EVTOL cluster collaborative task and charging planning method based on space-time uncertainty

Abstract

The invention discloses an eVTOL cluster collaborative task and charging planning method based on space-time uncertainty, and the method comprises the following steps: S1, building an eVTOL dynamic model based on the flight dynamic characteristics of a four-rotor eVTOL, and building an eVTOL energy consumption and charging model based on the motor system efficiency and the battery discharge characteristics; s2, collaborative decision making and dynamic adjustment are carried out through reinforcement learning algorithms based on double-layer DQN, navigation point coordinates are output, and the reinforcement learning algorithms based on the double-layer DQN comprise an outer-layer DQN agent decision algorithm and an inner-layer DQN agent decision algorithm; s3, inputting the navigation point coordinates into an A * algorithm, and generating an optimal reference path by the A * algorithm according to a multi-dimensional cost function; s4, dynamically adjusting the course angle and the flight speed of the eVTOL according to the optimal reference path by utilizing a PID (Proportion Integration Differentiation) controller until the eVTOL reaches the target node; through multi-agent collaborative decision and online optimization, the space-time uncertainty in the urban airspace is effectively dealt with, and the maximization of the eVTOL cluster operation efficiency and economic benefits is realized.

Description

Technical Field

[0001] This invention relates to the field of collaborative control technology for unmanned aerial vehicles (UAVs), and in particular to a method for collaborative task and charging planning of eVTOL clusters based on spatiotemporal uncertainty. Background Technology

[0002] eVTOL clusters refer to collaborative systems composed of multiple electric vertical takeoff and landing (eVTOL) aircraft. With the rapid growth of urban air traffic demand, eVTOL clusters have become a key technology for solving urban traffic congestion due to their flexibility. In the airspace, multiple eVTOL aircraft work together to complete complex tasks through collaborative control, task allocation, and path planning. However, in dynamic urban airspace, eVTOL clusters must simultaneously address multiple spatiotemporal uncertainties, such as travel order allocation, complex traffic flows, and sudden weather changes. This places stringent demands on the cluster's collaborative task planning and energy management.

[0003] Existing eVTOL clusters mostly adopt centralized scheduling strategies, using deterministic assumptions (such as fixed battery capacity and fixed charging time) for task allocation and charging planning. However, in actual operation, uncertainties exist, such as battery capacity limitations, uneven distribution of charging facilities, real-time traffic changes, and battery decay. This disconnects eVTOL cluster task planning from charging scheduling, making it difficult to adjust according to real-time conditions. This can easily lead to task interruptions (such as insufficient battery power to complete tasks) or resource idleness (such as idle charging facilities). As the eVTOL cluster scales up, processing efficiency decreases, making it difficult to meet real-time decision-making requirements.

[0004] Therefore, there is an urgent need for a distributed planning method that can adapt to spatiotemporal uncertainties and achieve efficient collaborative operation of eVTOL clusters. Summary of the Invention

[0005] In view of the above-mentioned prior art, the present invention provides a method for eVTOL cluster collaborative task and charging planning based on spatiotemporal uncertainty, which mainly solves the technical problems existing in the background art.

[0006] To achieve the above objectives, the technical solution of this invention is implemented as follows: A method for eVTOL cluster collaborative task and charging planning based on spatiotemporal uncertainty includes the following steps: Step S1: Establish an eVTOL dynamic model based on the flight dynamics characteristics of the quadcopter eVTOL, and establish an eVTOL energy consumption and charging model based on the efficiency of the motor system and the discharge characteristics of the battery. Step S2: Use a reinforcement learning algorithm based on two-layer DQN to perform collaborative decision-making and dynamic adjustment, and output navigation point coordinates. The two-layer DQN reinforcement learning algorithm includes an outer layer DQN agent decision-making algorithm and an inner layer DQN agent decision-making algorithm. Step S3: Input the navigation point coordinates into the A* algorithm. The A* algorithm generates the optimal reference path based on the multi-dimensional cost function. Step S4: Use the PID controller to dynamically adjust the heading angle and flight speed of the eVTOL according to the optimal reference path until the eVTOL reaches the target node.

[0007] Optionally, the eVTOL dynamic model in step S1 is:

[0008] in, , , These are the roll angle, pitch angle, and yaw angle along the eVTOL body coordinate system, respectively. , , They are , , angular velocity; , , They are , , angular acceleration; , , It is the relative velocity along the X, Y, and Z axes in a coordinate system with Earth as the reference point; , , It refers to the acceleration along the X, Y, and Z axes in a coordinate system with Earth as the reference point; , , It is the moment of inertia of the eVTOL body about the body coordinate system; , , These are the air damping coefficients of the drone along the X, Y, and Z axes, respectively. It is the moment of inertia of a single rotor about its own axis of rotation; It is the overall effective angular velocity of the four rotors of the eVTOL; , , , These are the actual control inputs for the four rotors of the eVTOL. Controlling the lifting and lowering movement of the eVTOL Controlling the roll motion of the eVTOL Controlling the pitch motion of the eVTOL Controlling the yaw motion of the eVTOL; The air drag coefficient, The lift coefficient; , , , These represent the rotational speeds of the four rotors of the eVTOL; The distance from the motor to the center of mass. For the total mass of eVTOL, It is the acceleration due to gravity. , , These are the air damping coefficients of eVTOL on the x, y, and z axes of the body coordinate system, respectively.

