Unmanned aerial vehicle charging network dynamic optimization system and method based on multi-agent cooperation

Abstract

The invention discloses an unmanned aerial vehicle charging network dynamic optimization system and method based on multi-agent cooperation. The method comprises the following steps: S1, carrying out system initialization; s2, real data acquisition and processing are carried out; s3, carrying out edge layer rapid decision making; s4, cloud layer global optimization is carried out through dynamic game equilibrium calculation; s5, carrying out multi-agent collaboration and task execution; and S6, carrying out dynamic updating and maintenance. According to the unmanned aerial vehicle charging network dynamic optimization system and method based on multi-agent cooperation, the dynamic game ant colony algorithm is combined with the contract network protocol, the task completion rate can be effectively improved, the algorithm carries out multi-target cost coefficient comprehensive optimization, the path length is shortened, and task time consumption is reduced; the utilization rate of the fast-charging pile is stabilized, the vacancy rate of the slow-charging pile is reduced, the cycle life of the battery is prolonged, and the service life loss of charging pile equipment is reduced; and the ultra-short-term scheduling response delay and the long-term planning full-period maintenance cost are reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent transportation systems, and specifically relates to a system and method for dynamic optimization of a drone charging network based on multi-agent collaboration. Background Art

[0002] Traditional path planning algorithms (such as Dijkstra and A*) struggle to cope with the coupled constraints of dynamic traffic flow and drone battery degradation. Load balancing strategies for charging stations and battery swap stations lack predictability in both temporal and spatial dimensions, leading to localized overloads or idleness. Multi-drone collaborative scheduling suffers from information silos, preventing global optimization. Furthermore, there is a lack of dynamic charging path planning models that consider battery health status. A two-way game mechanism between drone task queues and charging networks has not been established. Furthermore, there is a lack of charging resource allocation strategies with multiple time granularities (minutes and hours).

[0003] Therefore, further improvements are made to the above problems. Summary of the Invention

[0004] The main purpose of the present invention is to provide a dynamic optimization system and method for drone charging networks based on multi-agent collaboration. By combining the dynamic game ant colony algorithm with the contract network protocol, the task completion rate can be effectively improved. The algorithm performs comprehensive optimization of multi-objective cost coefficients, shortens the path length, and reduces task time. It stabilizes the utilization rate of fast charging piles, reduces the idle rate of slow charging piles, extends the battery cycle life, and reduces the life loss of charging pile equipment. It also reduces ultra-short-term scheduling response delays and long-term planning full-cycle maintenance costs.

[0005] To achieve the above objectives, the present invention provides a method for dynamic optimization of a UAV charging network based on multi-agent collaboration, comprising the following steps:

[0006] Step S1: Initialize the system;

[0007] Step S2: collecting and processing real number data;

[0008] Step S3: Make a quick decision at the edge layer;

[0009] Step S4: Perform global optimization of the cloud layer through dynamic game equilibrium calculation;

[0010] Step S5: multi-agent collaboration and task execution;

[0011] Step S6: Perform dynamic update and maintenance.

[0012] As a further preferred technical solution of the above technical solution, step S1 is specifically implemented as follows:

[0013] Step S1.1: Hardware device information upload and parameter configuration. After the drone is powered on, it uploads individual information and initial parameters to the cloud layer through the communication module. After the charging pile is started, it reports the initial status.

[0014] Step S1.2: Data preprocessing and model loading. The edge nodes of the edge layer load the traffic flow prediction model based on LSTM-CNN, initialize the historical traffic data cache, and the cloud layer initializes the parameters of the dynamic game ant colony algorithm to generate the initial pheromone matrix.

[0015] As a further preferred technical solution of the above technical solution, step S2 is specifically implemented as follows:

[0016] Step S2.1: Data is collected through the perception layer. The drone collects data including power, temperature, and location once a second time and uploads it to the edge layer. The charging pile monitors information including load and temperature and uploads it to the cloud layer.

[0017] Step S2.2: Data fusion and noise reduction: Filter the GPS coordinates of the drone to reduce the position error. If the power data drops suddenly beyond the preset value, the redundant sensor switch is triggered to eliminate the abnormal value.

[0018] As a further preferred technical solution of the above technical solution, step S3 is specifically implemented as follows:

[0019] Step S3.1: Use the traffic flow prediction model to predict, input historical traffic data and intersection surveillance video, and output the path congestion index at a preset time in the future;

[0020] Step S3.2: Perform local path planning using dynamic game ant colony algorithm;

[0021] Step S 3.3: Emergency task processing, if the UAV battery level is less than the safety threshold, the edge node directly assigns the nearest low-load charging pile.

