Multi-unmanned aerial vehicle nest cooperative site selection method for forest patrol

Abstract

The invention discloses a multi-unmanned aerial vehicle nest cooperative site selection method for forest patrol. The method comprises the steps of obtaining geographic information data and performing data preprocessing; the method comprises the following steps: extracting main ridge lines of a forest region, comprehensively considering three factors of geographical environment, infrastructure and safety guarantee, and preliminarily pre-selecting a plurality of unmanned aerial vehicle nest candidate points; comprehensively considering the electric quantity, the coverage rate, the anti-interference performance and the terrain influence of the unmanned aerial vehicle, constructing a multi-target optimization site selection model, and determining a target function and a constraint condition of the multi-target optimization site selection model; an improved NSGA-II algorithm is adopted to solve the multi-objective optimization site selection model, and an optimal site selection layout scheme of the unmanned aerial vehicle nest is obtained; a Pareto optimal solution set is calculated based on the fuzzy set theory, optimal indexes of all solutions are calculated and sorted, and the solution with the maximum optimal value is a compromise solution of the multi-target problem. According to the method, the cooperative operation efficiency of the unmanned aerial vehicle group is effectively improved, the nest construction and operation cost is remarkably reduced, and the use efficiency and the resource allocation capability of the nest are improved.

Description

Technical Field

[0001] The present invention belongs to the field of planning and site selection of drone nests, and in particular relates to a collaborative site selection method for multiple drone nests for forest inspection. Background Art

[0002] As an important part of the earth's ecosystem, forests carry many key functions, such as conserving water resources, maintaining water and soil, regulating climate, and maintaining biodiversity. However, with the increase of human activities and the impact of global climate change, forests are facing increasingly serious threats such as fires, pests and diseases, and illegal logging. These problems not only damage forest resources, but also destroy the ecological balance. Therefore, it is particularly important to conduct forest inspections and monitoring in a timely and effective manner.

[0003] Traditional forest inspection methods mainly rely on manual ground inspections and aerial monitoring by fixed-wing aircraft. Although these methods can monitor forests to a certain extent, they have obvious limitations, such as limited inspection range, large impact of terrain, high cost and difficulty in real-time monitoring. Especially in complex terrain conditions, the effect of traditional methods is often unsatisfactory.

[0004] In recent years, with the development of drone technology and its wide application in various fields, drones have become one of the ideal tools for forest inspections due to their rapid deployment, high flexibility, relatively low cost, and ability to achieve real-time data transmission. Especially in mountainous forest environments, drones can overcome terrain obstacles, efficiently perform large-area inspection tasks, and promptly detect and respond to problems such as forest fires, pests and diseases, and illegal activities. Therefore, the use of multiple drones to work together to carry out automated forest inspections is gradually becoming a development trend to replace traditional inspection methods.

[0005] The efficient operation of the drone inspection system is inseparable from the reasonable deployment of the nest. The nest is not only the key hub for drone take-off and landing, but also the main facility for its charging and storage. The reasonable choice of the nest location directly affects the inspection efficiency and the overall performance of the system. Although domestic and foreign scholars have conducted a lot of research in the field of drone nest layout optimization and have made certain progress, current research is mostly focused on urban environments and less on complex forest terrain. The existing site selection optimization model has limited processing capabilities under multi-objective and multi-constraint conditions, especially in dealing with the impact of terrain on drone flight performance. In addition, existing methods often operate each facility independently and do not fully consider the coordination of multiple machines, resulting in unsatisfactory application results in mountainous forest environments.

[0006] Therefore, the problem of site selection optimization for drone nests in forest areas still needs further in-depth research and improvement to ensure the efficient operation of drone inspection systems in complex natural environments. Summary of the invention

[0007] Purpose of the invention: The present invention proposes a collaborative site selection method for multiple UAV nests for forest inspection, which effectively improves the collaborative operation efficiency of UAV swarms, significantly reduces the construction and operation costs of nests, and improves the utilization efficiency and resource allocation capabilities of nests.

