Formation control method, system and equipment for photovoltaic panel cleaning unmanned aerial vehicle group, and medium

Abstract

The invention relates to the technical field of unmanned aerial vehicle formation control, and provides a formation control method, system, device and medium for a photovoltaic panel cleaning unmanned aerial vehicle group, and the method comprises the steps: regularly obtaining the task execution information of the unmanned aerial vehicle group, combining a deep learning model to obtain a target formation, carrying out the formation switching, and carrying out the cleaning of a to-be-cleaned photovoltaic panel according to the distribution information of the to-be-cleaned photovoltaic panel. Based on a preset path planning algorithm, cluster task path information in a formation transformation period is obtained, and according to the target formation and a preset formation, an optimization target function is established, and based on an improved differential evolution algorithm, formation establishment information is obtained; and performing formation control on the unmanned aerial vehicle group based on a formation control task sequence obtained by performing formation cleaning task division according to the group task path information, a preset formation control task optimization model and formation establishment information. According to the invention, the formation of the unmanned aerial vehicle group can be adaptively replaced, the anti-interference performance and flexibility of the formation are ensured, and the photovoltaic panel cleaning efficiency and the operation safety of the unmanned aerial vehicle group are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of drone formation control, and in particular to a formation control method, system, equipment and medium for a photovoltaic panel cleaning drone group. Background Art

[0002] Excessive dust accumulation on photovoltaic panels will reduce their power generation efficiency. Using multiple drones to coordinate cleaning can effectively improve the cleaning efficiency of photovoltaic panels.

[0003] Existing formation control methods for drone swarms primarily establish inter-drone connections based on a specific topological matrix, maintain a rigid formation using a fixed distance constraint matrix, and implement self-adjustment of the drone swarm's formation by designing a formation controller based on consistency theory. However, certain application limitations exist when it comes to photovoltaic panel cleaning tasks: 1) Targeted cleaning of only those photovoltaic panels with dust accumulation can improve drone cleaning efficiency, but the specific topological matrix and rigid formation make it difficult to simultaneously address multiple objectives while ensuring formation stability, reducing overall cleaning efficiency. 2) The need to perform cleaning tasks close to the photovoltaic panels necessitates real-time consideration of collision and obstacle avoidance during both formation maintenance and expansion control. However, existing consistency-based formation control methods struggle to incorporate collision and obstacle avoidance into state adjustments, making it difficult to ensure drone safety. Therefore, there is an urgent need for a formation control method that can improve photovoltaic panel cleaning efficiency while ensuring the safety of drone swarms. Summary of the Invention

[0004] The purpose of the present invention is to provide a formation control method for a swarm of photovoltaic panel cleaning drones, which is based on a deep learning-based formation adaptive transformation strategy, combined with a formation information dynamic optimization mechanism that takes balancing communication delay, formation anti-interference and formation flexibility as optimization goals and is solved by an improved differential evolution algorithm, and a formation maintenance and formation scaling task coordination control mechanism that considers drone safety to perform reliable formation control on a swarm of photovoltaic panel cleaning drones. While achieving adaptive replacement of the drone swarm's formation, it can ensure the formation's anti-interference and flexibility, thereby improving the photovoltaic panel cleaning efficiency and the safety of the drone swarm's operation.

[0005] In order to achieve the above objectives, a formation control method, system, equipment and medium for a swarm of photovoltaic panel cleaning drones are provided.

[0006] In a first aspect, an embodiment of the present invention provides a formation control method for a swarm of photovoltaic panel cleaning drones, the method comprising the following steps:

[0007] Obtaining mission execution information of the drone swarm according to a preset formation change cycle, and obtaining a target formation of the drone swarm based on the mission execution information and a pre-built deep learning model;

[0008] After the formation of the drone group is switched to the target formation, the group's task path information within the current formation change cycle is obtained based on a preset path planning algorithm according to the distribution information of the photovoltaic panels to be cleaned;

[0009] According to the target formation and the preset formation formation optimization objective function, formation formation information is obtained based on the improved differential evolution algorithm; the formation formation information includes a communication topology matrix and a distance constraint matrix; the preset formation formation optimization objective function is constructed with balancing communication delay, formation anti-interference and formation flexibility as the optimization goal;

[0010] The formation clearing tasks are divided according to the group task path information to obtain a formation control task sequence, and the formation control of the drone group is performed according to the formation control task sequence, a preset formation control task optimization model and the formation formation information.

[0011] In a second aspect, an embodiment of the present invention provides a formation control system for a photovoltaic panel cleaning drone swarm, the system comprising:

[0012] A target formation acquisition module is used to obtain task execution information of the drone swarm according to a preset formation change cycle, and obtain the target formation of the drone swarm based on the task execution information and a pre-built deep learning model;

[0013] A path information acquisition module is used to obtain the group mission path information within the current formation change cycle based on the preset path planning algorithm according to the distribution information of the photovoltaic panels to be cleaned after the formation of the group of drones is switched to the target formation;

[0014] a formation information acquisition module, configured to obtain formation formation information based on the target formation shape and a preset formation formation optimization objective function, based on an improved differential evolution algorithm; the formation formation information includes a communication topology matrix and a distance constraint matrix; the preset formation formation optimization objective function is constructed with a balance between communication delay, formation anti-interference and formation flexibility as the optimization goal;

[0015] The formation control module is used to divide the formation clearing tasks according to the group task path information, obtain a formation control task sequence, and perform formation control on the drone group according to the formation control task sequence, a preset formation control task optimization model and the formation formation information.

[0016] In a third aspect, an embodiment of the present invention further provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0017] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0018] The present invention provides a formation control method, system, device and medium for a photovoltaic panel cleaning drone swarm. The method realizes obtaining task execution information of the drone swarm according to a preset formation transformation cycle, obtaining the target formation of the drone swarm according to the task execution information and a pre-built deep learning model, switching the formation of the drone swarm to the target formation, obtaining the swarm task path information within the current formation transformation cycle based on the distribution information of the photovoltaic panels to be cleaned and based on a preset path planning algorithm, then obtaining the swarm task path information based on the target formation formation and a preset formation formation optimization objective function constructed with balancing communication delay, formation anti-interference and formation flexibility as optimization objectives, obtaining formation formation information including a communication topology matrix and a distance constraint matrix based on an improved differential evolution algorithm, dividing the formation cleaning tasks according to the swarm task path information to obtain a formation control task sequence, and performing formation control on the drone swarm according to the formation control task sequence, the preset formation control task optimization model and the formation formation information. Compared with the existing technology, the formation control method of the photovoltaic panel cleaning drone swarm is based on the deep learning formation adaptive transformation strategy, combined with the formation formation information dynamic optimization mechanism that takes balancing communication delay, formation anti-interference and formation flexibility as optimization goals and adopts the improved differential evolution algorithm to solve, and the formation maintenance and formation scaling task coordination control mechanism considering the safety of drones to achieve reliable control of the photovoltaic panel cleaning drone swarm formation. It can not only adaptively change the formation of the drone swarm according to environmental changes and mission requirements, but also ensure the anti-interference and flexibility of the swarm formation, thereby effectively improving the photovoltaic panel cleaning efficiency and the safety of the drone swarm operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 1 is a flow chart of a formation control method for a swarm of photovoltaic panel cleaning drones according to an embodiment of the present invention;