[0009] Optionally, the eVTOL energy consumption and charging model in step S1 includes an eVTOL energy consumption model and an eVTOL charging model; The eVTOL energy consumption model is as follows:

[0010] in, eVTOL from the start To the finish line energy consumption For eVTOL at flight speed The baseline cruise power is as follows. To assist in system power consumption, Additional power to counteract the effects of wind fields, The time taken to travel from the starting point i to the ending point j; The eVTOL charging model is as follows:

[0011] in, Let t represent the constant power of the charging station and t be the time. To increase electricity consumption.

[0012] Optionally, step S2 specifically includes: Step S21: Make decisions using the outer DQN agent decision-making algorithm and output the coordinate positions of each eVTOL target node; Step S22: Input the coordinates of each eVTOL target node into the inner DQN agent decision-making algorithm for decision-making, and output the navigation point coordinates and speed of each eVTOL from its current position to the target node; Step S23: Upon reaching a navigation point, the inner DQN agent decision-making algorithm optimizes based on the real-time state and regenerates new navigation point coordinates and velocity.

[0013] Optionally, the outer DQN agent decision-making algorithm in step S2 includes: Define the state space of the agent as follows:

[0014] in, For state space, The current position coordinates of eVTOL; The remaining battery power of the eVTOL; The current task status includes whether a task is being executed and the coordinates of the task being executed; This includes relevant information about other intelligent agents, such as the location and battery level of nearby agents; For known task requests, include the location of the task's start and end points, time requirements, and whether the task has already been accepted; The status of the charging station includes its location coordinates and whether it is available. Define the action space of the intelligent agent as:

[0015] in, For the action space, ( , , The coordinates of the target location include the starting coordinates of all unassigned eVTOLs, the ending coordinates of all unassigned eVTOLs, and the coordinates of all charging stations. The agent optimizes itself through reward and punishment mechanisms, receiving either a reward or a punishment for each action taken. The reward and punishment function is as follows:

[0016]

[0017]

[0018]

[0019] Where R is the reward / punishment function; It is a reward for completing the task; It's a smart charging reward; Penalty for exceeding the time limit; It's a penalty for low battery. It is a penalty for flight energy consumption; The reward coefficient for smart charging; The number of intelligent agents queuing ahead; This is the timeout penalty coefficient; This refers to the timeout duration. Energy cost coefficient; It refers to the energy consumed.

[0020] Optionally, the inner-layer DQN agent decision-making algorithm in step S2 includes: Define the state space:

[0021] in, For state space, For the speed of the intelligent agent, , , For the position of the agent, Remaining battery power This represents the total number of coordinates for which a decision needs to be made. n An index representing a specific navigation point; Define the action space:

[0022] Where A is the action space. The expected speed of the navigation point to be decided. , , The location of the navigation point to be decided; Reward and punishment functions:

[0023]

[0024]

[0025]

[0026] Where R is the reward / punishment function; It is a reward for completing the task; It's a path length penalty; The path length weighting coefficient is d; d is the path length. This is the energy consumption penalty coefficient; It is the energy consumed; It is a time-consuming penalty; t is the time consumption penalty coefficient; t is the time consumed to complete the task.

[0027] Optionally, step S3 specifically includes: Step S31: Calculate the cost between the starting node and the target node based on the multi-dimensional cost function; Step S32: Create the OPEN set and the CLOSED set, put the starting node into the OPEN set, and set the initial state of the CLOSED set to empty; Step S33: Determine if the OPEN set is empty. If it is empty, adjust the starting node. If it is not empty, take the node with the lowest cost in the OPEN set and add it to the CLOSED set, and set that node as the current node. Step S34: If the current node is the destination, end the search, check the CLOSED set, and backtrack to generate the optimal path. Step S35: If the current node is not the endpoint, traverse all adjacent nodes of the current node, calculate the new cost according to the multi-dimensional cost function, and update the adjacent nodes that are not in the OPEN set to the OPEN set. Step S36: Repeat steps S33-S35 until the target node is reached, and output the final optimal reference path.