[0008] Technical solution: The method for collaborative site selection of multiple drone nests for forest inspection described in the present invention comprises the following steps:

[0009] (1) Obtain geographic information data and perform data preprocessing;

[0010] (2) Extract the main ridgelines in the forest area, comprehensively consider the three factors of geographical environment, infrastructure and security, and preliminarily pre-select multiple candidate drone nests;

[0011] (3) Comprehensively consider the UAV power, coverage, anti-interference and terrain influence, build a multi-objective optimization site selection model, and determine its objective function and constraints;

[0012] (4) The improved NSGA-II algorithm is used to solve the multi-objective optimization site selection model and obtain the optimal site layout plan for the drone nest;

[0013] (5) Based on the fuzzy set theory, the Pareto optimal solution set obtained in step (4) is calculated, and the optimization index of each solution is calculated and ranked. The solution with the largest optimization value is the compromise solution to the multi-objective problem.

[0014] Furthermore, the geographic information data described in step (1) includes high-precision DEM data, annual meteorological information, power supply network, and signal coverage network.

[0015] Furthermore, the data preprocessing process described in step (1) is as follows:

[0016] Calculate the Z-Score and IQR of the DEM data elevation value to measure the degree of deviation between the data and the average value of the sample set. IQR is the difference between the upper quartile and the lower quartile. The Z-Score calculation formula is as follows:

[0017]

[0018] Where h is a single observed elevation value, μ is the mean of the sample elevation, and σ is the standard deviation of the sample elevation;

[0019] For small outliers and missing values, bilinear interpolation is used, which replaces or estimates the value based on the four surrounding known pixel values:

[0020]

[0021] Wherein, (lon, lat) represents the horizontal and vertical coordinates of the pixel point, and h(lon, lat) represents the height value of the pixel point (lon, lat); the small range of abnormal values ​​and missing values ​​are data where the number of pixels affected is less than 5% of the total number of pixels;

[0022] For large-scale outliers and missing values, use drones to take aerial photos of the area, and then use GIS software to generate new high-precision DEM data to replace the original erroneous data;

[0023] The latitude and longitude coordinates of the forest area boundary are extracted from the open street map data to form an initial boundary point set P = {P1, P2, ..., P n}, where P i =(lon i ,lat i ), use the system function to find the convex hull point set C = {C1, C2, ..., C m}, where C i =(lon i ,lat i ), and there is a point set relationship The regular convex polygon area formed by the convex hull point set is used as the forest area, and the original DEM data is clipped accordingly; the minimum circumscribed rectangle of the convex hull area is calculated to obtain the coordinates of its four vertices

[0024] The forest area is gridded, and the side length of the square grid is defined as L according to the accuracy requirements of the inspection task to ensure that each grid can reasonably represent the geographical area within the scope; the DEM data is divided into a series of square grids with a side length of L, and the elevation values ​​in each grid are aggregated to represent the elevation value of the grid.

[0025] Furthermore, the process of extracting the main ridgeline of the forest area in step (2) is as follows:

[0026] The key grid features are defined as the starting and ending points of the opposite terrain trends and flat terrain; the terrain grid is cross-sectionally scanned in multiple directions to identify grids that meet the key grid features; the cross-sectional scan is performed along the main direction, and the diagonal direction is considered to capture more details; all grid points that meet the key grid features are marked and extracted; the Kruskal algorithm is used to calculate the minimum spanning tree for the extracted key grid point set to connect all key points while ensuring that the total weight of the tree is minimized; the branches in the minimum spanning tree are analyzed, the branches that do not meet the main ridgeline features are removed, and the main path is retained; the point set MST of the main ridgeline processed by the minimum spanning tree is retained as MST = {M1, M2, … M t}, where M i =(lon i,lat i ).

[0027] Furthermore, the implementation process of preliminarily selecting multiple candidate drone nest points in step (2) is as follows:

[0028] Geographic environmental factors process digital elevation data to obtain the height and slope information of each point to evaluate the suitability of the terrain; at the same time, analyze the local wind direction and wind speed throughout the year, and simulate the surface wind speed to ensure that the climate conditions in the site selection area are suitable for drone flights;

[0029] Infrastructure factors obtain power supply data to ensure that the site selection area is within the power supply reachable area to meet the energy needs of the machine nest; at the same time, obtain signal coverage data, consider the signal coverage area, and ensure unimpeded communication between the drone and the control center;

[0030] Safety factors: Obtain no-fly zone data to avoid setting up aircraft nests in no-fly zones, ensure flight safety and comply with relevant regulations.