[0022] As a further preferred technical solution of the above technical solution, for the dynamic game ant colony algorithm in step S3.2, by simulating the ant pheromone mechanism, the drone selects a path based on the real-time updated pheromone concentration and introduces an elasticity coefficient. When the path congestion index exceeds the critical value, the selection probability of the road section is reduced, thereby improving the detour rate. The specific implementation is as follows:

[0023] Step S3.2.1: The state transition probability formula is:

[0024]

[0025] α is the pheromone concentration weight: controls the influence of pheromone concentration on the selection probability; β is the heuristic factor weight: adjusts the influence of heuristic information on the selection probability; δ is the multi-objective cost weight: adjusts the influence of the multi-objective cost coefficient on path selection;

[0026] Pheromone concentration τ ij (t): reflects historical path selection preferences and is dynamically adjusted through iterative updates;

[0027] Heuristic Factor Take the inverse of the path distance, that is Guide the drone to choose the shortest path;

[0028] Multi-objective cost coefficient Comprehensively consider the path length d ij 、Traffic congestionρ ij , remaining power ΔE k and charging pile life cs : Among them, ω1, ω2, ω3, and ω4 are empirical weights determined by the hierarchical analysis method. MaxCost represents the maximum comprehensive cost value that may occur in the current scenario, which is specifically expressed as MaxCost = ω1·d max +ω2·ρ max +ω3·ΔE max +ω4·Age max ;

[0029] Feasibility function: Ensure that the drone's battery level is not lower than the safety threshold before it reaches the charging station;

[0030] Pheromone update rule: The pheromone matrix is updated according to Updates; drone contributions: Where Q is the pheromone intensity constant, L k is the path length of UAV k, Delay k Delay time for the task; contribution of charging pile Where λ is the load feedback coefficient, D avg is the average distance between charging stations, Reflects the UAV path selection preference, Reflect the charging pile load feedback and form a two-way optimization closed loop;

[0031] Step S3.2.2: Dynamic traffic constraint processing, develop a traffic flow prediction model based on LSTM-CNN: combine spatiotemporal features to predict future traffic flow, the formula is:

[0032]

[0033] Step S3.2.3: Construct real-time path elasticity coefficient φ ij :When the congestion index ρ ij Exceeding the critical value ρ critical When ρ ij >ρ critical =0.7, the path selection tends to be detour;

[0034]

[0035] Among them, k is the core control parameter that controls the real-time path elasticity coefficient;

[0036] Step S3.2.4: Update the state transition probability and exclude the paths that do not meet the feasibility function;

[0037] Step S3.2.5: Generate a preset number of candidate paths and select the optimal path based on probability.

[0038] As a further preferred technical solution of the above technical solution, for step S4, the conflict between the drone mission execution requirements and the charging station resource allocation is balanced through game theory to achieve a global Nash equilibrium, which is specifically implemented as follows:

[0039] Step S4.1: Design of the drone-side utility function:

[0040]

[0041] Among them, Priority k Set different task priorities for logistics and inspection, and adjust them automatically remotely; E is the power urgency; k is the current power; E full Fully charged; TravelTime k The weight of flight time is determined by analytic hierarchy process. The weight parameters α1, α2, and α3 are determined by fitting historical data with analytic hierarchy process.

[0042] Step S4.2: Design of utility function on the charging pile side:

[0043]

[0044] Among them, S cs is the load benefit, β1 is the benefit weight; DeviceAge cs Indicates the equipment age of the charging pile, T life is the design device life threshold of the charging pile CS, which is used to normalize the device age. represents the equipment life utility, β2 is the equipment life attenuation weight, EnergyCost cs To balance economic benefits and equipment health for energy costs, β3 is the weight coefficient of energy costs;

[0045] Step S4.3: Nash equilibrium conditions:

[0046] Drone path optimization: solving by gradient descent method Route k Generate local optimal paths for UAV path selection;

[0047] Charging pile service rate adjustment: solution ServiceRate cs Dynamically adjust the service rate for the charging pile Make the load stable within the preset range.

[0048] As a further preferred technical solution of the above technical solution, step S5 is specifically implemented as follows:

[0049] Step S5.1: Task allocation based on the contract network protocol. The cloud layer publishes the task pool, and the drones bid and calculate the bid scores. The cloud layer evaluates the bids and selects the drone with the highest score. If the scores are the same, they are randomly assigned. Dynamic task allocation design based on the contract network protocol:

[0050] Contract Network Agreement: UAVs compete for tasks through bidding. The evaluation process comprehensively considers factors including priority and remaining battery power to improve resource matching efficiency and support task timeout reallocation logic. The formula is:

[0051]

[0052] Step S5.2: Charging pile service scheduling. The fast charging pile strategy is to reduce the service rate when the load rate is greater than the threshold, and guide the drone to divert traffic through pheromone feedback; the battery swap station strategy is to trigger the isolation mechanism and stop task allocation when the battery health is less than the threshold.