[0028] Optionally, the multi-dimensional cost function in step S3 is:

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035]

[0036] Where F(n) is the multi-dimensional cost function; G(n) is the actual cost from the starting point to the current node n; H(n) is the heuristic cost from the current node n to the target node; , , )and( , , The coordinates of the current node n and the target node j are respectively the three-dimensional coordinates of the current node n and the target node j. The cost is the path length. It is the actual route length eVTOL travels from the starting point to the ending point; It is the distance length coefficient; This comes at the cost of dynamic airspace risks. To predict the congestion probability of road segment (i, j) based on historical and real-time data; This is the expected delay time; This represents the congestion penalty coefficient. It is the cost of weather impacts; It is the wind field intensity; It is the wind field penalty coefficient; It is the precipitation penalty coefficient; It refers to the intensity of rainfall; It represents the flight altitude and energy consumption cost; z represents the absolute flight altitude. It has a high energy consumption coefficient; Base offset; It represents the cost of avoiding no-fly zones and sensitive areas; l is the shortest distance from the current location to the boundary of the restricted area; For a safe buffer distance; This represents the penalty coefficient for restricted areas.

[0037] Optionally, the expression for dynamically adjusting the heading angle and flight speed of the eVTOL using a PID controller based on the waypoint in step S4 is as follows:

[0038] Where u(t) is the output control quantity, responsible for the correction command of heading angle and flight speed; e(t) is the Euclidean distance deviation between the current position coordinates and the target node coordinates; , , These are the proportional, integral, and differential gain coefficients, respectively.

[0039] The optimal reference path is discretized to obtain path points. A PID controller is then used to dynamically adjust the heading angle and flight speed of the eVTOL based on the path points until the eVTOL reaches the target node.

[0040] Optionally, the PID controller can adaptively adjust according to the real-time status; the PID controller can also feed back the actual flight data to the two-layer DQN reinforcement learning algorithm to update the waypoints; The actual flight data includes the eVTOL's motor speed, actual flight speed, and tracking error.

[0041] The beneficial effects of the present invention are as follows: The present invention provides a method for collaborative task and charging planning of eVTOL cluster based on spatiotemporal uncertainty, which regards each eVTOL as an independent intelligent agent, and performs global and local collaborative decision-making based on the outer layer DQN intelligent agent decision algorithm and the inner layer DQN decision algorithm, thereby generating navigation points and providing a basis for subsequent generation of optimal reference paths; The A* algorithm can be used to integrate airspace dynamic risks, weather effects and no-fly zone restrictions in path planning to generate a robust optimal reference path. Subsequently, task allocation and charging plans can be dynamically adjusted based on the real-time status of eVTOL, and continuous optimization can be achieved through PID path tracking feedback data.

[0042] This invention effectively addresses the spatiotemporal uncertainties in urban airspace through multi-agent collaborative decision-making and online optimization, maximizing the operational efficiency and economic benefits of eVTOL clusters. It can be extended to scenarios such as regional cargo drone scheduling and urban emergency response, and has good portability and application value. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating an eVTOL cluster collaborative task and charging planning method based on spatiotemporal uncertainty provided in an embodiment of the present invention. Figure 2 This is a flowchart illustrating step S2 provided in an embodiment of the present invention; Figure 3 This is a flowchart illustrating step S3 provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the eVTOL dynamic model provided in an embodiment of the present invention. Detailed Implementation

[0044] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. In the following description, the expression "some embodiments" refers to a subset of all possible embodiments; however, it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict.

[0045] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.

[0046] It should be understood that the present invention can be embodied in various forms and should not be construed as being limited to the embodiments set forth herein. Rather, providing these embodiments will make the disclosure thorough and complete, and will fully convey the scope of the invention to those skilled in the art. Furthermore, the terminology used herein is intended only to describe particular embodiments and is not intended to limit the invention. When used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “compose” and / or “comprising,” when used in this specification, identify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups. When used herein, the term “and / or” includes any and all combinations of the associated listed items.

[0047] It should also be noted that when an element is referred to as being "fixed to" another element, it can be directly attached to the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "inner," "outer," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0048] To fully understand this invention, a detailed structure will be presented in the following description to illustrate the technical solution proposed by this invention. Optional embodiments of the invention are described in detail below; however, in addition to these detailed descriptions, the invention may have other embodiments.

[0049] Example 1 Please refer to the attached document. Figure 1 and attached Figure 4 This application provides a method for eVTOL cluster collaborative task and charging planning based on spatiotemporal uncertainty, including the following steps: Step S1: Establish an eVTOL dynamic model based on the flight dynamics characteristics of the quadcopter eVTOL, and establish an eVTOL energy consumption and charging model based on the efficiency of the motor system and the discharge characteristics of the battery. Step S2: Use a reinforcement learning algorithm based on two-layer DQN to perform collaborative decision-making and dynamic adjustment, and output navigation point coordinates. The two-layer DQN reinforcement learning algorithm includes an outer layer DQN agent decision-making algorithm and an inner layer DQN agent decision-making algorithm. Step S3: Input the navigation point coordinates into the A* algorithm. The A* algorithm generates the optimal reference path based on the multi-dimensional cost function. Step S4: Use the PID controller to dynamically adjust the heading angle and flight speed of the eVTOL according to the optimal reference path until the eVTOL reaches the target node.