[0031] Furthermore, the objective function of the multi-objective optimization site selection model in step (3) includes:

[0032] The total number of actual machine nests is the smallest:

[0033]

[0034] In the formula, x i Indicates the use of the candidate point of the machine nest, which is a 0-1 variable, 1 means building a machine nest at the candidate point i, and 0 means not building a machine nest at the candidate point i;

[0035] The selected machine nest has the largest total coverage of the terrain:

[0036]

[0037] In the formula, Cover i Indicates the coverage of the terrain by candidate nest i.

[0038] Furthermore, the constraints of the multi-objective optimization site selection model in step (3) include:

[0039] The overlap ratio between any two selected nests is less than β:

[0040]

[0041] In the formula, O ij represents the overlap ratio between candidate nests i and j;

[0042] The proportion of any selected nest area blocked by significant terrain is less than γ:

[0043]

[0044] Where D k (2) Indicates the proportion of the candidate machine nest area blocked by the terrain line;

[0045] The effective coverage area ratio of any selected nest is greater than

[0046]

[0047] Where D k (3) Indicates the ratio of the valid area of ​​the candidate machine nest;

[0048] Each selected nest has at least one other selected nest in its coverage area:

[0049]

[0050] The drone starts from any point in the terrain to ηd max There is at least one machine nest within the range:

[0051]

[0052] Where, d max is the UAV's cruising range, I{d kp <η·d max} is the indicator function, when d kp ≤η·d max The value is 1 when , otherwise it is 0, and p is any point in the terrain.

[0053] Furthermore, the implementation process of step (4) is as follows:

[0054] (41) Population initialization: Each candidate nest point has only two states: selected to build a nest or not selected to build a nest, and binary coding is used; each nest selection scheme corresponds to an individual in the population, and each gene value of the individual chromosome corresponds to the state of the candidate nest; the chromosome length is set to the number of candidate nest points, that is, N; the chromosome genotype X k =(x1,x2,…,x N ), x i The state corresponding to the i-th candidate point of the machine nest; the population size is M, and the population is initialized by random generation. Each individual represents a machine nest location scheme;

[0055] (42) Fitness evaluation: The fitness evaluation index is set based on the two objective functions of maximizing coverage and minimizing the number of nests. A penalty term is added to the fitness function. Penalties are accumulated when the coverage overlap between nests exceeds the set threshold. Penalties are increased for nests blocked by terrain. Penalties are accumulated for nests with low effective coverage. Each nest must contain at least one spare nest within its coverage range, otherwise penalties are accumulated:

[0056] P=P overlap +P blocked +P effective +P redundancy (10)

[0057]

[0058]

[0059] Where P is the sum of penalty terms, P overlap , P blocked , P effective , P redundancy They are overlap rate penalty, terrain blocking penalty, effective coverage penalty, and redundant coverage penalty; I{overlap i,j >β} is an indicator function, which takes 1 when the overlap ratio of nests i and j is greater than β, and takes 0 otherwise; I{blocked i >γ} means that the value is 1 when the area of ​​the machine nest blocked by the terrain is greater than γ, otherwise it is 0; Indicates that the effective coverage rate of the machine nest is less than When it is 1, otherwise it is 0; I{nearbycount i <1} indicates that if there are no other nests within the coverage of the nest, the value is 1, otherwise it is 0; F1 is the fitness value of the target maximum coverage rate, and F2 is the fitness value of the target minimum number of nests; the larger the F1 value and the smaller the F2 value, the better the site selection plan;

[0060] (43) Environmental selection: First, the individuals are sorted by non-domination, and the individuals with the lowest non-domination level are selected as the priority objects; when the number of optimal solution sets exceeds the preset external archive capacity, selection is made based on the crowding distance of the individuals to ensure diversity; in the process of selecting the solution set, individuals with a larger crowding distance will be retained first to ensure that the solution set is evenly distributed on the Pareto frontier and avoid excessive concentration in local areas;

[0061] Individual i has a certain objective function f j Crowding distance:

[0062]

[0063] In the formula, F j(i) represents the value of the jth fitness function of the ith individual;

[0064] The total crowding distance of individual i is:

[0065]