[0053] As a further preferred technical solution of the above technical solution, step S6 is specifically implemented as follows:

[0054] Step S6.1: Pheromone matrix update;

[0055] Step S6.2: Battery health management, calculate and record the attenuation factor of the drone battery and the temperature of the charging station.

[0056] As a further preferred technical solution of the above technical solution, multi-time granularity resource allocation is performed according to different time periods, which is specifically implemented as follows:

[0057] First, ultra-short-term scheduling: Based on real-time allocation of current load, it resolves conflicts between sudden demand and immediate resources. When the drone's battery power falls below the safety threshold, the nearest low-load charging pile is assigned in real time with a response time of ≤30 seconds. The weight of emergency tasks is increased by 50%, and fast charging pile resources are given priority. The formula is:

[0058]

[0059] Among them, S capacity is the total service capacity of the charging pile; Priority k is the mission importance priority coefficient, which indicates the urgency and importance level of the UAV’s current mission; The percentage of the drone’s current remaining power to the total power; S available The remaining allocable charging resources of the charging pile;

[0060] Second, short-term scheduling: Avoid future resource bottlenecks based on predictive models, including:

[0061] LSTM prediction model: It inputs historical traffic flow data and outputs a load forecast for the next 30 minutes. The formula is:

[0062]

[0063] Reserve spare capacity:

[0064]

[0065] Where μ is the adjustment coefficient, which weights the predicted average load to balance the size of the reserved capacity; σ is the adjustment coefficient, which weights the complex variance of the prediction to indicate the degree of consideration of the load fluctuation risk; Var() is the variance prediction of the charging pile load in the future t+1, which is used to measure the degree of fluctuation of the predicted load;

[0066] Third, long-term planning: Optimize the layout of charging piles based on historical data, reduce the cost of the entire life cycle, and optimize the layout of the charging network and the long-term utilization of resources. The formula is:

[0067] Deployment cs =Optimization(∑ k TravelTime k +∑ cs MaintenanceCost cs );

[0068] Among them, ∑ k TravelTime k The total flight time of all UAVs from their current location to the target charging station; ∑ csMaintenanceCost cs The total maintenance cost of all charging piles within 1 to 24 hours;

[0069] Fourth, battery health status perception model:

[0070] Introducing battery attenuation factor: reflects the impact of battery cycle times on capacity. When the number of cycles exceeds N cycle After that, the capacity decays exponentially, and the formula is:

[0071]

[0072] Among them, N charge The cumulative number of times the drone is charged, N cycle is the battery cycle life threshold, when ξ k When (t) = 0, a maintenance alarm is triggered to avoid a sudden drop in charging efficiency or safety hazards caused by battery aging;

[0073] Fifth, establish a nonlinear relationship between charging efficiency and battery temperature: when the temperature deviates from T opt When the charging efficiency η cs (T) decreases according to Gaussian distribution, and the efficiency distribution after temperature correction is:

[0074]

[0075] Where η0 is the nominal charging efficiency; T is the real-time temperature of the battery; T opt is the optimal temperature reference value; σ is the temperature tolerance parameter.

[0076] The beneficial effects of the present invention are:

[0077] 1. The dynamic game ant colony algorithm combined with the contract network protocol can effectively improve the task completion rate. The algorithm performs comprehensive optimization of multi-objective cost coefficients, shortens the path length, and reduces task time.

[0078] 2. Stabilize the utilization rate of fast charging piles, reduce the idle rate of slow charging piles, extend the battery cycle life, and reduce the life loss of charging pile equipment.

[0079] 3. Reduce ultra-short-term scheduling response delays and long-term planning full-cycle maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 is a schematic diagram of the system of the present invention. DETAILED DESCRIPTION

[0081] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are for illustrative purposes only, and those skilled in the art will readily appreciate other obvious variations. The basic principles of the present invention defined in the following description may be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the present invention.

[0082] In the preferred embodiments of the present invention, those skilled in the art should note that the drone, charging pile, etc. involved in the present invention can be regarded as prior art.

[0083] Preferred embodiment.

[0084] The present invention discloses a method for dynamic optimization of a UAV charging network based on multi-agent collaboration, comprising the following steps:

[0085] Step S1: Initialize the system;

[0086] Step S2: collecting and processing real number data;

[0087] Step S3: Make a quick decision at the edge layer (ultra-short-term scheduling of 1 to 5 minutes);

[0088] Step S4: Perform global optimization of the cloud layer through dynamic game equilibrium calculation;

[0089] Step S5: multi-agent collaboration and task execution;

[0090] Step S6: Perform dynamic update and maintenance.

