Path planning method for cross-domain scheduling of photovoltaic cleaning trolley by rotor unmanned aerial vehicle

Abstract

The invention discloses a path planning method for cross-domain scheduling of a photovoltaic cleaning trolley by a rotor unmanned aerial vehicle, and belongs to the field of unmanned aerial vehicle application. According to the method, a tournament selection method and a roulette selection method are combined to serve as a selection operator, a segmentation crossover operator is adopted, a traditional genetic algorithm is improved, and the method can meet the requirement that an unmanned aerial vehicle dispatches a cleaning trolley to work in a complex three-dimensional space; the problem that the genetic algorithm is low in convergence speed and prone to falling into local optimum when applied in a two-dimensional mode is solved, the cost loss of the whole power generation area can be minimized, and benefits can be maximized.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle applications, and in particular to a path planning method for a rotary-wing unmanned aerial vehicle to dispatch a photovoltaic cleaning vehicle across domains. Background Art

[0002] In recent years, photovoltaic power generation has been increasingly used. At present, photovoltaic power generation technology has become universal and low-cost, and distributed photovoltaic power generation (such as rooftop photovoltaic power generation technology) has been vigorously promoted and applied. Distributed photovoltaic power generation is different from traditional large-scale photovoltaic power plants in that, in order to improve the utilization rate of solar energy and space utilization efficiency, photovoltaic panels in a certain area are usually distributed in a three-dimensional spatial state (that is, there are obvious differences in high and low positions in space). For the cleaning of photovoltaic panels in this situation, the current solution is mostly still manual cleaning, which is risky and costly, and is not suitable as the best cleaning solution.

[0003] In order to solve the above problems, there are reports on a distributed cleaning method that uses a rotor UAV to dispatch a photovoltaic cleaning vehicle in three dimensions across domains. The use of a rotor UAV to dispatch a photovoltaic cleaning vehicle in three dimensions across domains involves UAV path planning.

[0004] UAV path planning means the process of planning the optimal or suboptimal flight path that meets the requirements from the starting point to the end point of the flight by using relevant algorithms before the UAV takes off, taking into account the flight performance, arrival time, energy consumption and constraints of the surrounding environment of the UAV. It is one of the key technologies of the flight mission planning system and the basic technical guarantee for the UAV to achieve autonomous navigation and autonomous cruising. Therefore, the research on UAV path planning is very important. Specifically: for photovoltaic power generation in a certain area, considering factors such as power generation efficiency and cleaning cost, how to plan the path of the rotor UAV to dispatch the cleaning car can improve the efficiency of the entire power generation area; through reasonable path planning, the cleaning requirements of the distributed photovoltaic power generation area can be met, and the cleaning car can be dispatched across regions in a complex three-dimensional space to the designated photovoltaic panel.

[0005] For the path planning of the 3D cross-domain dispatching photovoltaic cleaning vehicle of the wing UAV, the traditional trajectory planning methods include A* algorithm, sparse A* algorithm, D* algorithm, Dijkstra algorithm and dynamic programming method. These traditional trajectory planning methods are currently difficult to apply to the complex environment of the photovoltaic power generation area with uneven heights. In terms of speeding up the solution time of trajectory planning, the intelligent optimization algorithm has a good performance. Among them, the common intelligent optimization algorithms include genetic algorithm, simulated annealing, particle swarm algorithm, neural network algorithm, etc. Although the intelligent optimization algorithm can quickly generate trajectories in a two-dimensional environment, it converges slowly in a three-dimensional environment.

[0006] It should be noted that the current three-dimensional path planning problems for drones are mostly for military drones. Obstacle avoidance and height restrictions are used as the main constraints in complex three-dimensional spaces. Objects with heights in three-dimensional space are usually regarded as threats and avoided directly. This makes it impossible to dispatch cleaning vehicles in complex spaces with uneven heights.

[0007] At the same time, the current mainstream path planning algorithms are not suitable for path planning in three-dimensional complex spaces, and different algorithms have problems such as being easily trapped in local optimality, long convergence time, premature convergence, and high cost. For example, the method reported in the paper "Photovoltaic panel cleaning and grading task planning based on improved genetic algorithm, Li Cuiming, etc." is also only used for two-dimensional space path planning and is not suitable for three-dimensional space.