[0066] (44) Two-point crossover and binary bit mutation: The two-point crossover method is used, which randomly selects two crossover points of the parent chromosome and exchanges the gene fragments between the two points; the crossover operation is followed by a mutation operation, which uses a binary bit mutation operation; that is, one or more bits in the individual chromosome are randomly selected for mutation, changing gene 0 to 1, or gene 1 to 0;

[0067] (45) Update the population: Select the newly generated offspring together with the parent individuals to construct the next generation population; first, perform non-dominated sorting on all individuals to determine the dominance level of each individual; for individuals within the same dominance level, use the elite retention tournament selection method. In the same dominance level, individuals with greater diversity are retained through tournament selection to ensure the diversity of the solution set, while giving priority to retaining individuals with higher non-dominated levels;

[0068] (46) Termination of iteration: The algorithm terminates when the preset maximum number of iterations is reached or when the Pareto optimal solution set no longer changes significantly within several consecutive generations. The final output solution set is the Pareto optimal solution set for decision makers to choose the optimal solution.

[0069] Furthermore, the implementation process of step (5) is as follows:

[0070] Calculate the optimal index u(i) of each solution and sort them. The solution with the largest optimal value is the compromise solution to the multi-objective problem:

[0071]

[0072] u(i)=u1(i)+u2(i)(21)

[0073] In the formula, u j (i) is the optimal value of the jth objective function value of the ith solution, f j (i) is the objective function value of the jth objective function value of the ith solution, f jmin and f jmax is the minimum and maximum of the j-th objective function values.

[0074] Beneficial effects: Compared with the prior art, the present invention has the following beneficial effects:

[0075] 1. The present invention not only considers the coverage rate of drone nests, but also takes into account factors such as the number of nests, overlap rate, and terrain occlusion, and establishes a multi-objective site selection optimization model; by minimizing the number of nests and maximizing the coverage rate, it achieves maximum resource utilization and improved inspection efficiency; at the same time, the method also considers the impact of terrain occlusion on drone flight, and by analyzing the obstruction of the ridge line to the nest coverage area, it avoids unnecessary resource waste and improves the inspection effect;

[0076] 2. The present invention adopts an improved NSGA-II algorithm to solve the multi-objective optimization model. Compared with the traditional NSGA-II algorithm, the initial hierarchy is constructed based on the candidate points screened by the site selection criteria, and a penalty mechanism is introduced to increase the penalty term for the area blocked by the terrain and reduce its fitness value, thereby effectively improving the convergence speed and solution accuracy of the algorithm; the improved algorithm can better handle multi-objective optimization problems and avoid the emergence of local optimal solutions;

[0077] 3. Based on fuzzy set theory, the present invention ranks the various solutions in the Pareto optimal solution set according to the preferred indicators, and selects the solution with the largest preferred value as the final compromise solution; considering the decision maker's preferences and actual needs, the machine nest layout solution that is more in line with the actual situation is selected, thereby improving the scientificity and rationality of the decision. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0079] The present invention is further described in detail below with reference to the accompanying drawings.

[0080] like Figure 1 As shown, the present invention proposes a multi-UAV nest collaborative site selection method for forest inspection, which specifically includes the following steps:

[0081] Step 1: Geographic information data acquisition and preprocessing

[0082] Query geographic information data according to the latitude and longitude of the target area, including high-precision DEM data, annual weather information, power supply network, and signal coverage network. Integrate and import into the GIS system.

[0083] Identification of outliers and missing values: Calculate the Z-Score and IQR of the DEM data elevation value to measure the degree of deviation between the data and the average value of the sample set. IQR is the difference between the upper quartile (Q3) and the lower quartile (Q1). The Z-Score calculation formula is as follows:

[0084]

[0085] Where: h is a single observed elevation value, μ is the sample elevation mean, σ is the sample elevation standard deviation, if the elevation of a point satisfies |Z|>3 or is lower than Q1-1.5×IQR or higher than Q3+1.5×IQR, then the value is marked as an outlier. Identify NaN values ​​and -9999 in the data and mark them as missing values.

[0086] Outlier and missing value processing: For small-scale outliers and missing values ​​(the number of affected pixels is less than 5% of the total number of pixels), the bilinear interpolation method is used, that is, the value is replaced or estimated based on the surrounding four known pixel values:

[0087]

[0088] Where: (lon, lat) represents the horizontal and vertical coordinates of the pixel point, and h(lon, lat) represents the height value of the pixel point (lon, lat).

