

[0069] The vehicle module is used to define the static physical properties and dynamic state of the vehicle, and to simulate the changes in its state of charge as it is driven, charged, and discharged. The charging process simulates the nonlinear characteristics of the constant current-constant voltage two-stage process.

[0070] The path module is used to define the static attributes and dynamic traffic conditions of a path, and uses an exponential speed-density relationship model that takes into account background congestion and the occupancy of autonomous vehicles to dynamically calculate the travel speed of road segments.

[0071] The region module is used to define the static resources and dynamic vehicle sets of a region, and to simulate the event flow of vehicle arrival, order matching, dispatch matching, departure, patrol within the region, and charging / discharging status updates;

[0072] The node module is used to simulate the process of a vehicle passing through various nodes in the path;

[0073] The operations module is used to simulate the state changes of vehicle operation strategies.

[0074] This embodiment describes the simulation logic through three core elements: static attributes, dynamic attributes, and events, and divides them into five functional modules: (1) Vehicle module: defines the basic physical and technical attributes of the vehicle; (2) Area module: describes the vehicle's passenger pick-up, displacement, cruising, and charging / discharging processes within the area; (3) Path module: simulates the dynamic driving process considering road congestion and vehicle interaction; (4) Node module: describes the process of the vehicle passing through various nodes in the path; (5) Operation module: mainly represents the state changes of the autonomous taxi's operation strategy. Among them, the vehicle module and the operation module are not physical modules, but are embedded in the area node and path modules, respectively used to simulate the state changes of the vehicle during operation and the changes in the system's operation strategy. The overall logical architecture of the simulation model is shown in Figure 5.

[0075] Let h be the serial number of the shared autonomous vehicle, and H be the set of vehicles. The vehicle module simulates changes in vehicle battery level, speed, and location, as well as state changes in vehicle operation commands during operation. This module does not contain specific events; events are invoked from other modules for concrete manifestation. Vehicle static attributes include fixed input attributes related to the autonomous shared vehicle, such as battery capacity and minimum driving charge, as well as coefficients related to vehicle driving power, expressed as:

[0076] ;

[0077] ;

[0078] ;

[0079] ;

[0080] ;

[0081] in, Let h be the set of static parameters for vehicle h; This is the battery capacity of vehicle h, which is also the maximum battery capacity of the vehicle; This represents the minimum amount of electricity required for vehicle use, calculated as 20% of the total electricity consumption. For the quality of the vehicle; Let h be the frontal area of ​​the vehicle; For auxiliary power; For transmission efficiency; air density; This is the rolling resistance coefficient; It is the acceleration due to gravity; This is the drag coefficient. This is the set of dynamic parameters for vehicle h; For vehicle h in time Battery life at that time; Power consumption per unit time; For vehicle h in time The speed at that time; The timestamp for the change in the state of vehicle h; A set of vehicle movement processes determined by tasks such as scheduling and order matching; For the vehicle's charging and discharging status at the station, set For discharge, For charging, Vehicles are left idle; For the set of vehicle on-site status, if =0, the vehicle is in a charging / discharging parking space. =1, the vehicle is in a public parking space; For vehicle h at time Remaining mileage on the current path; The estimated departure time is the current estimated departure time. Due to real-time changes in road conditions, the estimated arrival time of vehicles from the departure route is dynamically changing. The number of users serving vehicle h; This represents the revenue generated by vehicle h upon completing an order; This indicates the benefit generated by vehicle h's discharge; This indicates the cost incurred when vehicle h is temporarily parked in a public parking space; This indicates that vehicle h incurred charges due to charging. For traction power, This refers to the road slope.

[0082] When a vehicle is charging, the change in its battery level exhibits nonlinear characteristics. This nonlinearity is primarily reflected in the dynamic changes in charging current or power with time and battery state (such as State of Charge (SOC) and temperature). It mainly presents an initial constant-current-constant-voltage (CC-CV) state. In the initial charging phase, the constant-current (CC) stage occurs, with a constant current and battery voltage increasing approximately linearly with increasing SOC (affected by internal resistance and polarization effects). During this phase, the charging power P = I × VP = I × V gradually increases, and the charging speed (energy input rate) exhibits a "quasi-linear" behavior. However, when the voltage reaches its upper limit and switches to constant-voltage (CV) mode, the current decays exponentially, and the charging power decreases nonlinearly, as shown below:

[0083] ;

[0084] = +( - ) / ;

[0085] in, Let h be the battery state of vehicle h during the intermediate time interval t. For charging speed, The threshold charge for switching from constant voltage to constant current. This indicates the time when vehicle h begins charging at station c. This indicates the time from the start of charging to the end of the constant current period. This is the charging function during the constant voltage phase. The charging start time.