[0020] Figure 2 Schematic diagram of the training construction of the deep learning module in an embodiment of the present invention;

[0021] Figure 3 1 is a flow chart of an improved differential evolution algorithm according to an embodiment of the present invention;

[0022] Figure 4 1 is a schematic diagram of a formation control process under a formation maintenance and scaling task in an embodiment of the present invention;

[0023] Figure 5 2 is a schematic structural diagram of a formation control system for a photovoltaic panel cleaning drone swarm according to an embodiment of the present invention;

[0024] Figure 6 is an internal structural diagram of a computer device according to an embodiment of the present invention;

[0025] Explanation of the accompanying symbols: 1. Target formation acquisition module; 2. Path information acquisition module; 3. Formation information acquisition module; 4. Formation control module. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and beneficial effects of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. Obviously, the embodiments described below are part of the embodiments of the present invention and are only used to illustrate the present invention, but are not used to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0027] In one embodiment, Figure 1 As shown, a formation control method for a photovoltaic panel cleaning drone swarm is provided, comprising the following steps:

[0028] S11. According to a preset formation change cycle, the task execution information of the drone group is obtained, and the target formation of the drone group is obtained based on the task execution information and a pre-built deep learning model; wherein, the preset formation change cycle can be understood as the time interval for changing the formation of the drone group. The specific duration can be set according to actual application requirements and is not specifically limited here.

[0029] The task execution information can be understood as the relevant information of the drone swarm performing the photovoltaic panel cleaning task corresponding to the current formation change cycle, which can be used to screen the optimal formation formation for the current formation change cycle, preferably including the current swarm position information, target position information and obstacle relative position information; wherein, the current swarm position information can be understood as the starting position of the drone swarm within the current formation change cycle, the target position information can be understood as the target photovoltaic panel position where the drone swarm needs to perform the cleaning task within the current formation change cycle, and the obstacle relative position information can be understood as the relative position between the obstacle and the target position. Specifically, the step of obtaining the target formation formation of the drone swarm based on the task execution information and the pre-built deep learning model includes:

[0030] Traverse the various formations in the preset formation library, and input the various formations and the task execution information into the deep learning model to predict the energy consumption of task execution, and obtain the corresponding formation energy consumption prediction value; wherein, the preset formation library can be understood as a formation database of the drone group pre-constructed based on the drone formation data in different scenarios. The specific construction steps can be referred to the relevant existing technology implementation and will not be described in detail here.

[0031] In principle, the deep learning model used to predict the energy consumption of different formations in this embodiment can adopt any neural network model that can achieve the corresponding function. However, in order to improve the efficiency of energy consumption prediction while ensuring the accuracy of energy consumption prediction, and thus provide a reliable guarantee for the formation switching efficiency of the drone group, this embodiment preferably builds the required deep learning model based on multilayer perceptron (MLP) training, such as Figure 2 As shown in Figure 2, the construction process of the deep learning model includes: setting up K groups of tasks containing random obstacles, random initial points, and random target points representing the photovoltaic panels to be cleaned in the simulation environment Gazebo; traversing various formations in the preset formation library, taking the formation, target point information, initial point information, and the relative position between the obstacle and the target point as input, performing task simulation and calculating the energy required for different formations to reach the target point; taking the formation, target point information, initial point information, and the relative position between the obstacle and the target point as input X, and taking the energy required for the corresponding task execution as output Y o , construct a training data set for the deep learning model; perform offline training on the multi-layer perceptron based on the training database, and finally obtain a deep learning model.

[0032] The minimum energy consumption prediction value among the formation energy consumption prediction values ​​corresponding to the various formation formations is obtained, and the formation formation corresponding to the minimum energy consumption prediction value is used as the target formation formation of the drone group.

[0033] This embodiment adopts a formation self-transformation strategy based on deep learning, which enables the drone group to adaptively change its formation in an interval period based on mission and environmental information, thereby improving the operating efficiency of drones in real-time missions.

[0034] S12. After the formation of the drone swarm is switched to the target formation, the swarm task path information within the current formation change cycle is obtained based on the preset path planning algorithm according to the distribution information of the photovoltaic panels to be cleaned; wherein the distribution information of the photovoltaic panels to be cleaned can be understood as the position distribution of the remaining uncleaned photovoltaic panels in the photovoltaic power station, which can be updated based on the position information of all photovoltaic panels in the photovoltaic power station and the position information of the photovoltaic panels that have been cleaned. The corresponding swarm task path information can be understood as the drone swarm execution cleaning task path obtained by performing cleaning task path rule analysis on the distribution information of the photovoltaic panels to be cleaned based on the formation change cycle duration, and the formation maneuvering information of the drone swarm at each moment in the path. The specific acquisition process may include:

[0035] First, according to the distribution information of the photovoltaic panels to be cleaned and the duration of the formation change cycle, the next target point (the location of the next photovoltaic panel to be cleaned) is selected. Based on the photovoltaic power station grid map and obstacle information, the A* algorithm is used to perform path planning to avoid obstacles and obtain the fleet task path. Assume that in a large-scale photovoltaic power station scenario, photovoltaic panels are widely distributed and there are some obstacles (such as maintenance vehicles, monitoring equipment, etc.). First, the map of the photovoltaic power station is gridded, the location of the photovoltaic panel to be cleaned is set as the target point, and the initial position of the drone is set as the starting point. The A* algorithm is used to comprehensively consider the distance cost from the starting point to the target point (such as Euclidean distance) and the heuristic cost of path search (such as the straight-line distance to the target point) to search for the initial planned path from the drone's initial position to the target photovoltaic panel location. During the search process, the cost of each grid node is continuously evaluated, and the node with the smallest cost is selected for expansion until the target point is found or it is determined that no path exists. For example, in a 100×100 grid map, the A* algorithm can quickly find the initial planned path from the lower left corner (initial position) to the upper right corner (target photovoltaic panel location), avoiding the obstacle area in the middle.

[0036] After obtaining the swarm mission path within the current formation change cycle through the above method steps, the formation maneuvering information of the drone swarm in the swarm mission path (including the speed, angular velocity and formation scaling information of the drone formation maneuver at each moment, etc.) can also be obtained based on the pre-set maneuvering time step as the desired tracking target when the subsequent formation controller performs formation control: 1) The speed information of the drone formation at each moment is calculated based on the distance between two adjacent points on the swarm mission path and the maneuvering time step; 2) The angular velocity information of the drone formation at each moment is calculated according to the turning situation of the drone on the swarm mission path; 3) The formation scaling strategy is determined according to the distribution density and area size of the photovoltaic panels to be cleaned; if the photovoltaic panels are densely distributed, the formation spacing needs to be reduced when approaching the target area to improve the cleaning efficiency; if the distribution is sparse, the formation spacing is appropriately expanded; by calculating the position offset of each maneuvering time step, the scaling information of the formation at each moment can be obtained, so that the drone formation can flexibly adjust the formation size according to the actual situation and efficiently complete the photovoltaic panel cleaning task.