[0050] Specifically, the eVTOL dynamics model can describe the three-dimensional motion of the eVTOL in urban airspace, that is, the relationship between its motion state (position, velocity, attitude angle) and control input (motor speed); combining motor system efficiency and battery discharge characteristics, an eVTOL energy consumption and charging model is established. The energy consumption model can quantify the impact of flight strategy and environment on battery power during eVTOL flight, and the charging model describes the variation law of charging time and charging power. The dynamics model can ensure that subsequent path planning and eVTOL control conform to actual constraints, and the energy consumption model and charging model can provide a basis for subsequent scheduling decisions. Then, a reinforcement learning algorithm based on a two-layer DQN is used for collaborative decision-making and dynamic adjustment. The outer DQN is responsible for macro-level task allocation and global path planning. The input state is the overall state of the cluster, and the output is the target node. Each agent of the outer DQN decides whether to accept a task based on its own state (such as remaining battery power, current task queue, and surrounding charging station status) and the global state (such as task demand distribution and charging station queuing status). Whether it's performing a task, executing a mission, or heading to a charging station, the inner DQN is responsible for micro-dynamic adjustments. It operates within each eVTOL, taking the real-time state of a single eVTOL as input and outputting local navigation point corrections. In a continuous spatiotemporal coordinate system, the inner DQN plans a series of optimal navigation point coordinates between the starting point and the target point for each agent. Based on the series of navigation point coordinates generated by the inner DQN in step S2 and the corresponding velocities, the A* algorithm is input. The A* algorithm, combined with a multi-dimensional cost function, searches for the optimal reference path, ensuring that the planned optimal reference path can balance future charging costs and mission benefits. The optimal reference path obtained by the A* algorithm is a series of discrete path points. The PID controller dynamically adjusts the heading angle and flight speed of the eVTOL based on the series of discrete path points in the optimal reference path to achieve accurate path tracking. Through the above methods, the eVTOL cluster can operate efficiently and collaboratively in a spatiotemporally uncertain environment, significantly improving the overall operational efficiency and economic benefits of the cluster.

[0051] As an optional implementation, the eVTOL dynamic model in step S1 is:

[0052] in, , , These are the roll angle, pitch angle, and yaw angle along the eVTOL body coordinate system, respectively. , , They are , , angular velocity; , , They are , , angular acceleration; , , It is the relative velocity along the X, Y, and Z axes in a coordinate system with Earth as the reference point; , , It refers to the acceleration along the X, Y, and Z axes in a coordinate system with Earth as the reference point; , , It is the moment of inertia of the eVTOL body about the body coordinate system; , , These are the air damping coefficients of the drone along the X, Y, and Z axes, respectively. It is the moment of inertia of a single rotor about its own axis of rotation; It is the overall effective angular velocity of the four rotors of the eVTOL; , , , These are the actual control inputs for the four rotors of the eVTOL. Controlling the lifting and lowering movement of the eVTOL Controlling the roll motion of the eVTOL Controlling the pitch motion of the eVTOL Controlling the yaw motion of the eVTOL; The air drag coefficient, The lift coefficient; , , , These represent the rotational speeds of the four rotors of the eVTOL; The distance from the motor to the center of mass. For the total mass of eVTOL, It is the acceleration due to gravity. , , These are the air damping coefficients of eVTOL on the x, y, and z axes of the body coordinate system, respectively.

[0053] As an optional implementation, the eVTOL energy consumption and charging model in step S1 includes an eVTOL energy consumption model and an eVTOL charging model. The eVTOL energy consumption model is as follows:

[0054] in, eVTOL from the start To the finish line energy consumption For eVTOL at flight speed The baseline cruise power is as follows. To reduce system power consumption, the auxiliary system includes avionics and flight control systems, cooling systems, and sensor systems; Additional power to counteract the effects of wind fields, The time taken to travel from the starting point i to the ending point j; The eVTOL charging model is as follows:

[0055] in, Let t represent the constant power of the charging station and t be the time. To increase electricity consumption.

[0056] As an optional implementation method, please refer to the appendix. Figure 2 Step S2 specifically includes: Step S21: Make decisions using the outer DQN agent decision-making algorithm and output the coordinate positions of each eVTOL target node; Step S22: Input the coordinates of each eVTOL target node into the inner DQN agent decision-making algorithm for decision-making, and output the navigation point coordinates and speed of each eVTOL from its current position to the target node; Step S23: Upon reaching a navigation point, the inner DQN agent decision-making algorithm optimizes based on the real-time state and regenerates new navigation point coordinates and velocity.