[0089] For large-scale outliers and missing values ​​(the number of pixels affected is no less than 5% of the total number of pixels), drones are used to take aerial photos within the range, and then GIS software is used to generate new high-precision DEM data and replace the original erroneous data.

[0090] Convex hull formation: The Overpass plugin in QGIS is used to extract the longitude and latitude coordinates of the forest boundary from the OpenStreetMap (OSM) data to form an initial boundary point set P = {P1, P2, ..., P n}, where P i =(lon i ,lat i ), use the system function to find the convex hull point set C = {C1, C2, ..., C m}, where C i =(lon i ,lat i ), and there is a point set relationship The regular convex polygon area formed by the convex hull point set is used as the forest area, and the original DEM data is clipped accordingly. The minimum bounding box of the convex hull area is calculated to obtain the coordinates of its four vertices.

[0091] Regional gridding: Grid the forest area, define the side length of the square grid as L according to the accuracy requirements of the inspection task, and ensure that each grid can reasonably represent the geographical area within the range. Divide the DEM data into a series of square grids with a side length of L. Aggregate the elevation values ​​in each grid (take the maximum value) to represent the elevation value of the grid.

[0092] Step 2: Extract the main ridgelines in the forest area and conduct comprehensive analysis to determine the pre-selected location of the drone nest.

[0093] Extract the main ridgeline in the forest area: define the characteristics of the key grid as a) the opposite terrain trend before and after (from rising to falling); b) the starting point and end point of the flat terrain. Perform cross-sectional scans on the terrain grid in multiple directions to identify grids that meet the key grid characteristics. The cross-sectional scan is performed along the main direction (from south to north, from east to west), and the diagonal direction is considered to capture more details. Mark and extract all grid points that meet the key grid characteristics. Use the Kruskal algorithm to calculate the minimum spanning tree (MST) for the extracted key grid point set to connect all key points while ensuring that the total weight of the tree is minimized. Analyze the branches in the minimum spanning tree, remove those branches that do not meet the characteristics of the main ridgeline, and retain the main path. The point set MST that retains the main ridgeline after the minimum spanning tree processing is retained as {M1, M2,…M t}, where M i =(lon i ,lat i ).

[0094] Taking into account accessibility, geographical environment, communication conditions, and safety assurance, multiple candidate sites for drone nests are preliminarily pre-selected. Considering the impact of factors such as terrain, vegetation, and climate on drone flight and nest installation, selecting a high and open area will help improve communication quality and flight safety; ensuring that the site selection area has the necessary infrastructure support, such as power supply and communication network coverage, which are the basis for the normal operation of the drone nest; conducting a cost-benefit analysis among multiple candidate locations, and selecting a location with relatively low construction and maintenance costs and good coverage.

[0095] Read and preprocess the acquired terrain digital elevation data (DEM) to obtain the height and slope information of each point. Perform statistical analysis on the local wind direction and wind speed throughout the year, and simulate the surface wind speed of the terrain. Based on the principle of machine nest site selection, comprehensively study the power supply map information, signal coverage map information, no-fly zone information, and terrain data information, and preliminarily screen the set of candidate machine nest points. The requirements that the location of the machine nest must meet are as follows: power supply accessible area, signal coverage area, no-fly restriction area, flat ground, and smooth airflow area.

[0096] Step 3: Comprehensively consider the UAV power, coverage, anti-interference and terrain influence, and establish a multi-objective optimization site selection model with the minimum total number of drone nests and the maximum total coverage of the terrain by the drone nests.

[0097] The first objective function in the model indicates that the total number of actual machine nests to be built is minimized, as shown in the following formula:

[0098]

[0099] Where: x i Indicates the usage of the candidate point of the machine nest, where 1 means building a machine nest at candidate point i, and 0 means not building a machine nest at candidate point i.

[0100] The second objective function indicates that the total coverage of the terrain by the selected machine nest is maximized, as shown in the following formula:

[0101]

[0102] Where: Cover i Indicates the coverage of the terrain by candidate nest i.