[0037] S13. According to the target formation shape and the preset formation formation optimization objective function, the formation formation information is obtained based on the improved differential evolution algorithm; the formation formation information includes a communication topology matrix and a distance constraint matrix, and the communication topology matrix determines the communication connection between the UAVs, affecting the information transmission path in the multi-machine system, and the distance constraint matrix provides restrictions on the relative position relationship between the UAVs to ensure the stability of the formation.

[0038] The preset formation formation optimization objective function in this embodiment can be understood as a multi-objective optimization function for obtaining the communication topology matrix and distance constraint matrix of the drone group corresponding to the target formation formation, and used as the fitness evaluation function in the improved differential evolution algorithm. In order to ensure that the formation formation can take into account the communication efficiency, anti-interference ability and difficulty of formation deformation adjustment at the same time, this embodiment preferably constructs the preset formation formation optimization objective function with the balance of communication delay, formation anti-interference and formation flexibility as the optimization goal, that is, based on the obtained target formation formation, considering the communication delay, formation anti-interference and task allocation flexibility, the three sub-objective functions of communication delay, anti-interference and flexibility are designed respectively, and the three sub-objective functions are weightedly combined to form the final formation formation optimization objective function, which is expressed as:

[0039] J=w1·J delay +w2·J robust +w3·J flex

[0040] Where,

[0041]

[0042] Among them, J delay represents the total communication delay between all connected drones; t ij (k ij ) represents the communication delay between UAV i and UAV j, including sending delay, propagation delay and processing delay, and the propagation delay is mainly related to the distance constraint; a ij Represents the connection relationship variable between UAV i and UAV j, which is an element in the communication topology matrix, a ij ∈{0,1},a ij =1 means that UAV i and UAV j have communication connection, a ij =0 means there is no communication connection between UAV i and UAV j; k ij Represents the distance between UAV i and UAV j, which is an element in the distance constraint matrix; J robust represents the impact of external interference (such as airflow disturbance, etc.) on the UAV formation, and represents the anti-interference performance of the formation. It is the average value of the deviation of each UAV in the formation relative to the expected position obtained by Q disturbance simulations under the target formation; e q (a ij ,k ij ) means in a ij and k ij Under the constrained formation, the deviation of the relative expected position between UAV i and UAV j during the qth simulated disturbance; Q represents the total number of simulated disturbances; J flex Indicates the difficulty of changing the formation topology to adapt to the mission scenario; c ijrepresents the cost of changing the connection relationship and distance constraint between UAV i and UAV j, which can be set based on actual application requirements; and They represent the connection relationship between UAV i and UAV j before and after the formation topology structure changes; N represents the total number of UAVs in the UAV swarm; w1, w2 and w3 represent weight coefficients, which can be set according to actual application requirements, as long as w1+w2+w3=1 is satisfied.

[0043] In practical applications, the above t ij (k ij ), e q (a ij ,k ij ) and c ij The acquisition process is as follows:

[0044] 1) In actual application scenarios, assuming that the fixed parameter values ​​of the transmission delay and processing delay of the drone communication module are known (these values ​​are usually determined by the technical specifications of the communication module), the propagation delay can be calculated based on the distance k between drone i and drone j. ij And the signal propagation speed v is used to calculate, that is, the propagation delay is k ij / v; add the propagation delay, sending delay and processing delay to get t ij (k ij ); With t between all connected drones ij (k ij ), we can calculate the communication delay sub-objective function value J delay ;

[0045] 2) In the simulation environment, various simulated external disturbances (such as airflows of different strengths and directions) are imposed on the UAV formation, and the deviation of each UAV from the expected position under each disturbance is recorded. q (a ij ,k ij ); the deviation values ​​of all UAVs in the formation are statistically averaged to obtain the deviation of the formation under this simulated disturbance; after Q simulated disturbances, the deviation of the formation under each simulated disturbance is averaged to obtain the interference rejection objective function value J robust ;

[0046] 3) In actual calculation, C can be defined according to the specific mission requirements and the complexity of the formation adjustment. ij If changing the connection relationship requires reconfiguring the communication link, adjusting the control algorithm parameters, and other operations, and changing the distance constraint is relatively easy, and the corresponding cost of changing the distance constraint is negligible, then C can be quantified based on the computing resources and time costs required for these operations. ijIf reconfiguring the communication link requires a unit of computing resources and adjusting the control algorithm parameters requires b units of computing resources, then we can define C ij =a+b; calculate C by all possible changes in the connection relationship ij Then add up to get the flexibility objective function value J flex It should be noted that, in practical applications, if the cost of changing the distance constraint needs to be quantified, then when calculating the flexibility objective function value, it is only necessary to add the cost of changing the connection relationship and the cost of changing the distance constraint.

[0047] In order to obtain a reasonable communication topology matrix and distance constraint matrix, we use Figure 3 The improved differential evolution algorithm shown is used for optimization and solution: 1) randomly generating an initial population; 2) calculating the fitness of each parent individual; 3) performing mutation operations using the DE / rand / 1 operator and the DE / current-to-best / 1 operator; 4) performing crossover operations using binomial crossover; 5) calculating the fitness of the offspring and comparing it with the fitness of the parent to select the next generation population; 6) iterating repeatedly until the fitness is less than a threshold and the optimal individual is obtained. Specifically, the step of obtaining formation formation information based on the improved differential evolution algorithm according to the target formation formation and the preset formation formation optimization objective function includes:

[0048] According to the target formation, a set of initial solutions of the communication topology matrix and the distance constraint matrix are randomly generated as original individuals to construct an initial population; that is, each original individual in the initial population represents a possible UAV formation connection and distance configuration.

[0049] The fitness value of each of the original individuals in the initial population is calculated based on the preset formation optimization objective function; the lower the fitness value, the better the solution is in meeting the communication delay, anti-interference and flexibility requirements.

[0050] A number of individuals to be mutated are randomly selected from the initial population, and mutation operations are performed on each of the individuals to be mutated based on a dynamic scaling factor to generate corresponding mutated individuals; wherein the mutation operation generates new individuals by perturbing individuals in the population. In this embodiment, DE / rand / 1 and DE / current-to-best / 1 operators are used to perform the mutation operation; in order to effectively control the degree of mutation, avoid the algorithm from falling into local optimality and enhance local search capability, this embodiment preferably dynamically adjusts the scaling factor based on the iteration progress: in the early stage of iteration, the scaling factors of different individuals are obtained by random value sampling to avoid the algorithm from falling into local optimality; in the later stage, the scaling factors are adjusted according to the number of iterations. As the evolutionary generations increase, when the population gradually converges, the values ​​of the scaling factors F of all individuals are reduced to enhance local search capability. The specific dynamic scaling factor can be expressed as:

[0051]

[0052] Among them, F min ,F max They represent the lower and upper limits of the dynamic scaling factor, respectively; U(·) represents uniform distribution; EF max Indicates the maximum number of iterations, gen is the current number of iterations; F p,gen Represents the dynamic scaling factor of the p-th individual in the current iteration.