[0057] Specifically, by using an outer DQN for global decision-making and an inner DQN for local decision-making, the outer DQN ensures the efficiency of task allocation and path planning for the entire eVTOL cluster through global optimization. Simultaneously, considering charging resource coordination, the outer DQN passes global decisions to the inner DQN, which then makes local adjustments based on these decisions. The inner DQN responds to environmental disturbances (such as battery degradation) through real-time adjustments, ensuring local path optimization for individual eVTOLs. This dual-layer DQN approach allows for the optimization of both the global strategy of the eVTOL cluster and the local strategy of individual eVTOLs, enabling collaborative path planning. It should be noted that the target node coordinates output by the outer DQN are the coordinates of the destination that the eVTOL needs to reach, depending on whether it is executing a task or heading to a charging station. If it is executing a task, it is the destination coordinate of the task; if it is heading to a charging station, it is the destination coordinate of that charging station. The navigation points output by the inner DQN are a subset of high-quality, continuous points selected by the inner DQN from all points between the starting point and the target node, which can also be understood as path points. The outer DQN selects the optimal target node and provides it to the inner DQN. The inner DQN finds a series of navigation points between the starting point and the target node and provides them to the A* algorithm. The A* algorithm then selects a unique continuous point as the optimal reference path and provides it to the PID. The PID supervises the eVTOL to move along the optimal reference path, and finally, the eVTOL reaches the target node according to the optimal reference path.

[0058] As an optional implementation, the outer DQN agent decision-making algorithm in step S2 includes: Define the state space of the agent as follows:

[0059] in, For state space, The current position coordinates of eVTOL; The remaining battery power of the eVTOL; The current task status includes whether a task is being executed and the coordinates of the task being executed; This includes relevant information about other intelligent agents, such as the location and battery level of nearby agents; For known task requests, include the location of the task's start and end points, time requirements, and whether the task has already been accepted; The status of the charging station includes its location coordinates and whether it is available. Define the action space of the intelligent agent as:

[0060] in, For the action space, ( , , The coordinates of the target location include the starting coordinates of all unassigned eVTOLs, the ending coordinates of all unassigned eVTOLs, and the coordinates of all charging stations. The agent optimizes itself through reward and punishment mechanisms, receiving either a reward or a punishment for each action taken. The reward and punishment function is as follows:

[0061]

[0062]

[0063]

[0064] in, It is a reward for completing the task; It's a smart charging reward; Penalty for exceeding the time limit; It's a penalty for low battery. It is a penalty for flight energy consumption; The reward coefficient for smart charging; The number of intelligent agents queuing ahead; This is the timeout penalty coefficient; This refers to the timeout duration. Energy cost coefficient; It refers to the energy consumed.

[0065] Specifically, It is a reward for completing a task; successfully completing a travel order will earn you a positive reward. It is a smart charging reward. When the battery is low, the smart agent can choose to go to a "high cost-performance" charging station to complete the charging and receive a positive reward. It is a low battery penalty; when the remaining battery is below the safety threshold, a large negative reward is applied to prevent the task from being interrupted. The reward and punishment function of the outer DQN aims to guide the agent to learn to weigh "immediate task benefits" against "future charging costs". That is, when making decisions, the agent not only focuses on completing the task at hand, but also considers the subsequent charging needs, so as to avoid the failure of subsequent tasks or the need for longer charging time due to excessive power consumption, thereby affecting the overall efficiency.

[0066] As an optional implementation, the inner-layer DQN agent decision-making algorithm in step S2 includes: Define the state space:

[0067] in, For state space, For the speed of the intelligent agent, , , For the position of the agent, Remaining battery power This represents the total number of coordinates for which a decision needs to be made. n An index representing a specific navigation point; Define the action space:

[0068] Where A is the action space. The expected speed of the navigation point to be decided. , , The location of the navigation point to be decided; Reward and punishment functions:

[0069]

[0070]

[0071]

[0072] in, It is a reward for completing the task; It's a path length penalty; The path length weighting coefficient is d; d is the path length. This is the energy consumption penalty coefficient; It is the energy consumed; It is a time-consuming penalty; t is the time consumption penalty coefficient; t is the time consumed to complete the task.

[0073] Specifically, the reward and punishment function of the inner DQN aims to guide the agent to learn to balance the agent's needs to complete the task while saving energy and time as much as possible. In other words, while executing the navigation points planned by the outer layer, the inner DQN makes local adjustments to optimize energy and time.