[0008] Therefore, there is an urgent need in this field for an improved path planning method for three-dimensional cross-domain scheduling of photovoltaic cleaning vehicles by rotorcraft UAVs. Summary of the invention

[0009] The purpose of the present invention is to provide a path planning method for a rotorcraft UAV to dispatch a photovoltaic cleaning vehicle across domains, so as to solve the above problems.

[0010] In order to achieve the above object, the technical solution adopted by the present invention is as follows: a path planning method for a rotor UAV to dispatch a photovoltaic cleaning vehicle across domains, comprising the following steps: S1: Based on the known environmental information of photovoltaic panel distribution, a spatial three-dimensional model of the photovoltaic panel distribution area is established, and the spatial three-dimensional coordinates of each photovoltaic panel are input; S2: According to the specified starting point information, the drone flies to the specified photovoltaic panel and deploys the cleaning vehicle; S3: The cleaning car performs cleaning work, while the UAV system performs path planning and obtains the next navigation target; S4: After the cleaning car completes the cleaning task of the photovoltaic panel, the drone re-equips the cleaning car to go to the next planned target and deploys the car to work; S5: Repeat S4 until a cleaning cycle ends or the drone stops working after running out of energy. After the energy is replenished, the drone continues to work along the predetermined path until a cleaning cycle ends. S6: The drone recovers the cleaning vehicle and performs maintenance on the entire system; In step S3, the path planning is performed based on a cost function F. ; in, is the total cost of the drone in the vertical direction during a cleaning cycle, is the total flight time of the drone in a cleaning cycle, is the total distance the drone flies in a cleaning cycle, , , They are , , The weight coefficient, k is a natural number, , , The calculation formula is: ; ; ; In the formula, m is the number of different photovoltaic panels, and M is the maximum number of photovoltaic panels in the area. , is the Z-axis coordinate of different photovoltaic panels; is the three-dimensional Euclidean distance between different photovoltaic panels, and v is the flight speed of the drone.

[0011] In order to solve the problem that mainstream path planning algorithms in three-dimensional space are not applicable, the present invention provides a "three-dimensional space-time joint optimization" method, which encodes the position of photovoltaic panels into three-dimensional space coordinates, for example: P1 (x1, y1, z1), P2 (x2, y2, z2)..., directly processes the three-dimensional path, and integrates multiple constraints such as time, space, and energy consumption, with the ultimate goal of minimizing costs, so as to maximize the efficiency of the cleaning task and minimize energy consumption, and finally can efficiently and accurately plan the optimal path in three-dimensional space.

[0012] The problem of maximizing power generation benefits can be simplified to finding the path with the minimum cost function through an improved genetic algorithm, that is, finding After comparing k path planning solutions, the path planning solution with the smallest F is the optimal path solution for the three-dimensional cross-domain dispatching of the photovoltaic cleaning vehicle by the rotor UAV.

[0013] As a preferred technical solution, the weight coefficient adopts a dynamic adjustment strategy, and the specific adjustment formula is: ; ; ; ; in, is the current iteration number, is the maximum number of iterations; using a dynamically adjusted weight mechanism can quickly reduce the path length in the initial stage ( Leading), later optimize flight time and vertical energy consumption ( leading).

[0014] As a further preferred technical solution, during path planning, a hybrid selection operator combining the tournament selection method with the roulette selection method is used, and a method combining adaptive crossover mutation probability and hybrid mutation strategy is used.

[0015] As a further preferred technical solution, the adaptive crossover mutation probability formula is as follows: ; ; in, is the crossover probability, which decreases gradually with the number of iterations; is the mutation probability, which increases gradually with the number of iterations. The crossover operator based on the segmentation rule makes it easier for the algorithm to jump out of the local optimum and enhances the stability of the algorithm; The selection of mutation strategy is determined by experimental testing. The experimental data is shown in Table 1 below: Table 1 Comparison of different mutation strategies , Through experimental comparison, we found that: (1) Inversion mutation can converge quickly but is prone to fall into local optimality (diversity score 0.65) (2) Insertion mutation can increase diversity but converges slowly (convergence algebra 380); (3) The hybrid strategy uses 80% inversion mutation to ensure convergence speed and 20% insertion mutation to maintain population diversity, achieving the optimal balance between convergence speed and solution quality.