[0053] Based on the binomial crossover model and the adaptive crossover probability, a crossover operation is performed on each of the mutant individuals and the original individuals in the initial population to generate multiple test individuals. The adaptive crossover probability is determined based on the principle that the greater the fitness value of the parent generation, the greater the crossover probability. That is, the crossover probability is adaptively adjusted according to the evolutionary process. The parent generation with lower fitness has a greater crossover probability. As the fitness of the parent generation increases, the crossover probability gradually decreases. It can be expressed as:

[0054]

[0055] Among them, Cr0 represents the initial crossover probability; fit c Indicates the fitness value of the c-th parent in the current iteration; Cr c represents the crossover probability of the c-th parent in the current iteration; C represents the number of parents.

[0056] Calculating the fitness value of each of the test individuals according to the preset formation optimization objective function, and comparing the fitness value of each of the test individuals with the fitness value of each of the original individuals, and selecting target individuals to enter the next generation population based on the principle of minimizing the fitness value;

[0057] Repeat the mutation, crossover and selection operations on the next generation population until a preset iteration termination condition is reached to obtain the optimal individual, and use the optimal individual as the formation formation information.

[0058] The communication topology matrix and distance constraint matrix corresponding to the optimal individual obtained through the above method steps can provide the parameters of communication connection and distance constraint for the UAV swarm control system, which can be used for the subsequent control of the UAV swarm formation, facilitating the realization of functions such as obstacle avoidance, collision avoidance and formation maintenance.

[0059] This embodiment adopts a dynamic optimization mechanism for formation formation information based on an improved differential evolution algorithm with the optimization goal of balancing communication delay, formation anti-interference and formation flexibility to obtain the optimal communication topology matrix and optimal distance constraint matrix of the target formation. This can ensure the effectiveness of formation formation while reducing communication delay and improving the anti-interference and flexibility of the UAV formation.

[0060] S14: Divide the formation clearing tasks according to the swarm mission path information to obtain a formation control task sequence, and perform formation control on the swarm of drones based on the formation control task sequence, a preset formation control task optimization model, and the formation formation information. The formation control task sequence includes a formation holding task and a formation scaling task; the corresponding preset formation control task optimization model includes a formation holding task optimization sub-model and a formation scaling task optimization sub-model.

[0061] In this embodiment, the drone sub-control system in the drone swarm formation control system is constructed based on a leader-follower framework, and each drone sub-control system implements operational control of each drone based on a distributed model predictive controller and a drone single dynamics model. The single-machine dynamics model is a model that takes into account factors such as the drone's speed and acceleration in the x-axis, y-axis, and z-axis directions, the air resistance torque coefficient, mass, the virtual control variable for controlling the aircraft's lift, the angle, and the angular velocity. It is used to describe the relationship between the drone's motion state and the control input and can be expressed as:

[0062]

[0063] in, and Represents the speed of the drone in the x-axis, y-axis and z-axis directions respectively; and They represent the acceleration of the UAV in the x-axis, y-axis and z-axis directions respectively; K1, K2, K3, K4, K5 and K6 represent the corresponding air resistance torque coefficients in the six degrees of freedom; M represents the mass of the UAV; u1 represents the virtual control quantity for controlling the lift of the aircraft, θ, φ, Respectively represent the pitch angle, roll angle and yaw angle of the drone, and Represent the angular velocities corresponding to the pitch angle, roll angle, and yaw angle respectively; and Respectively represent the angular acceleration corresponding to the pitch angle, roll angle and yaw angle; u θ ,u φ , They represent the virtual control quantities of the angular torque corresponding to the pitch angle, roll angle and yaw angle respectively; I1, I2, I3 represent the moment of inertia of the quadrotor UAV along the three coordinate axes in the body coordinate system; d represents the distance from the propeller axis of the UAV to the center of mass of the UAV.

[0064] The UAV swarm formation control system in this embodiment is a multi-machine system based on a single-machine dynamics model combined with a leader-follower architecture, which can be expressed as:

[0065] x i (k+1)=Ax i (k)+Bu i (k),i=1,2,…N

[0066] Among them, x i (k) and x i (k+l) represents the state variables of UAV i at time k and time k+l respectively; u i (k) represents the control input of the drone sub-control system of drone i in the multi-machine system; A and B represent the coefficient matrices of the equation respectively; N represents the total number of drones.

[0067] It should be noted that in the entire UAV group formation control process, the Lth UAV is used as the pilot UAV, and the other UAVs are used as follower UAVs to perform formation control and task execution. For the convenience of description, x is used in the following text. L (k),u L (k) represent the state variables and control inputs of the pilot UAV respectively; at the same time, a in the communication topology matrix iL =1 indicates that the follower UAV i can obtain the control information of the lead UAV L through this communication link. This communication connection mode ensures that in the formation, the follower UAV can receive the instructions from the lead UAV in real time, providing a basis for subsequent following and formation maintenance.

[0068] Taking into account the actual process of executing the task of cleaning photovoltaic panels by a swarm of drones, each drone needs to move in a determined formation to get as close as possible to the photovoltaic panel to be cleaned (target point). When the drone arrives at the target point, the drone's coverage area needs to be expanded to perform the cleaning work, and after completing the task, the distance between drones needs to be tightened again to maintain the formation and move to the next target point to efficiently complete the application requirements of the cleaning task. This embodiment preferably divides the task of cleaning photovoltaic panels by a drone formation into two parts: a formation maintenance task and a formation scaling task. Specifically, the formation cleaning task is divided according to the swarm task path information to obtain a formation control task sequence, and the steps of performing formation control on the drone swarm according to the formation control task sequence, the preset formation control task optimization model and the formation formation information include:

[0069] According to the swarm task path information, the formation cleaning tasks are divided based on the principle of executing the formation maintenance task during the drone swarm behavior and executing the formation scaling task when the drone swarm starts or ends photovoltaic panel cleaning, and the formation control task sequence is obtained; that is, it is necessary to judge whether the positions of each node in the swarm task path in the swarm task path information belong to the path starting point, path intermediate node and path end point. If the path node is the path starting point, it is necessary to execute the formation scaling task (the formation needs to be contracted); if the path node is the path intermediate node, it is necessary to execute the formation maintenance task; if the path node is the path end point, it is necessary to execute the formation scaling task (the formation needs to be enlarged); in the process of the drone swarm executing the formation photovoltaic panel cleaning task, it is necessary to judge the node type of the current position node in real time (whether it is the path starting point, path intermediate node or path end point), and execute the corresponding type of formation control task according to the corresponding node type.