[0074] As an optional implementation method, please refer to the appendix. Figure 3 Step S3 specifically includes: Step S31: Calculate the cost between the starting node and the target node based on the multi-dimensional cost function; Step S32: Create the OPEN set and the CLOSED set, put the starting node into the OPEN set, and set the initial state of the CLOSED set to empty; Step S33: Determine if the OPEN set is empty. If it is empty, adjust the starting node. If it is not empty, take the node with the lowest cost in the OPEN set and add it to the CLOSED set, and set that node as the current node. Step S34: If the current node is the destination, end the search, check the CLOSED set, and backtrack to generate the optimal path. Step S35: If the current node is not the endpoint, traverse all adjacent nodes of the current node, calculate the new cost according to the multi-dimensional cost function, and update the adjacent nodes that are not in the OPEN set to the OPEN set. Step S36: Repeat steps S33-S35 until the target node is reached, and output the final optimal reference path.

[0075] Specifically, after obtaining a series of navigation point coordinates from the inner DQN algorithm, the A* algorithm is used to generate local paths between the navigation points, generating an optimal reference path that takes into account distance, energy consumption, and spatiotemporal risks. These navigation point coordinates will serve as the target points for each call to the A* algorithm to guide the agent to the task point or charging station.

[0076] As an optional implementation, the multi-dimensional cost function in step S3 is characterized by:

[0077]

[0078]

[0079]

[0080]

[0081]

[0082]

[0083]

[0084] Where F(n) is the multi-dimensional cost function; G(n) is the actual cost from the starting point to the current node n; H(n) is the heuristic cost from the current node n to the target node; , , )and( , , The coordinates of the current node n and the target node j are respectively the three-dimensional coordinates of the current node n and the target node j. The cost is the path length. It is the actual route length eVTOL travels from the starting point to the ending point; It is the distance length coefficient; This comes at the cost of dynamic airspace risks. To predict the congestion probability of road segment (i, j) based on historical and real-time data; This is the expected delay time; This represents the congestion penalty coefficient. It is the cost of weather impacts; It is the wind field intensity; It is the wind field penalty coefficient; It is the precipitation penalty coefficient; It refers to the intensity of rainfall; It represents the flight altitude and energy consumption cost; z represents the absolute flight altitude. It has a high energy consumption coefficient; Base offset; It is about avoiding the costs of no-fly zones and sensitive areas; It is the shortest distance from the current location to the boundary of the restricted area, which includes no-fly zones and sensitive areas; For a safe buffer distance; This represents the penalty coefficient for restricted areas.

[0085] Specifically, The cost is the path length, which serves as the basic cost to ensure the path is as short as possible. This comes at the cost of dynamic airspace risks in order to address the uncertainty of traffic congestion; This is the congestion penalty coefficient. The larger the value, the higher the degree of avoidance of time uncertainty. This cost guides eVTOL to avoid areas with a high probability of congestion. This is the cost of meteorological impacts, taking into account the effects of wind field intensity and precipitation intensity on flight safety and energy consumption in order to cope with the uncertainty of weather changes; It is the relationship between flight altitude and energy consumption, and it has a non-linear relationship with both flight altitude and energy consumption. This function encourages the selection of the cruise altitude with optimal energy consumption under the premise of safety. It is the cost of avoiding no-fly zones and sensitive areas, and adopts a graded penalty mechanism to ensure safety. This design imposes an infinite penalty on intrusion behavior and a gradient penalty on approach behavior to achieve effective avoidance. When eVTOL flies in dynamic urban airspace, its cost is affected by spatiotemporal uncertainties such as flight time, energy consumption, airspace traffic flow, and weather changes. Therefore, a multi-dimensional cost function comprehensively evaluates path length, flight altitude, airspace dynamic risks, and meteorological conditions, enabling the A* algorithm to generate an optimal reference path with low energy consumption and high reliability through the multi-dimensional cost function.

[0086] The multidimensional cost function can be continuously calibrated and corrected based on actual flight data and environmental feedback, thereby improving the prediction accuracy and robustness of subsequent path planning and better adapting to spatiotemporal uncertainties.

[0087] As an optional implementation, the expression for dynamically adjusting the heading angle and flight speed of the eVTOL using a PID controller based on the waypoint in step S4 is as follows:

[0088] Where u(t) is the output control quantity, responsible for the correction command of heading angle and flight speed; e(t) is the Euclidean distance deviation between the current position coordinates and the target node coordinates; , , These are the proportional, integral, and differential gain coefficients, respectively.

[0089] The optimal reference path is discretized to obtain path points. A PID controller is then used to dynamically adjust the heading angle and flight speed of the eVTOL based on the path points until the eVTOL reaches the target node.

[0090] As an optional implementation, the PID controller can adaptively adjust according to the real-time status; the PID controller can also feed back the actual flight data to the reinforcement learning algorithm of the two-layer DQN to update the waypoints. The actual flight data includes the eVTOL's motor speed, actual flight speed, and tracking error.