[0091] Specifically, step S1 is implemented as follows:

[0092] Step S1.1: Hardware device information upload and parameter configuration. After the drone is powered on, it uploads individual information and initial parameters to the cloud layer through the communication module. After the charging pile is started, it reports the initial status.

[0093] Step S1.2: Data preprocessing and model loading. The edge nodes of the edge layer load the traffic flow prediction model based on LSTM-CNN, initialize the historical traffic data cache, and the cloud layer initializes the parameters of the dynamic game ant colony algorithm to generate the initial pheromone matrix.

[0094] More specifically, step S2 is implemented as follows:

[0095] Step S2.1: Data is collected through the perception layer. The drone end collects data including power, temperature, and location once every first time (preferably 5 seconds) and uploads it to the edge layer (the edge layer decides whether to upload it to the cloud layer based on the data type. The data that needs to be uploaded to the cloud layer include: data required for global optimization, model training data, and abnormal event records; the data that does not need to be uploaded to the cloud layer include: real-time control instructions, high-frequency collected raw data, and temporary cached data). The charging pile end monitors information including load and temperature at the second time (preferably 1 second) and uploads it to the cloud layer.

[0096] Step S2.2: Data fusion and noise reduction: Filter the GPS coordinates of the drone to reduce the position error. If the power data drops by more than a preset value (preferably 10%, which may be a fault), trigger the redundant sensor switch to eliminate the abnormal value.

[0097] Specifically, step S3 is implemented as follows:

[0098] Step S3.1: Perform a prediction using a traffic flow prediction model (in other embodiments, a Transformer time series prediction or a real-time correction based on a Kalman filter may be used instead of the LSTM-CNN fusion model), input historical traffic data and intersection surveillance footage, and output a path congestion index for a preset time in the future (preferably 5 minutes);

[0099] Step S3.2: Perform local path planning using a dynamic game ant colony algorithm (in other embodiments, a reinforcement learning-Q learning algorithm or a mixed integer linear programming algorithm may be used instead of the dynamic game ant colony algorithm);

[0100] Step S 3.3: Emergency task processing, if the UAV battery level is less than the safety threshold, the edge node directly assigns the nearest low-load charging pile.

[0101] More specifically, for the dynamic game ant colony algorithm in step S3.2 (the dynamic game ant colony algorithm is a hybrid algorithm that combines the traditional ant colony optimization algorithm with game theory ideas, aiming to solve the collaborative optimization problem of multi-agent systems in the environment. The core function of the present invention using this algorithm is to simulate the ant pheromone mechanism. The drone selects a path based on the real-time updated pheromone concentration (pheromone not only reflects the pros and cons of historical paths but also dynamically adjusts path preferences based on real-time data such as dynamic traffic flow and charging pile load) and introduces an elasticity coefficient. When the path congestion index exceeds the critical value, the selection probability of the road section is reduced, thereby improving the detour rate. The specific implementation is as follows:

[0102] Step S3.2.1: The state transition probability formula is:

[0103]

[0104] Among them: α is the pheromone concentration weight: controls the influence of pheromone concentration on the selection probability; β is the heuristic factor weight: adjusts the influence of heuristic information on the selection probability; δ is the multi-objective cost weight: adjusts the influence of the multi-objective cost coefficient on path selection;

[0105] Pheromone concentration τ ij (t): reflects historical path selection preferences and is dynamically adjusted through iterative updates;

[0106] Heuristic Factor Take the inverse of the path distance, that is Guide the drone to choose the shortest path;

[0107] Multi-objective cost coefficient Comprehensively consider the path length d ij 、Traffic congestionρ ij , remaining power ΔE k and charging pile life cs : Among them, ω1, ω2, ω3, and ω4 are empirical weights determined by the hierarchical analysis method. MaxCost represents the maximum comprehensive cost value that may occur in the current scenario, which can be specifically expressed as MaxCost = ω1·d max +ω2·ρ max +ω3·ΔE max +ω4·Age max , which can be static values (historical data or scene characteristics) or dynamic values (real-time traffic, battery, etc. status updates);

[0108] Feasibility function: Ensure that the drone's battery level is not lower than the safety threshold (E reserve );

[0109] Pheromone update rule: The pheromone matrix is updated according to Updates; drone contributions: Where Q is the pheromone intensity constant, L k is the path length of UAV k, Delay k Delay time for the task; contribution of charging pile Where λ is the load feedback coefficient, D avg is the average distance between charging stations, Reflects the UAV path selection preference, Reflect the charging pile load feedback and form a two-way optimization closed loop;