[0016] Therefore, this embodiment adopts a hybrid mutation strategy, in which 80% inversion mutation ensures the convergence speed and 20% insertion mutation maintains population diversity, so as to achieve a balance between the convergence speed and the solution quality.

[0017] In order to improve the convergence speed of the algorithm, the present invention proposes an improved genetic algorithm (IGA) to solve this problem. A hybrid selection operator that combines the tournament selection method with the roulette selection method simultaneously uses a method that combines adaptive crossover mutation probability with a hybrid mutation strategy, which significantly improves the convergence speed of the algorithm.

[0018] The above algorithm is preferably designed and implemented using MATLAB R2016b.

[0019] Compared with the prior art, the advantages of the present invention are: (1) For photovoltaic power generation in a certain area, the present invention can solve the current market gap in the automated cleaning mode for distributed photovoltaic panels, reduce cleaning risks and cleaning costs, and at the same time, considering factors such as power generation efficiency and cleaning costs, the present invention can minimize the cost loss and maximize the benefits of the entire power generation area; (2) The present invention can meet the cleaning requirements of distributed photovoltaic power generation areas and can dispatch cleaning vehicles across regions in a complex three-dimensional space to designated photovoltaic panels with minimal loss; (3) The path planning method for the three-dimensional cross-domain dispatching of photovoltaic cleaning vehicles by rotorcraft UAVs based on the improved genetic algorithm of the present invention can not only meet the needs of dispatching cleaning vehicles to work in complex three-dimensional space, but also propose a new idea to solve the problem of slow convergence speed and easy to fall into local optimum when the genetic algorithm is applied in two-dimensional mode; the improved genetic algorithm of the present invention uses a dynamic evolution mechanism, self-adjustment of weights + adaptive mutation strategy, avoids manual parameter tuning, and can significantly improve the convergence speed of the algorithm; (4) The three-dimensional spatiotemporal joint optimization processing method of the present invention enables the algorithm to have a wider range of application scenarios. By directly processing three-dimensional data and integrating multiple constraints such as time, space and energy consumption, the accuracy and efficiency of the algorithm can be improved. The method of the present invention can solve most path planning problems in three-dimensional space and has broad prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 The flowchart of the cross-domain dispatching of photovoltaic cleaning vehicles by the rotary-wing UAV of the present invention; Figure 2 is the coordinate system of each photovoltaic panel in the space of the embodiment of the present invention; Figure 3 This is a flow chart of path planning using an improved genetic algorithm in the present invention; Figure 4 2D graph of scheduling paths for improving genetic algorithm planning; Figure 5 A 3D diagram of the scheduling path planned by the improved genetic algorithm of an embodiment of the invention; Figure 6 A graph showing the variation of F with the number of iterations solved by the improved genetic algorithm of an embodiment of the invention; Figure 7 This is the graph of F solved by the traditional genetic algorithm as the number of iterations changes; Figure 8 This is a graph showing how F of an improved genetic algorithm based on a combination of a tournament selection method and a roulette wheel selection method changes with the number of iterations. DETAILED DESCRIPTION

[0021] The present invention will be further described below in conjunction with embodiments.

[0022] Example:

[0023] See also Figure 1 A path planning method for a rotary-wing UAV to dispatch a photovoltaic cleaning vehicle across domains includes the following steps: S1: Based on the known environmental information of photovoltaic panel distribution, a spatial three-dimensional model of the photovoltaic panel distribution area is established, and the spatial three-dimensional coordinates of each photovoltaic panel are input; S2: According to the specified starting point information, the drone flies to the specified photovoltaic panel and deploys the cleaning vehicle; S3: The cleaning car performs cleaning work, while the UAV system performs path planning and obtains the next navigation target; S4: After the cleaning car completes the cleaning task of the photovoltaic panel, the drone re-equips the cleaning car to go to the next planned target and deploys the car to work; S5: Repeat S4 until a cleaning cycle ends or the drone stops working after running out of energy. After the energy is replenished, the drone continues to work along the predetermined path until a cleaning cycle ends. S6: The drone recovers the cleaning vehicle and performs maintenance on the entire system; It should be noted that the technology of using a rotor drone to dispatch a cleaning vehicle and the obstacle avoidance technology of the drone are both well-known technologies in the art and will not be described in detail here. Since the distribution range of distributed photovoltaic panels is not too wide, that is, the area that needs to be cleaned is generally not too large, the present invention does not consider the impact of environmental factors in different regions on photovoltaic power generation; at the same time, the photovoltaic panels in the same area are cleaned, and the environmental information is regarded as known information (such as the vertical height and horizontal position (spatial coordinates) of each photovoltaic panel distribution, etc.); In this embodiment, the spatial coordinates of each photovoltaic panel are shown in Table 2 below; Table 2 Spatial coordinates of each photovoltaic panel , In this embodiment, in step S1, In the vertical direction, since the UAV performs vertical displacement when dispatching the cleaning vehicle at different heights, it will consume the energy of the entire system. Therefore, the height dimension is added to calculate the vertical energy consumption cost. The formula is as follows: ; Indicates the total height change of the drone in the vertical direction (Z axis), reflecting the lifting energy consumption, where: , is the vertical (Z-axis) coordinate of different photovoltaic panels; Due to the addition of the vertical direction, the construction method of the traditional Hamilton graph is optimized, and the three-dimensional Euclidean distance is used instead of the two-dimensional Manhattan distance. The solar photovoltaic panel cleaning problem is transformed into a traveling salesman problem (TSP). Different photovoltaic panels in the space are set to different spatial coordinates, such as Figure 2 As shown; Figure 2 In the formula, X is the horizontal coordinate of the geometric center of each photovoltaic panel; Y is the vertical coordinate of the geometric center of each photovoltaic panel; and the altitude is the altitude information of each photovoltaic panel, representing the vertical coordinate. The position of each photovoltaic panel is a three-dimensional spatial coordinate, for example: P1 (x1, y1, z1), P2 (x2, y2, z2)... Each point represents the specific position of the photovoltaic panel, and is numbered with a number. Each photovoltaic panel is regarded as a graph node, and the edge weight between nodes is the three-dimensional Euclidean distance to construct a Hamilton graph, using the three-dimensional Euclidean distance to replace the two-dimensional Manhattan distance. The specific formula is as follows; ; in, is the three-dimensional Euclidean distance between different photovoltaic panels, , is the horizontal coordinate of different photovoltaic panels, , is the vertical coordinate of different photovoltaic panels, , are the vertical coordinates of different photovoltaic panels; therefore, the path planning of the cleaning vehicle is to find an optimal Hamilton loop that satisfies the condition that the cost function F of the present invention is minimum; In step S3, the path planning is performed based on a cost function F. ; in is the total cost of the drone in the vertical direction during a cleaning cycle, is the total flight time of the drone in a cleaning cycle, is the total distance that the drone flies in a cleaning cycle, that is, the sum of the three-dimensional Euclidean distances that the drone flies in a cleaning cycle. k is a natural number. The specific formula is: ; ; in is the three-dimensional Euclidean distance between different photovoltaic panels, v is the flight speed of the drone, and in this embodiment, v=5m / s; , , They are , , The weight coefficient of In this embodiment, the weight coefficient adopts a dynamic adjustment strategy, and the specific adjustment formula is: ; ; ; ; in, is the current iteration number, is the maximum number of iterations.

[0024] In order to improve the convergence speed of the algorithm, this embodiment adopts an improved genetic algorithm, see Figure 3 , when planning the path, a hybrid selection operator combining the tournament selection method with the roulette selection method is used, and a method combining the adaptive crossover mutation probability and the hybrid mutation strategy is used; the formula for the adaptive crossover mutation probability is as follows: ; ; in, is the crossover probability, which decreases gradually with the number of iterations; is the mutation probability.