[0086] Let l represent the symbol of a path, and L be the set of connecting paths between different areas of the road network. The path module is used to simulate the process of vehicle scheduling and travel tasks, including the changes in the status of each path and vehicle as it travels from the starting area to the destination area.

[0087] Vehicle static attributes include fixed input attributes related to the path, and path static attributes. Represented as:

[0088] ;

[0089] in, , , These represent the length of path k, the number of lanes, and the congestion density, respectively. Let k be the capacity of path k. . Let K be the free-flow velocity. Vehicle road occupancy rate, i.e., exogenous congestion not caused by SAVs; and Let k be the starting and ending regions of path k.

[0090] Path dynamic attributes The attributes that include path changes due to time and the arrival of autonomous vehicles are represented as:

[0091] ;

[0092] in, Let k represent the set of vehicles on path k. For vehicles on path k at time step The average speed over time.

[0093] Considering the impact of the number of vehicles on the route and road congestion on speed, Described as an exponential function model, it is expressed as:

[0094]

[0095] ;

[0096] ;

[0097] in, and For the given coefficients, Let K be the free-flow velocity along path k. and The road conditions are respectively and The average speed of vehicles on the road, road conditions, and road length. and number of lanes And it is related to vehicle congestion rate, and This represents the remaining capacity in the background of the path. Unlike the linear speed model, when a road segment is completely blocked, the value of the exponential function model does not drop to 0, but remains at the blocking speed, indicating that vehicles are still moving slowly, which more realistically reflects the actual operating conditions.

[0098] The vehicle arrival path process is as follows: when the vehicle arrives from the area... Upon entering path k, a "vehicle arrived" event is triggered, updating the vehicle set for that path. and vehicle status change timestamps And trigger the "Route and Vehicle Status Update Event" to Given the vehicle speed in the previous update phase, and obtain the vehicle's average power in the previous update phase. This updates the vehicle's battery level. And update the speed of vehicles along the route based on the route status. and all vehicles in the route Remaining mileage updated Estimated departure time of the vehicle and made an appointment at a certain time. Trigger the vehicle departure process.

[0099] The vehicle departure process determines whether a vehicle has left the route; if so, the time... When a vehicle triggers a "vehicle departure event", the set of vehicles on the path is updated. This simultaneously triggers the "Route Vehicle Status Update Event" and the "Vehicle Arrival Event" of the region module.

[0100] Let i be the symbol of a region, and Z be the set of regions. Each region module has a dedicated charging station. Assume that users pick up and use vehicles within the region. Nodes represent the connection points between paths. There are no dedicated parking spaces or vehicle retrieval processes. Vehicles only arrive at and leave at nodes.

[0101] static attributes of the region Represented as:

[0102] ;

[0103] in, For charging rate, For discharge rate, For users in region i, the average expected time from booking to pickup for vehicles in their local area. Based on the actual customer demand in region i at different times, The number of charging stations to be replaced in the region. Waiting time when reserving a vehicle for a user; Set upper thresholds for battery charging and discharging. The threshold values ​​for battery charging and discharging. For time Electricity price at that time Parking fees for public parking spaces.

[0104] Regional dynamic attributes Represented as:

[0105] ;

[0106] in, In order to be in Vehicles gather within the time zone. The number of vehicles that meet the minimum power requirements and have not been assigned a task; In order to be in The number of vehicles parked in designated parking spaces at any given time; In time period T, a user makes a vehicle reservation from region i to any region. Total travel demand; Let this be the probability per unit time that a user picks up a car in region i and travels to region j. ; For a short period of time User needs awaiting allocation = ; For a short period of time The order matching requirements are from region i to region j. Price the vehicles for time period T from i to j.

[0107] The vehicle arrival process is as follows: when a vehicle arrives at an area from the path module, a vehicle arrival event is triggered, and the vehicle set within the area is updated. Vehicle status change timestamps If the vehicle has no mission, i.e., the vehicle route... And meet the minimum power requirements. Update available vehicles If there are available dedicated V2G parking spaces. The vehicle enters an empty dedicated V2G parking space and updates. , If the vehicle's battery level is low The vehicle is discharging electricity. ;like Vehicle charging ,or and Vehicles are idle. Otherwise, when V2G professional parking spaces are not available, Enter a public parking space. , If vehicle h has a task, i.e. Update vehicle route collection This triggers the vehicle's departure process.