[0070] When executing the formation maintaining task, the formation control of each drone in the drone group is performed based on the drone sub-control system according to the corresponding formation maintaining task optimization sub-model and the formation formation information; wherein, the formation maintaining task optimization sub-model can be understood as taking into account that the failure of the follower drone to effectively follow the pilot drone will cause formation chaos, making it impossible to efficiently reach the cleaning area, poor formation maintenance may cause some photovoltaic panels to be unable to be cleaned or the cleaning effect is poor, and collision with obstacles will cause damage to the drone, which will increase the cleaning cost and time, etc., by simultaneously considering the actual needs of the follower drone following the pilot drone, the drone maintaining the formation, and the obstacle avoidance effect in the process, the formation maintaining problem optimization model for a single drone based on the state information of the neighboring drones is preferably designed. Specifically, the steps of constructing the formation maintaining task optimization sub-model include:

[0071] Based on a comprehensive analysis of formation error, obstacle safety distance deviation, following error and control increment, the formation keeping task cost function is obtained; wherein, the formation error can be understood as the offset between each UAV (to evaluate the formation keeping effect), the obstacle safety distance deviation can be understood as the situation where the UAV is away from the obstacle (to drive the UAV away from the obstacle), the following error can be understood as the offset between the UAV and the pilot UAV (to evaluate the following effect of the pilot UAV), and the control increment can be understood as the change in the control amount input by the UAV sub-control system; specifically, the formation keeping task cost function is expressed as:

[0072] J 1,i (l)=ψ i ·e i,formation (l)+υ i ·e i,safe (l)+ξ i ·e i,follow (l)+ζi ctrl_inc i (l)

[0073] Where,

[0074]

[0075] e i,follow (l)=||x i (k+l|k)-x L (k+l|k)-b i-L || 2

[0076] ctrl_inc i (l)=||u i (k+l|k)|| 2

[0077] Among them, J i (l) represents the cost function value of the formation-keeping task of UAV i at the lth step, i = 1, 2, ... N, N represents the total number of UAVs, l = 1, 2, ... N1, N1 represents the total number of steps; e i,formation (l), e i,safe (l), e i,follow (l) and ctrl_inc i (l) respectively represent the formation error, obstacle safety distance deviation, following error and control increment of UAV i at the lth step time; ||·|| 2 represents the two-norm; ||h i (k+l|k)-Ω n || 2 represents the distance between UAV i and the obstacle center; h i (k+l|k) represents the location information of UAV i at time k+l; Ω n is the coordinate information of the nth obstacle; O represents the total number of obstacles; d safe Indicates the safety distance threshold between the drone and the obstacle; x i (k+l|k) and x L (k+l|k) represents the position status information of UAV i and pilot UAV L at time k+l respectively; represents the estimated position state information of UAV j at time k+l; b represents the position state offset information of UAV i and UAV j at time k+l; i-j represents the expected posture state offset information between UAV i and UAV j; x i (k+l|k)-x L (k+l|k) represents the position state offset information between UAV i and the pilot UAV L; bi-L represents the expected posture state offset information between UAV i and the pilot UAV L; u i (k+l|k) represents the control input of the drone sub-control system corresponding to drone i; ψ i ,υ i ,ξ i ,ζ i They represent the weight coefficients of formation error, obstacle safety distance deviation, following error and control increment respectively, and can be set and adjusted according to actual application requirements.

[0078] Based on the state input range, the control increment range and the formation formation constraint between UAVs, the formation maintenance task optimization constraint is obtained; the formation formation constraint between UAVs is obtained based on the formation formation information; specifically, the formation maintenance task optimization constraint is expressed as:

[0079]

[0080] Where R i and R j Represents the maximum radius of UAV i and UAV j respectively, which can be based on the k in the distance constraint matrix in the formation formation information. ij set up; represents the estimated position of UAV j at time k+l, l=1,2,…N1, N1 represents the total number of steps; X min and X max They represent the lower and upper limits of the UAV’s attitude state respectively; ΔU min and ΔU max Respectively represent the lower and upper limits of the control increment.

[0081] The formation keeping task optimization sub-model is obtained according to the formation keeping task cost function and the formation keeping task optimization constraint.

[0082] The present embodiment provides a method for calculating the following error by using the posture offset between the UAV and the pilot UAV, evaluating the formation error by using the posture offset between each UAV, and evaluating the obstacle avoidance effect by using the distance between the UAV and the obstacle, while taking into account the control increment situation to design a formation keeping task optimization sub-model to optimize the formation keeping task. This can prompt the follower UAV to closely follow the pilot UAV, maintain the consistency of the overall movement direction of the formation, ensure the stability of the formation shape and the flight safety of the UAV, and thus achieve the stability of the UAV in approaching the photovoltaic panels to be cleaned while ensuring the stability of the formation and avoiding obstacles, effectively improving the execution efficiency and reliability of the cleaning task.

[0083] Based on the above formation maintenance task optimization sub-model, it can be seen that in actual formation maintenance, the operation control of a single UAV needs to consider the operation information of the other UAVs at the same time. Therefore, when optimizing the control of the following UAV i at a certain moment, it is necessary to use the operation information of UAV i at the previous moment and the control input of the pilot UAV to estimate the information of the other UAVs in order to coordinate the formation and prevent collisions. For example, at the kth moment, UAV i uses the trajectory information of UAV j, but at the kth moment, the actual information of UAV j cannot be obtained, so it is necessary to estimate the state of UAV j. First, the control input is estimated.

[0084]

[0085] in, represents the predicted value of the actual control input of UAV j at time k-1 for the future time k+l; u L (k+N1-1) represents the control input of the pilot UAV L at time k+N1-1; x j (k+N1-1|k-1) represents the state of UAV j at time k+N1-1; W j represents the gain matrix.

[0086] At the same time, if the following drone cannot obtain the actual operation information of the pilot drone in time due to various external factors (such as communication delay, signal interference, etc.) during the actual control process, the control input of the pilot drone must also be estimated. The control input estimation formula for the pilot drone at time k+1 is: in represents the actual control input prediction value of the pilot UAV L at time k for the future k+1 time; u L (k) is the control input of the pilot UAV L at time k, x i (k) represents the state of UAV i at time k, and K1 and K2 are gain matrices. Through this estimation formula, each UAV can calculate the estimated value of the control input of the pilot UAV at the next moment based on its own current state and the control input of the pilot UAV at the previous moment, thereby obtaining the control information of the pilot UAV, and then coordinating the formation and preventing collisions, thereby achieving formation maintenance and scaling tasks.

[0087] After the control inputs of the adjacent follower drones and the pilot drone are estimated through the above method steps, the corresponding state estimation values ​​can be obtained based on the obtained control inputs and the single-machine dynamics model. Expressed as:

[0088]

[0089] in, Represents the actual state value at time k-1 for the future time k+1; Indicates the use of The state estimate obtained by solving for it is used as the control input.