[0091] Specifically, the PID controller can adaptively adjust based on the real-time eVTOL status (e.g., when the remaining battery power is low, the proportional gain is appropriately reduced to smooth flight and save energy; when encountering strong crosswind disturbances, the derivative gain is increased to enhance system damping and suppress trajectory oscillations). Simultaneously, actual flight data is fed back to the central system to update the DQN strategy and correct the path model, ensuring that future navigation points better reflect real flight dynamics and energy consumption characteristics, thus forming a closed-loop optimization.

[0092] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for eVTOL cluster collaborative task and charging planning based on spatiotemporal uncertainty, characterized in that, Includes the following steps: Step S1: Establish an eVTOL dynamic model based on the flight dynamics characteristics of the quadcopter eVTOL, and establish an eVTOL energy consumption and charging model based on the efficiency of the motor system and the discharge characteristics of the battery. Step S2: Use a reinforcement learning algorithm based on two-layer DQN to perform collaborative decision-making and dynamic adjustment, and output navigation point coordinates. The two-layer DQN reinforcement learning algorithm includes an outer layer DQN agent decision-making algorithm and an inner layer DQN agent decision-making algorithm. Step S3: Input the navigation point coordinates into the A* algorithm. The A* algorithm generates the optimal reference path based on the multi-dimensional cost function. Step S4: Use the PID controller to dynamically adjust the heading angle and flight speed of the eVTOL according to the optimal reference path until the eVTOL reaches the target node.

2. The method for eVTOL cluster collaborative task and charging planning based on spatiotemporal uncertainty as described in claim 1, characterized in that, The eVTOL dynamic model in step S1 is: in, , , These are the roll angle, pitch angle, and yaw angle along the eVTOL body coordinate system, respectively. , , They are , , angular velocity; , , They are , , angular acceleration; , , It is the relative velocity along the X, Y, and Z axes of the Earth-based coordinate system; , , It refers to the acceleration along the X, Y, and Z axes in a coordinate system with Earth as the reference point; , , It is the moment of inertia of the eVTOL body about the body coordinate system; , , These are the air damping coefficients of the drone along the X, Y, and Z axes, respectively. It is the moment of inertia of a single rotor about its own axis of rotation; It is the overall effective angular velocity of the four rotors of the eVTOL; , , , These are the actual control inputs for the four rotors of the eVTOL. Controlling the lifting and lowering movement of the eVTOL Controlling the roll motion of the eVTOL Controlling the pitch motion of the eVTOL Controlling the yaw motion of the eVTOL; The air drag coefficient, The lift coefficient; , , , These represent the rotational speeds of the four rotors of the eVTOL; The distance from the motor to the center of mass. For the total mass of eVTOL, It is the acceleration due to gravity. , , These are the air damping coefficients of eVTOL on the x, y, and z axes of the body coordinate system, respectively.

3. The method for eVTOL cluster collaborative task and charging planning based on spatiotemporal uncertainty as described in claim 1, characterized in that, The eVTOL energy consumption and charging model in step S1 includes an eVTOL energy consumption model and an eVTOL charging model. The eVTOL energy consumption model is as follows: in, eVTOL from the start To the finish line energy consumption For eVTOL at flight speed The baseline cruise power is as follows. To assist in system power consumption, Additional power to counteract the effects of wind fields, The time taken to travel from the starting point i to the ending point j; The eVTOL charging model is as follows: in, Let t represent the constant power of the charging station and t be the time. To increase electricity consumption.

4. The method for eVTOL cluster collaborative task and charging planning based on spatiotemporal uncertainty as described in claim 1, characterized in that, Step S2 specifically includes: Step S21: Make decisions using the outer DQN agent decision-making algorithm and output the coordinate positions of each eVTOL target node; Step S22: Input the coordinates of each eVTOL target node into the inner DQN agent decision-making algorithm for decision-making, and output the navigation point coordinates and speed of each eVTOL from its current position to the target node; Step S23: Upon reaching a navigation point, the inner DQN agent decision-making algorithm optimizes based on the real-time state and regenerates new navigation point coordinates and velocity.