[0108] The user order allocation process is as follows: every... When the "Vehicle Status Update Event" is triggered, the vehicles in the area are updated first, and the time markers of all vehicles are updated. If the vehicle is in a V2G-only parking space, when Update battery level Update cost If the battery level is low Update the number of available vehicles ;if ,renew Update earnings .

[0109] Trigger "demand trigger and order allocation events" based on user arrival rate per unit time in each region. Based on arrival probability Randomly generated Actual travel demand during the time period If vehicles are available in region i Assign tasks according to orders Vehicles are matched with users' travel requests to area j, based on their battery level in descending order. This triggers a "vehicle departure process," updating the number of available vehicles in the area. Number of people served Vehicle expected revenue update , Let be the path length from region i to region j.

[0110] If all vehicles within region i have been matched and there are still unassigned users, then check the set of paths from other regions to this region. , If the vehicle h has a remaining distance And battery ,in To estimate the power consumption of a vehicle after it arrives in region i, the vehicle is added to the set of available vehicles in that region. In the middle, then press Match vehicles (h) with users in ascending order of size, and update the number of users served. Vehicle expected revenue update Using the Floyd shortest path algorithm, we calculate the number of nodes traversed from region i to region j, thus obtaining the set of intermediate nodes. Update vehicle mission route set Next, make an appointment. The "vehicle departure" process is triggered at any time, indicating that the vehicle has received the order and is heading to the user's location.

[0111] The vehicle dispatching process is as follows: when a vehicle in the area receives a dispatching task... Then, the "Schedule Vehicle Matching Task" is triggered. If vehicles are available in region i... The scheduling tasks are matched according to the travel distance from highest to lowest, and the vehicles are matched according to the battery level from highest to lowest. After matching, the vehicle departure process is triggered, and the regional vehicle set is updated. If all vehicles in the same region have been matched and there are still unassigned scheduling tasks, the set of paths from other regions to this region is checked. , Among the feasible vehicles, and according to the remaining distance of the vehicles. In ascending order, match vehicles h with scheduling tasks. For each vehicle matched with a scheduling task, calculate the number of nodes traversed from region i to region j to obtain the set of intermediate nodes. Update vehicle mission route set .

[0112] The vehicle departure process is as follows: When a vehicle receives a vehicle order or dispatch task, a "vehicle departure event" is triggered. This involves iterating through all vehicles in region i, and if vehicle h in region h has a movement task, then... ,Pick The first element j updates the vehicle set in region i. ;if ,renew And update the vehicle's battery level, if If the power consumption remains unchanged, ,renew Update charging fees ;if ,renew Update discharge revenue Last updated vehicle status change timestamp This triggers the "Vehicle Arrival at Road Process" module, allowing the vehicle to enter the path.

[0113] Vehicle movement within the area is as follows: When a vehicle within the area receives a task to move to a parking space within the area, a "Dynamic Vehicle Movement Event within the Area" is triggered. If the vehicle moves from a designated parking space to a public parking space, the number of vehicles in the designated parking space is updated. Parking status Vehicle charging / discharging status If a vehicle moves from a public parking space to a designated parking space, update the parking status. ,if Vehicle discharge ,if The vehicle has entered charging mode. ,if Vehicles are idle .

[0114] The vehicle passage process is as follows: The vehicle passage process only occurs in the node module. When vehicle h arrives at node k, it is based on the vehicle path set. Determine the next node or region the vehicle will reach, then leave the node and head to the designated region or node, and update... .

[0115] S4. Using the comprehensive operational efficiency index as the optimization target, the parameters of the dynamic pricing and charging / discharging threshold strategy are iteratively optimized using a full-cycle optimization method, and the optimized parameters are used to update the dynamic pricing and charging / discharging threshold strategy for the next long time interval.

[0116] In an optional embodiment of the present invention, the full-cycle optimization method in step S4 adopts a nested optimization framework. In the outer loop, a convergent optimization method based on the optimal foreground region random search and a coordinate search method are used to search in the decision space containing discrete variables to generate candidate discrete solutions.