[0090] When executing the formation scaling task, the formation control of each drone in the drone group is performed based on the drone sub-control system according to the corresponding formation scaling task optimization sub-model and the formation formation information; wherein, the formation scaling task optimization sub-model can be understood as taking into account that in the photovoltaic panel cleaning task, when the drone formation reaches the target photovoltaic panel area or leaves the cleaned photovoltaic panel area, it is necessary to adjust the formation size according to the distribution of the photovoltaic panels. If the scaling trajectory cannot be accurately tracked, it will lead to unreasonable cleaning coverage, affecting the cleaning effect, and the change in the distance between drones during the scaling process will make it easy for collision accidents to occur between drones. The formation scaling problem optimization model for a single drone based on the state information of neighboring drones is preferably designed based on the consideration of both the scaling trajectory tracking error and the collision avoidance problem between drones. Specifically, the steps of constructing the formation scaling task optimization sub-model include:

[0091] Based on the comprehensive analysis of the safety distance deviation between UAVs, the UAV trajectory deviation and the control increment, the formation scaling task cost function is obtained; the safety distance deviation between UAVs is obtained based on the formation formation information; the formation scaling task cost function is expressed as:

[0092]

[0093] Where,

[0094]

[0095] ctrl_inc i (l)=||u i (k+l|k)|| 2

[0096] Among them, J 2,i (l) represents the cost function value of the formation scaling task of UAV i at the lth step time, i = 1, 2, ... N, N represents the total number of UAVs, l = 1, 2, ... N1, N1 represents the total number of steps; e i,space (l), e i,trace (l) and ctrl_inc i (l) represent the inter-UAV safety distance deviation, UAV trajectory deviation and control increment of UAV i at the lth step time respectively; h i (k+l|k) represents the location information of UAV i at time k+l; represents the estimated position information of UAV j at time k+l; represents the position distance between UAV i and UAV j; x i (k+l|k) and They represent the posture state information of UAV i at time k+l and the expected trajectory posture state information respectively; represents the deviation between the posture state information of UAV i and the expected trajectory; R i and R j Represents the maximum radius of UAV i and UAV j respectively, which can be based on the k in the distance constraint matrix in the formation formation information. ij Setting; η represents the set expansion safety distance; and The weight coefficients for the safety distance deviation between drones, drone trajectory deviation, and control increment are respectively represented and can be set and adjusted according to actual application requirements;

[0097] Based on the state input range, control increment range and obstacle avoidance constraints, the optimization constraints of the formation scaling task are obtained. Specifically, the optimization constraints of the formation scaling task are expressed as:

[0098]

[0099] Where, d safe Indicates the safety distance threshold between the drone and the obstacle; X min and X max They represent the lower and upper limits of the UAV’s attitude state respectively; ΔU min and ΔU max Respectively represent the lower limit and upper limit of the control increment; Ω n is the coordinate information of the nth obstacle, n = 1, 2, ..., O, and O represents the total number of obstacles.

[0100] The formation scaling task optimization sub-model is obtained according to the formation scaling task cost function and the formation scaling task optimization constraint.

[0101] This embodiment considers both scaling trajectory tracking and collision avoidance between drones when optimizing the formation scaling task. In terms of scaling trajectory tracking, the error between the drone and the desired trajectory is calculated to guide the drone to move according to the preset scaling strategy, ensuring that each drone can accurately reach the designated position when the formation expands or reduces the combined area. In terms of collision avoidance between drones, the position distance between each drone is used to evaluate the collision avoidance risk, combined with the set expansion safety distance, to avoid collisions between drones during the formation scaling process. This not only improves the trajectory tracking accuracy during the formation scaling control process, making it possible to more accurately control the formation coverage area and improve the comprehensiveness and uniformity of photovoltaic panel cleaning, but also reduces the incidence of collision accidents between drones, reduces drone damage and mission interruptions caused by collisions, and further improves the overall efficiency and economic benefits of the cleaning task.

[0102] It should be noted that the subsystem optimization models of the formation maintenance and formation scaling tasks designed for the UAV photovoltaic panel cleaning task can use the distributed model predictive controller designed based on the distributed MPC (Model Predictive Control) control algorithm to solve the optimal control input during the execution of the corresponding task, so as to effectively achieve the formation maintenance and scaling trajectory tracking of the UAV swarm, obstacle avoidance between UAVs and obstacles, and collision avoidance between UAVs; among them, when using the distributed model predictive controller to solve the optimal control input during the execution of the corresponding task, the formation maneuver information of the UAV swarm at each moment in the swarm task path information can be used as a more important optimization target for control optimization. In order to facilitate the understanding of the scheme for formation maintenance and formation scaling control of the single-machine dynamics model of each UAV based on the distributed model predictive controller in the actual photovoltaic panel cleaning scenario, the following is combined Figure 4 A detailed description of the fleet control during the photovoltaic panel cleaning task in a large photovoltaic power station where photovoltaic panels are widely distributed and there are some obstacles (such as temporary equipment on the maintenance channel):

[0103] 1) Formation maintenance control: The primary task is to keep the drones in a stable formation, get as close as possible to the photovoltaic panels to be cleaned, and avoid obstacles along the way.

[0104] In practical applications, the reference trajectory of the pilot UAV can be determined according to the photovoltaic panel cleaning task planning (swarm task path information). At the same time, each follower UAV i obtains its own position information and speed information, as well as the position information and speed information of the pilot UAV, surrounding obstacles and other follower UAVs through sensors and communication systems, and calculates the various cost values ​​in the corresponding formation maintenance task cost function. Then, based on the corresponding formation maintenance task optimization constraints, the distributed MPC control algorithm is used to optimize and solve in the prediction time domain and the control time domain to obtain the optimal control input at time k that minimizes the formation maintenance task cost function. The distributed MPC control algorithm predicts the system state at multiple moments in the future and optimizes the control input based on the prediction results to achieve the best control effect.

[0105] To obtain the optimal control input After that, it can be input into the single-machine dynamics model of UAV i, and the optimal control input Calculate the acceleration of drone i Then get the speed and position x i (k+1)=x i (k)+v i(k+1)Δt (Δt is the time step), thereby controlling the movement of the UAVs to fly towards the target photovoltaic panel while keeping the formation stable and avoiding obstacles.

[0106] 2) Formation zoom control: When the UAV formation approaches the target photovoltaic panel area, the formation zoom operation needs to be initiated to expand the cleaning coverage area. Or when the UAV formation is about to leave the photovoltaic panel area after cleaning, the formation zoom operation needs to be initiated to shrink the formation to facilitate the maintenance of the formation in subsequent flights.