[0025] In this embodiment, the objective function is converted into a fitness function by a dynamic linear calibration method, which can avoid the fitness function difference being too small, thereby weakening the selection function of the genetic algorithm. The conversion formula is: ; ; in, is the fitness function; is the objective function (cost function); is the maximum objective function; For the The selection pressure adjustment value of the individual generation, where b and c are constants, is the baseline scaling factor for dynamically adjusting the fitness to ensure that the fitness value is positive; is the attenuation factor for dynamically adjusting fitness, ,make sure Follow After multiple verifications, this embodiment takes b=600; c=0.99; Combine the tournament selection method and the roulette selection method as the selection operator; Tournament selection is a selection method based on competition. First, some parent individuals are randomly selected to participate in a tournament (i.e. competition), and then the individuals with the best fitness are selected to enter the next generation. The size of the tournament is usually predetermined, and usually selected from a smaller group of individuals. Roulette selection is a selection method based on the proportion of individual fitness. Individuals with higher fitness occupy a larger area in the roulette wheel, so they have a greater probability of being selected; The tournament selection method is simple and efficient but prone to local optimality; the roulette selection method is simple and intuitive but prone to premature convergence and sensitive to fitness value differences. This embodiment adopts a combination of the two methods, and the specific hybrid process is divided into the following two stages: 1. Championship screening: randomly select 10% of individuals from the population and retain the top 20% of the elites in terms of fitness; 2. Roulette selection: Roulette selection is applied only to the elite group to ensure the inheritance of high-quality genes while maintaining diversity.

[0026] This combination helps to introduce some new parent individuals in each generation while maintaining the inheritance of individuals with higher fitness, taking into account the advantages of both selection strategies and maintaining a good balance and stability during the operation of the genetic algorithm; The commonly used sequential crossover (OX) operator may lead to a slower convergence speed in some cases. Therefore, a segmented crossover operator is proposed, which uses two operators, OX and self-crossover. When the required cost is greater than the set threshold Q, sequential crossover is used; when the required cost is less than Q, self-crossover is used; among them, a sensitivity analysis is performed on the threshold Q, and the specific results are shown in Table 3 below: Table 3 Results of sensitivity analysis of threshold Q , It can be seen from Table 3 that when the Q value is 180, the convergence speed is the fastest and the final cost is the lowest, so the Q value is 180 in this embodiment; At the same time, the use of adaptive crossover and mutation probability (the specific crossover and mutation probabilities are determined by the above formula) can avoid manual parameter tuning and improve the convergence speed of the algorithm; According to the above algorithm steps, the algorithm design is implemented using MATLAB R2016b.

[0027] In this embodiment, the population size is designed to be 150, the crossover probability and mutation probability are determined by the above formula, and the maximum number of iterations is 1000 for simulation. The results are as follows Figure 4 and Figure 5 As shown in the figure; the green mark in the figure is the starting point. It can be seen that the algorithm plans the path of the drone-dispatched cleaning vehicle and obtains the cleaning order as follows: ; The cost function of the output of this embodiment is ; The optimal path solved by the improved genetic algorithm in this embodiment changes with the number of iterations as shown in the following figure: Figure 6 As shown; Figure 6 In the equation, F is the current cost. Figure 5 It can be seen that when the algorithm iterates to 250 generations, the optimal path can be obtained.

[0028] It should be noted that the "traditional genetic algorithm" of the present invention refers to the standard genetic algorithm currently known and disclosed in the art. Based on this "standard genetic algorithm", many versions of "improved genetic algorithms" have emerged, but the current "improved genetic algorithms" are not applicable to three-dimensional space. Since the traditional genetic algorithm is not applicable to three-dimensional space, the "three-dimensional spatiotemporal joint optimization" method of the present invention is applied to the traditional genetic algorithm. The algorithm of the present invention is compared with the traditional genetic algorithm under the same parameters to solve the optimal path change with the number of iterations. Among them, the cost function , Fixed weight , , ; Changes such as Figure 7 As shown; Depend on Figure 7 It can be seen that in the same experimental sample, the traditional genetic algorithm fell into a local optimum when iterated to 633 generations, and the optimal cost was 340.8, which was significantly greater than the cost of the algorithm proposed in the present invention, and the convergence speed of the algorithm proposed in the present invention was about 400 generations faster than that of the traditional genetic algorithm. It can be seen that the algorithm proposed in the present invention can improve the convergence speed of the algorithm and obtain the optimal result more efficiently and accurately.