[0117] For each candidate discrete solution, an inner loop is started, the candidate discrete solution is fixed as a discrete variable, and the synchronous perturbation stochastic approximation method is used to optimize the corresponding continuous dynamic pricing and charging / discharging threshold strategy parameters to obtain the optimal continuous parameters and corresponding simulation profit under the discrete solution.

[0118] The optimal continuous parameters and simulation profits obtained from the inner loop are fed back to the outer loop to guide the next round of search.

[0119] This embodiment addresses dynamic pricing and charge / discharge threshold strategies with time-segment coupling characteristics within the operating cycle, employing a full-day, all-time joint optimization approach rather than simple segmented optimization. The simulation optimization framework is as follows: Figure 4 As shown, by updating the optimized pricing for each time period and the threshold input simulation model in different time periods, the global optimization of the long-term operation strategy is finally achieved. This framework fully leverages the efficiency of the optimal foreground region stochastic search convergence optimization method and the coordinate search method in finding the optimal discrete structure in high-dimensional discrete spaces (such as node capacity and fleet size), and the ability of the synchronous perturbation stochastic approximation method to quickly find the optimal continuous parameters in large-scale continuous spaces (such as dynamic pricing and charging / discharging thresholds). The framework consists of two loops: an outer loop manages the search for discrete variables, calls the inner loop for evaluation, and ensures overall convergence; the inner loop aims to quickly find and evaluate the optimal combination of continuous parameters for the given fixed discrete structure in the outer loop, thereby accurately obtaining the true profit potential of the structure.

[0120] In this embodiment, the outer loop performs a search in a decision space containing discrete variables based on the optimal foreground region random search convergence optimization method and the coordinate search method to generate candidate discrete solutions, including:

[0121] Set the maximum number of iterations for the outer loop. Standard sampling dimensions Number of internal and external loop conversions and the number of samples per round Initialize the visited solution set Profit Sample Evaluation Number Evaluation count counter Maximum number of simulations in a single iteration Current optimal decision vector and initial profit assessment ;in Let k be the initial decision vector, and k be the iteration number.

[0122] In the current iteration k, from the visited solution set Selecting the profit function The largest decision vector is taken as the current optimal decision vector. Among them are Discrete candidate solutions;

[0123] With the current optimal decision vector Centered on, combined with the current discrete solution With visited solution set Distances to other sample points in the range to construct the constraint range Defines the neighborhood boundary of the current search; where As the most promising region, To solve the space, To solve the currently visited set The reference point taken from it. This is the current optimal solution;

[0124] For discrete candidate solutions of Coordinate sampling is performed on each of the coordinate dimensions to generate a new discrete candidate solution set. ;in The coordinate dimension index. This refers to the number of coordinate dimensions.

[0125] Traverse the new discrete candidate solution set It will not be in the visited solution set. Candidate solutions join in And reset the corresponding profit sample evaluation number. and evaluation count counter Whenever the number of iterations reaches an integer multiple of the number of transitions between the inner and outer loops, the inner loop is executed to optimize the continuous variables of the candidate solution, and the profit of the optimal sample is recorded. ;

[0126] Determine if the number of iterations k in the outer loop has reached the maximum number of iterations. If the condition is not met, increment the outer loop iteration count k by 1 and redetermine the current optimal solution until the termination condition is met, then output the optimal decision variable. and corresponding profits .

[0127] This embodiment starts an inner loop for each candidate discrete solution, fixing the candidate discrete solution as a discrete variable, and uses a synchronous perturbation stochastic approximation method to optimize its corresponding continuous dynamic pricing and charge / discharge threshold strategy parameters, obtaining the optimal continuous parameters and corresponding simulation profit under that discrete solution, including:

[0128] Obtain the current discrete candidate solution passed from the outer loop. Set the maximum number of iterations for the inner loop. Parameters of the synchronous perturbation random approximation algorithm Minimum number of single-iteration simulations and high-precision simulation times ;

[0129] In each inner loop iteration k, the step size is calculated based on the parameters of the synchronization perturbation stochastic approximation method. and disturbance coefficient Generate symmetric perturbation vectors And for the current solution Apply positive and negative perturbations to construct upper and lower perturbation vectors respectively. and Profit estimates are obtained through a single simulation evaluation.

[0130] Calculate the approximate gradient of the objective function based on the profit estimate. ,in The sample number. Given the number of samples, update the continuous solution variables in the opposite direction of the gradient, and project the updated solution onto the feasible region. ;

[0131] Calculate the average profit of the current iteration. If the average profit is greater than the best profit recorded so far... Then update the current optimal solution. ;

[0132] Check if the inner loop has reached the maximum number of iterations; if so, terminate the loop and check the current optimal solution. implement Sub-high precision simulation, outputting the optimal continuous solution and the corresponding optimal profit of system operation .