[0110] Step S3.2.2: Dynamic traffic constraint processing, develop a traffic flow prediction model based on LSTM-CNN: combine spatiotemporal features (historical road section data + intersection images) to predict future traffic flow, the formula is:

[0111]

[0112] Step S3.2.3: Construct real-time path elasticity coefficient φ ij :When the congestion index ρ ij Exceeding the critical value ρ critical When ρ ij >ρ critical =0.7, the path selection tends to be detour;

[0113]

[0114] Among them, k is the core control parameter for controlling the real-time path elasticity coefficient: the larger the k value, the more the drone tends to avoid the path; the smaller the k value, the more likely the drone is to choose the path;

[0115] Step S3.2.4: Update the state transition probability and exclude the paths that do not meet the feasibility function; the update method is: comprehensive path length, traffic congestion (via φ ij After correction), remaining power, charging pile life generation Update τ based on the contribution of drones and charging piles ij ; Update the values of α, β, and δ after a fixed number of iterations.

[0116] Step S3.2.5: Generate a preset number (preferably 3) of candidate paths and select the optimal path based on probability.

[0117] Preferably, for step S4, the conflict between the drone mission execution requirements and the charging station resource allocation is balanced by game theory to achieve a global Nash equilibrium, which is specifically implemented as follows:

[0118] Step S4.1: Design of the drone-side utility function:

[0119]

[0120] Among them, Priority k Set different task priorities for logistics and inspection, and adjust them automatically remotely; is the battery urgency (when the battery is 30% lower, the urgency weight α2 increases from 0.3 to 0.5); E k is the current power; E full Fully charged; TravelTime kThe weight of flight time is determined by analytic hierarchy process. The weight parameters α1, α2, and α3 are determined by fitting historical data with analytic hierarchy process.

[0121] Step S4.2: Design of utility function on the charging pile side:

[0122]

[0123] Among them, S cs is the load benefit, β1 is the benefit weight (benefit weight of fast charging pile β1 = 0.7, slow charging pile β1 = 0.5, battery swap station β1 = 0.3); DeviceAge cs Indicates the equipment age of the charging pile, T life is the design device life threshold of the charging pile CS, which is used to normalize the device age. Represents the equipment life utility. The closer the equipment age is to its life limit, the smaller the value is and the lower the utility is. β2 is the equipment life attenuation weight (initial setting is β2 = 0.2. When the remaining equipment life is less than 20%, β2 is increased to 0.4 to extend the maintenance time). EnergyCost cs To balance economic benefits and equipment health for energy costs, β3 is the weight coefficient of energy costs;

[0124] Step S4.3: Nash equilibrium conditions:

[0125] Drone path optimization: solving by gradient descent method Generate local optimal paths for UAV path selection;

[0126] Charging pile service rate adjustment: solution Dynamically adjust the service rate for the charging pile Make the load stable within the preset range (70% to 90%).

[0127] Preferably, step S5 is specifically implemented as follows:

[0128] Step S5.1: Task allocation based on the contract network protocol. The cloud layer publishes the task pool, and the drones bid and calculate the bid scores. The cloud layer evaluates the bids and selects the drone with the highest score. If the scores are the same, they are randomly assigned. Dynamic task allocation design based on the contract network protocol:

[0129] Contract Network Agreement: UAVs compete for tasks through bidding. The evaluation process comprehensively considers factors including priority and remaining battery power to improve resource matching efficiency and support task timeout reallocation logic. The formula is:

[0130]

[0131] In other embodiments, an auction mechanism or a distributed negotiation solution may be used instead of the contract net protocol algorithm.

[0132] Step S5.2: Charging pile service scheduling. The fast charging pile strategy is to reduce the service rate when the load rate is greater than the threshold, and guide the drone to divert traffic through pheromone feedback; the battery swap station strategy is to trigger the isolation mechanism and stop task allocation when the battery health is less than the threshold.

[0133] Preferably, step S6 is specifically implemented as follows:

[0134] Step S6.1: Pheromone matrix update;

[0135] Step S6.2: Battery health management, calculate and record the attenuation factor of the drone battery and the temperature of the charging station.