[0107] In practical applications, it is necessary to determine the formation scaling strategy in advance according to the distribution of photovoltaic panels, such as uniform expansion or expansion according to a certain geometric shape, and obtain the expected trajectory information of each drone based on the group mission path information. And the real-time position distance r with other drones ij (k); then, based on the cost values ​​of each item in the formation scaling task cost function; and based on the corresponding formation scaling task optimization constraints, the distributed MPC control algorithm is used to optimize and solve in the prediction time domain and the control time domain to obtain the optimal control input at time k that minimizes the formation scaling task cost function. Then the control input This information is input into the drone's individual dynamics model, which calculates the drone's acceleration, velocity, and position, controlling the drone's zooming motion along the desired trajectory while avoiding collisions with other drones. During the zooming process, the position and relative distance of each drone are continuously updated, and the calculation and control process is repeated until the formation zooming mission is complete. After the cleanup mission is complete, the distance between drones may need to be tightened again, and the new formation may need to be maintained to reach the next target point, at which point the control process returns to the formation maintenance phase.

[0108] The embodiment of the present invention obtains the task execution information of the drone group according to the preset formation transformation cycle, obtains the target formation of the drone group according to the task execution information and the pre-built deep learning model, switches the formation of the drone group to the target formation, obtains the group task path information within the current formation transformation cycle based on the preset path planning algorithm according to the distribution information of the photovoltaic panels to be cleaned, and then forms an optimization objective function based on the target formation and a preset formation constructed with the optimization objectives of balancing communication delay, formation anti-interference and formation flexibility, obtains formation formation information including a communication topology matrix and a distance constraint matrix based on the improved differential evolution algorithm, and divides the formation cleaning tasks according to the group task path information to obtain a formation control task sequence. , and a technical solution for formation control of the drone swarm is proposed according to the formation control task sequence, the preset formation control task optimization model and the formation formation information, realizing a formation adaptive transformation strategy based on deep learning, combined with a dynamic optimization mechanism of formation formation information that takes balancing communication delay, formation anti-interference and formation flexibility as optimization goals and is solved by an improved differential evolution algorithm, and a formation maintenance and formation scaling task coordination control mechanism that considers drone safety to reliably control the photovoltaic panel cleaning drone swarm formation, which can not only adaptively change the drone swarm's formation according to environmental changes and task requirements, but also ensure the anti-interference and flexibility of the swarm formation, thereby effectively improving the photovoltaic panel cleaning efficiency and the drone swarm's operational safety, and has high practical value.

[0109] It should be noted that although the steps in the above flowchart are shown in sequence as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders.

[0110] In one embodiment, Figure 5 As shown, a formation control system for a photovoltaic panel cleaning drone group is provided, the system comprising:

[0111] Target formation acquisition module 1, used to obtain task execution information of the drone group according to a preset formation change cycle, and obtain the target formation of the drone group based on the task execution information and a pre-built deep learning model;

[0112] Path information acquisition module 2 is used to switch the formation of the drone group to the target formation, and then obtain the group task path information within the current formation change cycle based on the preset path planning algorithm according to the distribution information of the photovoltaic panels to be cleaned;

[0113] Formation information acquisition module 3 is used to obtain formation formation information based on the improved differential evolution algorithm according to the target formation formation and the preset formation formation optimization objective function; the formation formation information includes a communication topology matrix and a distance constraint matrix; the preset formation formation optimization objective function is constructed with a balance between communication delay, formation anti-interference and formation flexibility as the optimization goal;

[0114] The formation control module 4 is used to divide the formation clearing tasks according to the group task path information, obtain a formation control task sequence, and perform formation control on the drone group according to the formation control task sequence, a preset formation control task optimization model and the formation formation information.

[0115] Regarding the specific definition of the formation control system of the photovoltaic panel cleaning drone swarm, please refer to the definition of the formation control method of the photovoltaic panel cleaning drone swarm above. The corresponding technical effects can also be obtained equivalently, so they will not be repeated here. The various modules in the above-mentioned formation control system of the photovoltaic panel cleaning drone swarm can be implemented in whole or in part through software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0116] Figure 6 FIG. 1 shows an internal structure diagram of a computer device in one embodiment, which may be a terminal or a server. Figure 6 As shown, the computer device includes a processor, memory, a network interface, a display, a camera, and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When executed by the processor, the computer program can implement a formation control method for a swarm of photovoltaic panel cleaning drones. The display screen of the computer device can be a liquid crystal display or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or it can be buttons, a trackball, or a touchpad provided on the computer device housing, or it can be an external keyboard, touchpad, or mouse.

[0117] It can be understood by those skilled in the art that Figure 6The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computing device may include more or fewer components than shown in the figure, or combine certain components, or have the same component arrangement.

[0118] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0119] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are implemented.

[0120] In summary, the embodiments of the present invention provide a formation control method, system, equipment and medium for a swarm of photovoltaic panel cleaning drones. The formation control method for a swarm of photovoltaic panel cleaning drones can realize a formation adaptive transformation strategy based on deep learning, combined with a dynamic optimization mechanism for formation formation information that takes balancing communication delay, formation anti-interference and formation flexibility as optimization goals and is solved by an improved differential evolution algorithm, and a formation maintenance and formation scaling task coordination control mechanism that considers drone safety to reliably control the formation of a swarm of photovoltaic panel cleaning drones. It can not only adaptively change the formation of the drone swarm according to environmental changes and task requirements, but also ensure the anti-interference and flexibility of the swarm formation, thereby effectively improving the efficiency of photovoltaic panel cleaning and the safety of drone swarm operation.

[0121] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the various technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0122] The above-described embodiments merely represent several preferred implementations of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art can make several improvements and substitutions without departing from the technical principles of the present invention, and such improvements and substitutions should also be considered within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be based on the scope of protection of the claims.

Claims

1. A formation control method for a photovoltaic panel cleaning drone swarm, characterized in that: The method comprises the following steps: Obtaining mission execution information of the drone swarm according to a preset formation change cycle, and obtaining a target formation of the drone swarm based on the mission execution information and a pre-built deep learning model; After the formation of the drone group is switched to the target formation, the group's task path information within the current formation change cycle is obtained based on a preset path planning algorithm according to the distribution information of the photovoltaic panels to be cleaned; According to the target formation and the preset formation formation optimization objective function, formation formation information is obtained based on the improved differential evolution algorithm; the formation formation information includes a communication topology matrix and a distance constraint matrix; the preset formation formation optimization objective function is constructed with balancing communication delay, formation anti-interference and formation flexibility as the optimization goal; The formation clearing tasks are divided according to the group task path information to obtain a formation control task sequence, and the formation control of the drone group is performed according to the formation control task sequence, a preset formation control task optimization model and the formation formation information.

2. The formation control method for a photovoltaic panel cleaning drone group according to claim 1, characterized in that: The mission execution information includes current fleet position information, target position information and obstacle relative position information; The step of obtaining the target formation of the drone swarm based on the task execution information and the pre-built deep learning model includes: Traversing various formations in a preset formation library, and inputting various formations and the task execution information into the deep learning model to predict the energy consumption of task execution, and obtaining corresponding formation energy consumption prediction values; The minimum energy consumption prediction value among the formation energy consumption prediction values ​​corresponding to the various formation formations is obtained, and the formation formation corresponding to the minimum energy consumption prediction value is used as the target formation formation of the drone group.