[0133] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0134] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0135] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0136] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

[0137] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A hierarchical collaborative optimization method for autonomous taxi operation strategies, characterized in that, Includes the following steps: Acquire traffic network data, fleet status data, charging facility data, and user travel demand data for the target operating area; Based on the acquired data, a multi-scale hierarchical collaborative optimization strategy is constructed. Order allocation and vehicle route planning strategies are executed within a predefined short time interval, cross-regional vehicle dispatching and intra-regional cruising strategies are executed within a predefined medium time interval, and dynamic pricing and charging / discharging threshold strategies are applied within a predefined long time interval. The order allocation and vehicle route planning strategy, the cross-regional vehicle dispatching and intra-regional cruising strategy, and the dynamic pricing and charging / discharging threshold strategy are input into the autonomous driving taxi operation simulation model for full-cycle simulation operation, and the comprehensive operation efficiency index is output. Using the comprehensive operational efficiency index as the optimization target, the parameters of the dynamic pricing and charge / discharge threshold strategy are iteratively optimized using a full-cycle optimization method, and the optimized parameters are used to update the dynamic pricing and charge / discharge threshold strategy for the next long time interval.

2. The hierarchical collaborative optimization method for autonomous taxi operation strategy according to claim 1, characterized in that, The execution of order allocation and vehicle routing strategies includes: An integer programming model for order allocation is constructed within each short time interval with the optimization objective of maximizing short-term operating profit and is solved in real time. The constraints of the integer programming model for order allocation include the constraint of maximizing user travel demand for order allocation and the constraint of the number of vehicles to be dispatched for order allocation. Based on the order allocation results obtained from the solution, the shortest travel path from the origin to the destination of the vehicle is calculated using traffic network data.

3. The hierarchical collaborative optimization method for autonomous taxi operation strategy according to claim 1, characterized in that, Implementing cross-regional vehicle dispatching strategies includes: In each intermediate time interval, a multi-objective vehicle scheduling integer programming model is constructed with the primary optimization objective of maximizing user travel demand and the secondary optimization objective of minimizing vehicle travel time, and the model is solved in real time. The constraints of the multi-objective vehicle scheduling integer programming model include constraints on the number of vehicles to be scheduled and constraints on whether user travel demand meets vehicle restrictions.

4. The hierarchical collaborative optimization method for autonomous taxi operation strategy according to claim 1, characterized in that, The patrol strategy within the area includes: Based on the comparison between the predicted vehicle supply and the predicted user travel demand in the next intermediate time interval within the target operating area, different vehicle parking space transfer logics are executed. When the predicted number of vehicles supplied is greater than or equal to the number of users' travel needs, execute the vehicle parking space transfer logic with the goal of reducing parking fees. When the predicted number of vehicles is less than the number of users' travel needs, execute either a vehicle parking space transfer logic or a vehicle low-speed cruising logic aimed at increasing the number of available vehicles.

5. The hierarchical collaborative optimization method for autonomous taxi operation strategy according to claim 1, characterized in that, Dynamic pricing strategies link user travel demand through elastic demand functions, where elastic demand functions represent that user travel demand decreases exponentially as prices increase within the range of pricing parameters. The charging and discharging threshold strategy includes an upper charging threshold and a lower discharging threshold. When the vehicle's battery level is higher than the lower discharging threshold, the vehicle is controlled to discharge to the grid. When the vehicle's battery level is lower than the upper charging threshold, the vehicle is controlled to charge from the grid.

6. The hierarchical collaborative optimization method for autonomous taxi operation strategy according to claim 1, characterized in that, The simulation model for autonomous taxi operation includes: The vehicle module is used to define the static physical properties and dynamic state of the vehicle, and to simulate the changes in its state of charge as it is driven, charged, and discharged. The charging process simulates the nonlinear characteristics of the constant current-constant voltage two-stage process. The path module is used to define the static attributes and dynamic traffic conditions of a path, and uses an exponential speed-density relationship model that takes into account background congestion and the occupancy of autonomous vehicles to dynamically calculate the travel speed of road segments. The region module is used to define the static resources and dynamic vehicle sets of a region, and to simulate the event flow of vehicle arrival, order matching, dispatch matching, departure, patrolling within the region, and charging / discharging status updates; The node module is used to simulate the process of a vehicle passing through various nodes in the path. The operations module is used to simulate the state changes of vehicle operation strategies.