[0136] Preferably, multi-time granularity resource allocation is performed according to different time periods, and the specific implementation is as follows:

[0137] First, ultra-short-term scheduling (1-5 minutes): Based on real-time allocation of current load, it resolves conflicts between sudden demand and immediate resources. When the drone's battery level is below the safety threshold, the nearest low-load charging pile is assigned in real time with a response time of ≤30 seconds. The weight of emergency tasks is increased by 50%, and fast charging pile resources are given priority. The formula is:

[0138]

[0139] Among them, S capacity is the total service capacity of the charging pile; Priority k is the mission importance priority coefficient, which indicates the urgency and importance level of the UAV’s current mission; The percentage of the drone’s current remaining power to the total power; S available The remaining available charging resources for the charging piles. Second, short-term scheduling (15-30 minutes): avoid future resource bottlenecks based on the prediction model, including:

[0140] LSTM prediction model: It inputs historical traffic flow data and outputs a load forecast for the next 30 minutes. The formula is:

[0141]

[0142] Reserve spare capacity:

[0143]

[0144] Among them, is the adjustment coefficient, which weights the predicted average load to balance the size of the reserved capacity; σ is the adjustment coefficient, which weights the complex variance of the prediction, indicating the degree of consideration of the load fluctuation risk; Var() is the variance prediction of the charging pile load in the future t+1, which is used to measure the degree of fluctuation of the predicted load;

[0145] Third, long-term planning (1-24 hours): Optimize the layout of charging piles based on historical data, reduce the full life cycle cost, and optimize the charging network layout and long-term resource utilization. The formula is:

[0146] Deployment cs =Optimization(∑ k TravelTime k +∑ cs MaintenanceCost cs );

[0147] Among them, ∑ k TravelTime k The total flight time of all drones from their current location to the target charging pile (battery swap station); ∑ cs MaintenanceCost cs The total maintenance cost of all charging piles (battery swap stations) within 1 to 24 hours;

[0148] Fourth, battery health status perception model:

[0149] Introducing battery attenuation factor: reflects the impact of battery cycle times on capacity. When the number of cycles exceeds N ctcle After that, the capacity decays exponentially, and the formula is:

[0150]

[0151] Among them, N charge The cumulative number of times the drone is charged, N cycle is the battery cycle life threshold, when ξ k When (t) = 0, a maintenance alarm is triggered to avoid a sudden drop in charging efficiency or safety hazards caused by battery aging;

[0152] Fifth, establish a nonlinear relationship between charging efficiency and battery temperature: when the temperature deviates from T opt (25℃), charging efficiency η cs (T) decreases according to Gaussian distribution, and the efficiency distribution after temperature correction is:

[0153]

[0154] Where η0 is the nominal (battery temperature is T) charging efficiency; T is the real-time battery temperature; T opt is the optimal temperature reference value; σ is the temperature tolerance parameter.

[0155] like Figure 1 As shown, the present invention also discloses a dynamic optimization system for drone charging network based on multi-agent collaboration, including a physical layer, a perception layer, an edge layer, a cloud layer and an application layer, wherein:

[0156] Physical layer: includes physical devices such as drones and charging stations, responsible for data collection and command execution;

[0157] Perception layer: acquires real-time data including location, power consumption, and traffic through a sensor network, and uses Kalman filtering to reduce noise;

[0158] Edge layer: Deploy traffic flow prediction models to enable rapid local decision-making;

[0159] Cloud layer: Runs global optimization algorithms (such as dynamic game ant colonies) to generate long-term scheduling strategies;

[0160] Application layer: connects to specific logistics and inspection businesses, and defines task priorities and constraints.

[0161] It is worth mentioning that the technical features such as drones and charging piles involved in the patent application of this invention should be regarded as prior art. The specific structure, working principle and possible control method and spatial layout method of these technical features can adopt the conventional selection in the field and should not be regarded as the inventive point of this patent. This patent will not be further elaborated.

[0162] For those skilled in the art, it is still possible to modify the technical solutions described in the aforementioned embodiments, or to make equivalent replacements for some of the technical features therein. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A dynamic optimization method for drone charging network based on multi-agent collaboration, characterized in that: The following steps are involved: Step S1: Initialize the system; Step S2: collecting and processing real number data; Step S3: Make a quick decision at the edge layer; Step S4: Perform global optimization of the cloud layer through dynamic game equilibrium calculation; Step S5: multi-agent collaboration and task execution; Step S6: Perform dynamic update and maintenance.

2. The method for dynamic optimization of drone charging network based on multi-agent collaboration according to claim 1 is characterized in that: Step S1 is specifically implemented as follows: Step S1.1: Hardware device information upload and parameter configuration. After the drone is powered on, it uploads individual information and initial parameters to the cloud layer through the communication module. After the charging pile is started, it reports the initial status. Step S1.2: Data preprocessing and model loading. The edge nodes of the edge layer load the traffic flow prediction model based on LSTM-CNN, initialize the historical traffic data cache, and the cloud layer initializes the parameters of the dynamic game ant colony algorithm to generate the initial pheromone matrix.