[0103] The multi-objective optimization site selection model must meet the following constraints:

[0104] 1) The overlap rate between any two selected nests is less than β:

[0105]

[0106] Where: O ij Represents the overlap rate between candidate nests i and j. Reduce the overlap of monitoring areas between different nests, optimize resource allocation and utilization efficiency, avoid unnecessary repeated coverage, thereby improving inspection efficiency and reducing operating costs.

[0107] 2) The proportion of any selected nest blocked by significant terrain (main ridgeline) is less than γ:

[0108]

[0109] Where: D k (2) Indicates the proportion of the area of ​​the candidate machine nest that is blocked by the terrain line. When the drone crosses the ridge, it will consume more power because it needs to climb quickly in a short time, thereby reducing its total inspection endurance time; the wind speed is high at the ridge, and the flight stability of the drone will be affected in a high wind speed environment. Strong winds will not only make it more difficult for the drone to fly, but may also cause the flight trajectory to deviate and the vibration to increase, thereby affecting the quality of image acquisition and data transmission. At the same time, flying under high wind speeds will also cause greater hardware damage to the drone, increasing the cost of maintenance and replacement. If the drone frequently encounters ridge terrain that requires high climbs and strong winds during the inspection process, some tasks may not be completed as planned, affecting the reliability and stability of the overall inspection task.

[0110] 3) The effective coverage area ratio of any selected nest is greater than

[0111]

[0112] Where: D k (3) Indicates the proportion of the effective area of ​​the candidate nest (the part of the nest coverage circle that does not exceed the target area). When selecting the nest location, give priority to locations that can cover more effective inspection areas to avoid unnecessary coverage expansion to non-inspection areas, ensure that each nest can operate efficiently within its coverage area, reduce resource waste caused by covering non-effective areas, and improve the efficiency of the entire system and the quality of inspection task completion. This will optimize the overall deployment and planning of the nest.

[0113] 4) Each selected nest has at least one other selected nest in its coverage area:

[0114]

[0115] If a nest fails, the other two standby nests can quickly receive the affected drones and take over the tasks of the failed nest. This redundant design ensures that the drone inspection mission is not interrupted by the failure of a single nest, enhances the system's anti-interference ability and robustness, and improves the overall reliability of the system.

[0116] 5) The drone starts from any point in the terrain to ηd max There is at least one machine nest within the range:

[0117]

[0118] Where: d max is the UAV's cruising range, I{d kp ≤ηd max} is the indicator function, when d kp ≤η·d max The value is 1 when the drone is in the state of failure, otherwise it is 0; p is any point in the terrain. This ensures that when an accident occurs during the inspection, the drone can quickly and safely fly to a nest before the battery runs out, achieve safe landing and charging, increase the continuous working capacity of the system, and avoid inspection interruption and drone loss due to insufficient power.

[0119] Step 5: Use the improved NSGA-II algorithm to solve the site selection model and obtain the optimal site layout plan for the drone nest.

[0120] Population initialization: In the nest location optimization problem, each candidate nest point has only two states, namely, selected to build a nest and not selected to build a nest. Each candidate nest point has only two states, using binary coding. Each nest selection scheme corresponds to an individual in the population, and each gene value of the individual chromosome corresponds to the state of the candidate nest. The chromosome length is set to the number of candidate nest points, that is, N. Chromosome genotype Xk =(x1,x2,…,x N ), where x i The state of the i-th candidate nest point corresponds to the population size M, which is initialized by random generation, and each individual represents a nest location scheme.

[0121] Fitness evaluation: Fitness evaluation is performed on each individual. The fitness function is calculated based on the Pareto dominance relationship. That is, whether an individual is dominated by other individuals on all targets at the same time. If an individual is not inferior to other individuals on all targets and is dominant on at least one target, it is called the Pareto optimal solution. Based on this dominance relationship, the individuals in the population are stratified according to non-dominated sorting. Each layer contains individuals that do not dominate each other. The present invention sets the fitness evaluation index based on two objective functions: 1) the number of covered terrain points, that is, maximizing the coverage rate; 2) the number of selected machine nests, that is, minimizing the number of machine nests. The machine nest site selection scheme is comprehensively evaluated through these two objective functions. In addition, in order to improve the rationality of site selection and cope with complex forest environments, the following penalty items are added to the fitness function. Accumulate penalties for situations where the coverage overlap rate between machine nests exceeds the set threshold to avoid resource waste; increase penalties for machine nests blocked by terrain to ensure that the inspection route is not affected by terrain occlusion; each machine nest must contain at least one spare machine nest within its coverage range, otherwise accumulate penalties to enhance the system's fault tolerance and redundancy. The fitness function modified by the comprehensive penalty term is used to comprehensively evaluate the machine nest site selection plan.