7. The hierarchical collaborative optimization method for autonomous taxi operation strategy according to claim 6, characterized in that, The road segment speed is dynamically calculated using an exponential speed-density relationship model that considers background congestion and the occupancy of autonomous vehicles. in, For vehicles on road segment k, the time step The average speed at that time Let K be the free-flow velocity of vehicles on road segment k. It is an exponential function. Time step on segment k The vehicles gathered at that time Let k be the capacity of road segment k. Time step on segment k Vehicle road occupancy rate at that time The velocity attenuation coefficient, The congestion index is the speed from region i to region j on road segment k. This represents the congestion coefficient.

8. The hierarchical collaborative optimization method for autonomous taxi operation strategy according to claim 1, characterized in that, The full-cycle optimization method adopts a nested optimization framework. In the outer loop, a convergent optimization method based on the optimal foreground region and a coordinate search method are used to search in the decision space containing discrete variables to generate candidate discrete solutions. For each candidate discrete solution, an inner loop is started, the candidate discrete solution is fixed as a discrete variable, and the synchronous perturbation stochastic approximation method is used to optimize the corresponding continuous dynamic pricing and charging / discharging threshold strategy parameters to obtain the optimal continuous parameters and corresponding simulation profit under the discrete solution. The optimal continuous parameters and simulation profits obtained from the inner loop are fed back to the outer loop to guide the next round of search.

9. The hierarchical collaborative optimization method for autonomous taxi operation strategy according to claim 8, characterized in that, In the outer loop, a convergent optimization method based on the optimal foreground region random search and a coordinate search method are used to search in the decision space containing discrete variables, generating candidate discrete solutions, including: Set the maximum number of iterations in the outer loop, the standard sampling dimension, the number of inner and outer loop transitions, and the number of samples per round. Initialize the visited solution set, the number of profit sample evaluations, the evaluation count counter, the maximum number of simulations per single iteration, the current optimal decision vector, and the initial evaluated profit. In the current iteration, the solution that maximizes the profit function is selected from the visited solution set as the current optimal solution; Centered on the current optimal solution, and combining the distance between the current solution and other sample points in the visited solution set, a constraint range is constructed, and the neighborhood boundary of the current search is defined. Sub-coordinate sampling is performed on each coordinate dimension of the discrete decision variable to generate a new discrete candidate solution set; Traverse the new discrete candidate solution set, add candidate solutions that are not in the visited solution set and reset the corresponding evaluation count counter; whenever the number of iterations reaches an integer multiple of the number of inner and outer loop conversions, execute the inner loop to optimize the continuous variables of the candidate solution and record the profit of the best sample. Determine if the outer loop has reached the maximum number of iterations; if so, terminate the loop and output the optimal decision variable and the corresponding profit.

10. The hierarchical collaborative optimization method for autonomous taxi operation strategy according to claim 9, characterized in that, For each candidate discrete solution, an inner loop is initiated, fixing the candidate discrete solution as a discrete variable. A synchronous perturbation stochastic approximation method is used to optimize its corresponding continuous dynamic pricing and charging / discharging threshold strategy parameters, obtaining the optimal continuous parameters and corresponding simulation profit under that discrete solution, including: Obtain the current discrete candidate solution passed by the outer loop, and set the maximum number of iterations of the inner loop, the parameters of the synchronous perturbation random approximation algorithm, the minimum number of simulations per iteration, and the number of high-precision simulations; In each inner loop iteration, the step size and perturbation coefficient are calculated based on the parameters of the synchronous perturbation stochastic approximation method; a symmetric perturbation vector is generated and positive and negative perturbations are applied to the current solution to construct upper and lower perturbation vectors respectively; the corresponding profit estimate is obtained through a single simulation evaluation. Calculate the approximate gradient of the objective function based on the profit estimate, update the continuous solution variables in the opposite direction of the gradient, and project the updated solution into the feasible region. Calculate the average profit of the current iteration. If the average profit is greater than the currently recorded optimal profit, then update the current optimal solution. Determine if the number of iterations in the inner loop has reached the maximum number of iterations; if so, terminate the loop, perform a high-precision simulation on the current optimal solution, and output the optimal continuous solution and its corresponding optimal system operating profit.

Images