3. The method for dynamic optimization of drone charging network based on multi-agent collaboration according to claim 2 is characterized in that: Step S2 is specifically implemented as follows: Step S2.1: Data is collected through the perception layer. The drone collects data including power, temperature, and location once a second time and uploads it to the edge layer. The charging pile monitors information including load and temperature and uploads it to the cloud layer. Step S2.2: Data fusion and noise reduction: Filter the GPS coordinates of the drone to reduce the position error. If the power data drops suddenly beyond the preset value, the redundant sensor switch is triggered to eliminate the abnormal value.

4. The method for dynamic optimization of drone charging network based on multi-agent collaboration according to claim 3 is characterized in that: Step S3 is specifically implemented as follows: Step S3.1: Use the traffic flow prediction model to predict, input historical traffic data and intersection surveillance video, and output the path congestion index at a preset time in the future; Step S3.2: Perform local path planning using dynamic game ant colony algorithm; Step S 3.3: Emergency task processing, if the UAV battery level is less than the safety threshold, the edge node directly assigns the nearest low-load charging pile.

5. The method for dynamic optimization of drone charging network based on multi-agent collaboration according to claim 4 is characterized in that: For the dynamic game ant colony algorithm in step S3.2, by simulating the ant pheromone mechanism, the drone selects a path based on the real-time updated pheromone concentration and introduces an elasticity coefficient. When the path congestion index exceeds the critical value, the probability of selecting the road section is reduced, thereby improving the detour rate. The specific implementation is as follows: Step S3.2.1: The state transition probability formula is: α is the pheromone concentration weight: controls the influence of pheromone concentration on the selection probability; β is the heuristic factor weight: adjusts the influence of heuristic information on the selection probability; δ is the multi-objective cost weight: adjusts the influence of the multi-objective cost coefficient on path selection; Pheromone concentration τ ij (t): reflects historical path selection preferences and is dynamically adjusted through iterative updates; Heuristic Factor Take the inverse of the path distance, that is Guide the drone to choose the shortest path; Multi-objective cost coefficient Comprehensively consider the path length d ij 、Traffic congestionρ ij , remaining power ΔE k and charging pile life cs : Among them, ω1, ω2, ω3, and ω4 are empirical weights determined by the hierarchical analysis method. MaxCost represents the maximum comprehensive cost value that may occur in the current scenario, which is specifically expressed as MaxCost = ω1·d max +ω2·ρ max +ω3·ΔE max +ω4·Age max ; Feasibility function: Ensure that the drone's battery level is not lower than the safety threshold before it reaches the charging station; Pheromone update rule: The pheromone matrix is updated according to Updates; drone contributions: Where Q is the pheromone intensity constant, L k is the path length of UAV k, Delay k Delay time for the task; contribution of charging pile Where λ is the load feedback coefficient, D avg is the average distance between charging stations, Reflects the UAV path selection preference, Reflect the charging pile load feedback and form a two-way optimization closed loop; Step S3.2.2: Dynamic traffic constraint processing, develop a traffic flow prediction model based on LSTM-CNN: combine spatiotemporal features to predict future traffic flow, the formula is: Step S3.2.3: Construct real-time path elasticity coefficient φ ij :When the congestion index ρ ij Exceeding the critical value ρ critical When ρ ij >ρ critical =0.7, the path selection tends to be detour; Among them, k is the core control parameter that controls the real-time path elasticity coefficient; Step S3.2.4: Update the state transition probability and exclude the paths that do not meet the feasibility function; Step S3.2.5: Generate a preset number of candidate paths and select the optimal path based on probability.

6. The method for dynamic optimization of drone charging network based on multi-agent collaboration according to claim 5 is characterized in that: For step S4, the conflict between the UAV mission execution requirements and the charging station resource allocation is balanced through game theory to achieve a global Nash equilibrium. The specific implementation is as follows: Step S4.1: Design of the drone-side utility function: Among them, Priority k Set different task priorities for logistics and inspection, and adjust them automatically remotely; E is the power urgency; k is the current power; E full Fully charged; TravelTime k The weight of flight time is determined by analytic hierarchy process. The weight parameters α1, α2, and α3 are determined by fitting historical data with analytic hierarchy process. Step S4.2: Design of utility function on the charging pile side: Among them, S cs is the load benefit, β1 is the benefit weight; DeviceAge cs Indicates the equipment age of the charging pile, T life is the design device life threshold of the charging pile CS, which is used to normalize the device age. represents the equipment life utility, β2 is the equipment life attenuation weight, EnergyCost cs To balance economic benefits and equipment health for energy costs, β3 is the weight coefficient of energy costs; Step S4.3: Nash equilibrium conditions: Drone path optimization: solving by gradient descent method Route k Generate local optimal paths for UAV path selection; Charging pile service rate adjustment: solution ServiceRate cs Dynamically adjust the service rate for the charging pile Make the load stable within the preset range.