[0122] P=P overlap +P blocked +P effective +P redundancy (10)

[0123]

[0124]

[0125] Where: P is the sum of penalty terms, P overlap , P blocked , P effective , P redundancy They are overlap penalty, terrain blocking penalty, effective coverage penalty, and redundant coverage penalty. i,j >β} is an indicator function, which takes 1 when the overlap ratio of nests i and j is greater than β, and takes 0 otherwise. i >γ} means that the value is 1 when the area of ​​the machine nest blocked by the terrain is greater than γ, otherwise it is 0. Indicates that the effective coverage rate of the machine nest is less than If yes, it takes 1, otherwise it takes 0. i<1} indicates that if there are no other nests within the coverage of the nest, the value is 1, otherwise it is 0. F1 is the fitness value of target 1), and F2 is the fitness value of target 2). The larger the F1 value and the smaller the F2 value, the better the site selection plan.

[0126] Environmental selection: retain the Pareto optimal solution set from the current population. This step first sorts the individuals by non-dominance and selects the individuals with the lowest non-dominance level as the priority to be retained. When the number of optimal solution sets exceeds the preset external archive capacity, selection is made based on the crowding distance of the individuals to ensure diversity. In the process of selecting the solution set, individuals with a larger crowding distance will be retained first to ensure that the solution set is evenly distributed on the Pareto frontier and avoid excessive concentration in local areas.

[0127] Individual i has a certain objective function f j Crowding distance:

[0128]

[0129] Where: D j (i) represents the value of the jth fitness function of the ith individual.

[0130] The total crowding distance of individual i is:

[0131]

[0132] Two-point crossover and binary bit mutation: Through the crossover operation in the genetic algorithm, the genetic information of the parent individuals is recombined to generate new individuals. This paper adopts the two-point crossover method, which randomly selects two crossover points of the parent chromosome and exchanges the gene fragments between the two points. The crossover probability is set to 0.9. After the crossover operation, a mutation operation will be performed, using a binary bit mutation operation. That is, one or more bits in the individual chromosome are randomly selected for mutation, changing gene 0 to 1, or changing gene 1 to 0. The mutation probability is set to 0.1, and mutation generates new offspring individuals.

[0133] Update the population: Select the newly generated offspring together with the parent individuals to construct the next generation population. First, sort all individuals by non-dominance to determine the dominance level of each individual. For individuals in the same dominance level, the elite retention tournament selection method (etour) is used. In the same dominance level, individuals with greater diversity are retained through tournament selection to ensure the diversity of the solution set, while giving priority to retaining individuals with higher non-dominance levels.

[0134] Termination of iteration: Two termination conditions are used: the algorithm terminates when the preset maximum number of iterations is reached; the algorithm terminates when the Pareto optimal solution set no longer changes significantly in several consecutive generations. The final output solution set is the Pareto optimal solution set for decision makers to choose the optimal solution.

[0135] Step 6: Calculate the compromise solution through fuzzy set theory to provide decision support. For the Pareto optimal solution set solved in step 5, calculate the optimal index u of each solution i And sort them, the solution with the largest optimal value is the compromise solution to the multi-objective problem:

[0136]

[0137] u(i)=u1(i)+u2(i)(21)

[0138] Where: u j (i) is the optimal value of the jth objective function value of the ith solution, f j (i) is the objective function value of the jth objective function value of the ith solution, f jmin and f jmax is the minimum and maximum of the j-th objective function values.

[0139] For example, taking the typical mountainous terrain of Jiangsu Province, China as an example, the site selection plan for the machine nest is designed. FCER and TOPRANK are used as evaluation indicators, and the calculation process is as follows:

[0140]

[0141]

[0142] Among them, Cover i N represents the number of demand points covered by site selection scheme i in the terrain, i represents the corresponding required number of machine nests, and T represents the total number of demand points in the terrain. ij A represents the value of the jth attribute of the i-th solution. + and A - represent positive ideal solutions and negative ideal solutions respectively.