3. The formation control method for a photovoltaic panel cleaning drone group according to claim 1, characterized in that: The preset formation optimization objective function is expressed as: J=w1·J delay +w2·J robust +w3·J flex Where, Among them, J delay represents the total communication delay between all connected drones; t ij (k ij ) represents the communication delay between UAV i and UAV j; a ij k represents the connection relationship variable between drone i and drone j; ij represents the distance between UAV i and UAV j; J robust represents the impact of external interference on the UAV formation; e q (a ij ,k ij ) means in a ij and k ij Under the constraint, the deviation of the relative expected position between UAV i and UAV j during the qth simulated disturbance; Q represents the total number of simulated disturbances; J flex Indicates the difficulty of changing the formation topology to adapt to the mission scenario; c ij represents the cost of changing the connection relationship and distance constraint between UAV i and UAV j; and They represent the connection relationship between UAV i and UAV j before and after the formation topology structure changes; N represents the total number of UAVs in the UAV swarm; w1, w2 and w3 represent weight coefficients; J represents the formation formation optimization goal.

4. The formation control method for a photovoltaic panel cleaning drone group according to claim 1, characterized in that: The step of obtaining formation formation information based on the improved differential evolution algorithm according to the target formation shape and the preset formation formation optimization objective function includes: According to the target formation, a set of initial solutions of the communication topology matrix and the distance constraint matrix are randomly generated as original individuals to construct an initial population; Calculating the fitness value of each of the original individuals in the initial population based on the preset formation optimization objective function; Randomly selecting a number of individuals to be mutated from the initial population, and performing mutation operations on each of the individuals to be mutated based on a dynamic scaling factor to generate corresponding mutated individuals; the dynamic scaling factor is dynamically adjusted based on the iteration progress; Based on a binomial crossover model and an adaptive crossover probability, a crossover operation is performed on each of the mutated individuals and the original individuals in the initial population to generate a plurality of test individuals; the adaptive crossover probability is determined based on the principle that the greater the parent fitness value, the greater the crossover probability; Calculating the fitness value of each of the test individuals according to the preset formation optimization objective function, and comparing the fitness value of each of the test individuals with the fitness value of each of the original individuals, and selecting target individuals to enter the next generation population based on the principle of minimizing the fitness value; Repeat the mutation, crossover and selection operations on the next generation population until a preset iteration termination condition is reached to obtain the optimal individual, and use the optimal individual as the formation formation information.

5. The formation control method for a photovoltaic panel cleaning drone group according to claim 1, characterized in that: The formation control task sequence includes a formation maintaining task and a formation scaling task; the drone sub-control system of the drone swarm is constructed based on a leader-follower framework; The steps of dividing the formation clearing tasks according to the group task path information to obtain a formation control task sequence, and performing formation control on the drone group according to the formation control task sequence, a preset formation control task optimization model and the formation formation information include: According to the swarm task path information, the formation cleaning tasks are divided based on the principle of executing the formation maintaining task during the movement of the drone swarm and executing the formation scaling task when the drone swarm starts or ends photovoltaic panel cleaning, thereby obtaining the formation control task sequence; When executing the formation maintaining task, performing formation control on each drone in the drone group based on the drone sub-control system according to the corresponding formation maintaining task optimization sub-model and the formation forming information; When executing the formation scaling task, the sub-model and the formation formation information are optimized according to the corresponding formation scaling task, and formation control is performed on each drone in the drone group based on the drone sub-control system.

6. The formation control method for a photovoltaic panel cleaning drone group according to claim 5, characterized in that: The steps of constructing the formation keeping task optimization sub-model include: Based on a comprehensive analysis of formation error, obstacle safety distance deviation, following error, and control increment, the formation keeping task cost function is obtained; the formation keeping task cost function is expressed as: J 1，i (l)=ψ i and i，formation (l)+υ i ·and i，safe (l)+ξ i ·and i，follow (l)+ζ i ·ctrl_inc i (l) Among them, J 1,i (l) represents the cost function value of the formation-keeping task of UAV i at the lth step, i = 1, 2, ... N, N represents the total number of UAVs, l = 1, 2, ... N1, N1 represents the total number of steps; e i,formation (l), e i,safe (l), e i,follow (l) and ctrl_inc i (l) represent the formation error, obstacle safety distance deviation, following error and control increment of UAV i at the lth step time; ψ i ,υ i ,ξ i ,ζ i represents the weight coefficient; Based on the state input range, the control increment range and the formation formation constraint between the UAVs, a formation maintenance task optimization constraint is obtained; the formation formation constraint between the UAVs is obtained based on the formation formation information; The formation keeping task optimization sub-model is obtained according to the formation keeping task cost function and the formation keeping task optimization constraint.

7. The formation control method for a photovoltaic panel cleaning drone group according to claim 5, characterized in that: The steps of constructing the formation scaling task optimization sub-model include: Based on the comprehensive analysis of the safety distance deviation between UAVs, the UAV trajectory deviation and the control increment, the formation scaling task cost function is obtained; the safety distance deviation between UAVs is obtained based on the formation formation information; the formation scaling task cost function is expressed as: Among them, J 2,i (l) represents the cost function value of the formation scaling task of UAV i at the lth step time, i = 1, 2, ... N, N represents the total number of UAVs, l = 1, 2, ... N1, N1 represents the total number of steps; e i,space (l), e i,trace (l) and ctrl_inc i (l) represent the inter-UAV safety distance deviation, UAV trajectory deviation and control increment of UAV i at the lth step time respectively; and represents the weight coefficient; Based on the state input range, control increment range and obstacle avoidance constraints, the optimization constraints of the formation scaling task are obtained; The formation scaling task optimization sub-model is obtained according to the formation scaling task cost function and the formation scaling task optimization constraint.

8. A formation control system for a photovoltaic panel cleaning drone group, characterized in that: The system comprises: A target formation acquisition module is used to obtain task execution information of the drone swarm according to a preset formation change cycle, and obtain the target formation of the drone swarm based on the task execution information and a pre-built deep learning model; A path information acquisition module is used to obtain the group mission path information within the current formation change cycle based on the preset path planning algorithm according to the distribution information of the photovoltaic panels to be cleaned after the formation of the group of drones is switched to the target formation; a formation information acquisition module, configured to obtain formation formation information based on the target formation shape and a preset formation formation optimization objective function, based on an improved differential evolution algorithm; the formation formation information includes a communication topology matrix and a distance constraint matrix; the preset formation formation optimization objective function is constructed with a balance between communication delay, formation anti-interference and formation flexibility as the optimization goal; The formation control module is used to divide the formation clearing tasks according to the group task path information, obtain a formation control task sequence, and perform formation control on the drone group according to the formation control task sequence, a preset formation control task optimization model and the formation formation information.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images