[0029] Compared with the prior art that is closer to the present invention (i.e., the improved genetic algorithm based on the combination of tournament selection method and roulette selection method disclosed in the document “Photovoltaic panel cleaning and grading task planning based on improved genetic algorithm, Li Cuiming et al.”), since the prior art cannot be applied to three-dimensional space and the required cost function is different, the “three-dimensional spatiotemporal joint optimization” method is applied to this algorithm to facilitate comparison and control of a single variable. The parameter setting is consistent with that of this embodiment, and the results obtained are as follows: Figure 8 As shown; Depend on Figure 8It can be seen that in the same experimental sample, the algorithm obtained the lowest cost when iterated to 992 generations, and the optimal low cost was 368.3, which is significantly greater than the cost solved by the algorithm proposed in the present invention. From the trend of the curve, it can be inferred that the algorithm can still iterate to obtain better results, but the number of iterations is greater than 1000 times, which will not be discussed further here. From this, it can be seen that although the lowest cost sought by the improved algorithm may be lower than that of the standard genetic algorithm, the convergence speed of the algorithm proposed in the present invention is significantly improved compared with the improved algorithm, and the lowest cost route can be planned under the same conditions at a faster convergence speed. It can be seen from this that the algorithm proposed in the present invention can improve the convergence speed of the algorithm and obtain the optimal result more efficiently and accurately.

[0030] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A path planning method for a rotor UAV to dispatch a photovoltaic cleaning vehicle across domains, characterized in that: The steps include: S1: Based on the known environmental information of photovoltaic panel distribution, a spatial three-dimensional model of the photovoltaic panel distribution area is established, and the spatial three-dimensional coordinates of each photovoltaic panel are input; S2: According to the specified starting point information, the drone flies to the specified photovoltaic panel and deploys the cleaning vehicle; S3: The cleaning car performs cleaning work, while the UAV system performs path planning and obtains the next navigation target; S4: After the cleaning car completes the cleaning task of the photovoltaic panel, the drone re-equips the cleaning car to go to the next planned target and deploys the car to work; S5: Repeat S4 until a cleaning cycle ends or the drone stops working after running out of energy. After the energy is replenished, the drone continues to work along the predetermined path until a cleaning cycle ends. S6: The drone recovers the cleaning vehicle and performs maintenance on the entire system; In step S3, the path planning is performed based on a cost function F. ; in, is the total cost of the drone in the vertical direction during a cleaning cycle, is the total flight time of the drone in a cleaning cycle, is the total distance the drone flies in a cleaning cycle, , , They are , , The weight coefficient, k is a natural number, , , The calculation formula is: ； ； ； In the formula, m is the number of different photovoltaic panels, and M is the maximum number of photovoltaic panels in the area. , is the Z-axis coordinate of different photovoltaic panels; is the three-dimensional Euclidean distance between different photovoltaic panels, and v is the flight speed of the drone.

2. The path planning method for cross-domain dispatching of photovoltaic cleaning vehicles by rotorcraft UAVs according to claim 1 is characterized in that: The weight coefficient adopts a dynamic adjustment strategy, and the specific adjustment formula is: ； ； ； ； in, is the current iteration number, is the maximum number of iterations.

3. The path planning method for cross-domain dispatching of photovoltaic cleaning vehicles by a rotorcraft UAV according to claim 2 is characterized in that: When planning the path, a hybrid selection operator is used that combines the tournament selection method with the roulette selection method, and a method that combines the adaptive crossover mutation probability with the hybrid mutation strategy is used.

4. The path planning method for cross-domain dispatching of photovoltaic cleaning vehicles by a rotorcraft UAV according to claim 3 is characterized in that: The adaptive crossover mutation probability formula is as follows: ； ； in, is the crossover probability, which decreases gradually with the number of iterations; is the mutation probability, which increases gradually with the number of iterations.

5. The path planning method for cross-domain dispatching of photovoltaic cleaning vehicles by a rotorcraft UAV according to claim 4 is characterized in that: The algorithm design was implemented using MATLAB R2016b.

Images