7. The method for dynamic optimization of drone charging network based on multi-agent collaboration according to claim 6 is characterized in that: Step S5 is specifically implemented as follows: Step S5.1: Task allocation based on the contract network protocol. The cloud layer publishes the task pool, and the drones bid and calculate the bid scores. The cloud layer evaluates the bids and selects the drone with the highest score. If the scores are the same, they are randomly assigned. Dynamic task allocation design based on the contract network protocol: Contract Network Agreement: UAVs compete for tasks through bidding. The evaluation process comprehensively considers factors including priority and remaining battery power to improve resource matching efficiency and support task timeout reallocation logic. The formula is: Step S5.2: Charging pile service scheduling. The fast charging pile strategy is to reduce the service rate when the load rate is greater than the threshold, and guide the drone to divert traffic through pheromone feedback; the battery swap station strategy is to trigger the isolation mechanism and stop task allocation when the battery health is less than the threshold.

8. The method for dynamic optimization of drone charging network based on multi-agent collaboration according to claim 7 is characterized in that: Step S6 is specifically implemented as follows: Step S6.1: Pheromone matrix update; Step S6.2: Battery health management, calculate and record the attenuation factor of the drone battery and the temperature of the charging station.

9. The method for dynamic optimization of drone charging network based on multi-agent collaboration according to claim 8, characterized in that: Allocate resources at multiple time granularities according to different time periods. The specific implementation is as follows: First, ultra-short-term scheduling: Based on real-time allocation of current load, it resolves conflicts between sudden demand and immediate resources. When the drone's battery power falls below the safety threshold, the nearest low-load charging pile is assigned in real time with a response time of ≤30 seconds. The weight of emergency tasks is increased by 50%, and fast charging pile resources are given priority. The formula is: Among them, S capacity is the total service capacity of the charging pile; Priority k is the mission importance priority coefficient, which indicates the urgency and importance level of the UAV’s current mission; The percentage of the drone’s current remaining power to the total power; S available The remaining allocable charging resources of the charging pile; Second, short-term scheduling: Avoid future resource bottlenecks based on predictive models, including: LSTM prediction model: It inputs historical traffic flow data and outputs a load forecast for the next 30 minutes. The formula is: Reserve spare capacity: Where μ is the adjustment coefficient, which weights the predicted average load to balance the size of the reserved capacity; σ is the adjustment coefficient, which weights the complex variance of the prediction to indicate the degree of consideration of the load fluctuation risk; Var() is the variance prediction of the charging pile load in the future t+1, which is used to measure the degree of fluctuation of the predicted load; Third, long-term planning: Optimize the layout of charging piles based on historical data, reduce the cost of the entire life cycle, and optimize the layout of the charging network and the long-term utilization of resources. The formula is: Deployment cs =Optimization(∑ k TravelTime k +∑ cs MaintenanceCost cs ); Among them, ∑ k TravelTime k The total flight time of all UAVs from their current location to the target charging station; ∑ cs MaintenanceCost cs The total maintenance cost of all charging piles within 1 to 24 hours; Fourth, battery health status perception model: Introducing battery attenuation factor: reflects the impact of battery cycle times on capacity. When the number of cycles exceeds N cycle After that, the capacity decays exponentially, and the formula is: Among them, N charge The cumulative number of times the drone is charged, N cycle is the battery cycle life threshold, when ξ k When (t) = 0, a maintenance alarm is triggered to avoid a sudden drop in charging efficiency or safety hazards caused by battery aging; Fifth, establish a nonlinear relationship between charging efficiency and battery temperature: when the temperature deviates from T opt When the charging efficiency η cs (T) decreases according to Gaussian distribution, and the efficiency distribution after temperature correction is: Where η0 is the nominal charging efficiency; T is the real-time temperature of the battery; T opt is the optimal temperature reference value; σ is the temperature tolerance parameter.

10. A multi-agent collaborative unmanned aerial vehicle charging network dynamic optimization system, applied to a multi-agent collaborative unmanned aerial vehicle charging network dynamic optimization method according to any one of claims 1 to 9, characterized in that: It includes the physical layer, perception layer, edge layer, cloud layer and application layer, among which: Physical layer: includes physical devices such as drones and charging stations, responsible for data collection and command execution; Perception layer: acquires real-time data including location, power consumption, and traffic through a sensor network, and uses Kalman filtering to reduce noise; Edge layer: Deploy traffic flow prediction models to enable rapid local decision-making; Cloud layer: runs global optimization algorithms to generate long-term scheduling strategies; Application layer: connects to specific logistics and inspection businesses, and defines task priorities and constraints.

Images