5. The method for eVTOL cluster collaborative task and charging planning based on spatiotemporal uncertainty according to claim 4, characterized in that, The outer-layer DQN agent decision-making algorithm in step S2 includes: Define the state space of the intelligent agent as follows: in, For state space, The current position coordinates of eVTOL; The remaining battery power of the eVTOL; The current task status includes whether a task is being executed and the coordinates of the task being executed; This includes relevant information about other intelligent agents, such as the location and battery level of nearby agents; For known task requests, include the location of the task's start and end points, time requirements, and whether the task has already been accepted; The status of the charging station includes its location coordinates and whether it is available. Define the action space of the intelligent agent as follows: in, For the action space, ( , , The coordinates of the target location include the starting coordinates of all unassigned eVTOLs, the ending coordinates of all unassigned eVTOLs, and the coordinates of all charging stations. The agent optimizes itself through reward and punishment mechanisms, receiving either a reward or a punishment for each action taken. The reward and punishment function is as follows: Where R is the reward / punishment function; It is a reward for completing the task; It's a smart charging reward; Penalty for exceeding the time limit; It's a penalty for low battery. It is a penalty for flight energy consumption; The reward coefficient for smart charging; The number of intelligent agents queuing ahead; This is the timeout penalty coefficient; This refers to the timeout duration; Energy cost coefficient; It refers to the energy consumed.

6. The method for eVTOL cluster collaborative task and charging planning based on spatiotemporal uncertainty as described in claim 4, characterized in that, The inner-layer DQN agent decision-making algorithm in step S2 includes: Define the state space: in, For state space, For the speed of the intelligent agent, , , For the location of the agent, Remaining battery power This represents the total number of coordinates for which a decision needs to be made. n An index representing a specific navigation point; Define the action space: Where A is the action space. The desired speed of the navigation point to be decided. , , The location of the navigation point to be decided; Reward and punishment functions: Where R is the reward / punishment function; It is a reward for completing the task; It's a path length penalty; The path length weighting coefficient is d; d is the path length. This is the energy consumption penalty coefficient; It is the energy consumed; It is a time-consuming penalty; t is the time consumption penalty coefficient; t is the time consumed to complete the task.

7. The method for eVTOL cluster collaborative task and charging planning based on spatiotemporal uncertainty as described in claim 1, characterized in that, Step S3 specifically includes: Step S31: Calculate the cost between the starting node and the target node based on the multi-dimensional cost function; Step S32: Create the OPEN set and the CLOSED set, put the starting node into the OPEN set, and set the initial state of the CLOSED set to empty; Step S33: Determine if the OPEN set is empty. If it is empty, adjust the starting node. If it is not empty, take the node with the lowest cost in the OPEN set and add it to the CLOSED set, and set that node as the current node. Step S34: If the current node is the destination, end the search, check the CLOSED set, and backtrack to generate the optimal path. Step S35: If the current node is not the endpoint, traverse all adjacent nodes of the current node, calculate the new cost according to the multi-dimensional cost function, and update the adjacent nodes that are not in the OPEN set to the OPEN set. Step S36: Repeat steps S33-S35 until the target node is reached, and output the final optimal reference path.

8. The method for eVTOL cluster collaborative task and charging planning based on spatiotemporal uncertainty as described in claim 1, characterized in that, The multi-dimensional cost function in step S3 is: Where F(n) is the multi-dimensional cost function; G(n) is the actual cost from the starting point to the current node n; H(n) is the heuristic cost from the current node n to the target node; , , )and( , , The coordinates of the current node n and the target node j are respectively. The cost is the path length. It is the actual route length eVTOL travels from the starting point to the ending point; It is the distance length coefficient; This comes at the cost of dynamic airspace risks. To predict the congestion probability of road segment (i, j) based on historical and real-time data; This is the expected delay time; This represents the congestion penalty coefficient. It is the cost of weather impacts; It is the wind field intensity; It is the wind field penalty coefficient; It is the precipitation penalty coefficient; It refers to the intensity of rainfall; It represents the flight altitude and energy consumption cost; z represents the absolute flight altitude. It has a high energy consumption coefficient; Base offset; It represents the cost of avoiding no-fly zones and sensitive areas; l is the shortest distance from the current location to the boundary of the restricted area; For a safe buffer distance; This represents the penalty coefficient for restricted areas.

9. The method for eVTOL cluster collaborative task and charging planning based on spatiotemporal uncertainty as described in claim 1, characterized in that, The expression for dynamically adjusting the heading angle and flight speed of the eVTOL using a PID controller based on the path point in step S4 is as follows: Where u(t) is the output control quantity, responsible for the correction command of heading angle and flight speed; e(t) is the Euclidean distance deviation between the current position coordinates and the target node coordinates; , , These are the proportional, integral, and differential gain coefficients, respectively. The optimal reference path is discretized to obtain path points. A PID controller is then used to dynamically adjust the heading angle and flight speed of the eVTOL based on the path points until the eVTOL reaches the target node.

10. A method for eVTOL cluster collaborative task and charging planning based on spatiotemporal uncertainty according to any one of claims 1-9, characterized in that, The PID controller can adaptively adjust according to the real-time status; the PID controller can also feed back the actual flight data to the reinforcement learning algorithm of the two-layer DQN to update the waypoints; The actual flight data includes the eVTOL's motor speed, actual flight speed, and tracking error.

Images