[0143] Table 1 is a comparison of the experimental results of the method of the present invention and the mainstream site selection problem methods. It can be seen from Table 1 that the site selection scheme of the present invention is more cost-effective and more economical and efficient than the site selection schemes obtained by the greedy algorithm and the genetic algorithm. The method proposed by the present invention requires no more machine nests than other algorithms, and the coverage rate is higher than other algorithms. The FCER of the present invention is higher than that of other algorithms, indicating that the scheme provided by the present invention can achieve a higher coverage rate with less facility investment, thereby providing a higher cost-effectiveness. The TOPRANK of the present invention is lower than that of other algorithms, indicating that the scheme provided by the present invention can maintain a high coverage efficiency while approaching the ideal solution.

[0144] Table 1 Comparison of results of different methods

[0145]

[0146] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for collaborative site selection of multiple drone nests for forest inspection, characterized in that: The following steps are involved: (1) Obtain geographic information data and perform data preprocessing; (2) Extract the main ridgelines in the forest area, comprehensively consider the three factors of geographical environment, infrastructure and security, and preliminarily pre-select multiple candidate drone nests; (3) Comprehensively consider the UAV power, coverage, anti-interference and terrain influence, build a multi-objective optimization site selection model, and determine its objective function and constraints; (4) The improved NSGA-II algorithm is used to solve the multi-objective optimization site selection model and obtain the optimal site layout plan for the drone nest; (5) Based on the fuzzy set theory, the Pareto optimal solution set obtained in step (4) is calculated, and the optimization index of each solution is calculated and ranked. The solution with the largest optimization value is the compromise solution to the multi-objective problem.

2. The method for collaborative site selection of multiple drone nests for forest inspection according to claim 1 is characterized in that: The geographic information data described in step (1) includes high-precision DEM data, annual meteorological information, power supply network, and signal coverage network.

3. The method for collaborative site selection of multiple drone nests for forest inspection according to claim 1 is characterized in that: The data preprocessing process described in step (1) is as follows: Calculate the Z-Score and IQR of the DEM data elevation value to measure the degree of deviation between the data and the average value of the sample set. IQR is the difference between the upper quartile and the lower quartile. The Z-Score calculation formula is as follows: In the formula, h is a single observed elevation value, μ is the mean of sample elevation, and σ is the standard deviation of sample elevation; For small outliers and missing values, bilinear interpolation is used, which replaces or estimates the value based on the four surrounding known pixel values: In the formula, (lon, lat) represents the horizontal and vertical coordinates of the pixel point, and h(lon, lat) represents the height value of the pixel point (lon, lat); For large-scale outliers and missing values, use drones to take aerial photos of the area, and then use GIS software to generate new high-precision DEM data to replace the original erroneous data; The latitude and longitude coordinates of the forest area boundary are extracted from the open street map data to form an initial boundary point set P = {P1, P2, ..., P n }, where P i =(lon i ,lat i ), use the system function to find the convex hull point set C = {C1, C2, ..., C m }, where C i =(lon i ,lat i ), and there is a point set relationship The regular convex polygon area formed by the convex hull point set is used as the forest area, and the original DEM data is clipped accordingly; the minimum circumscribed rectangle of the convex hull area is calculated to obtain the coordinates of its four vertices The forest area is gridded, and the side length of the square grid is defined as L according to the accuracy requirements of the inspection task, ensuring that each grid can reasonably represent the geographical area within the scope; The DEM data is divided into a series of square grids with a side length of L, and the elevation values ​​in each grid are aggregated to represent the elevation value of the grid.

4. The method for collaborative site selection of multiple drone nests for forest inspection according to claim 1 is characterized in that: The process of extracting the main ridgeline of the forest area in step (2) is as follows: The key grid features are defined as the starting and ending points of the opposite terrain trends and flat terrain; the terrain grid is cross-sectionally scanned in multiple directions to identify grids that meet the key grid features; the cross-sectional scan is performed along the main direction, and the diagonal direction is considered to capture more details; all grid points that meet the key grid features are marked and extracted; the Kruskal algorithm is used to calculate the minimum spanning tree for the extracted key grid point set to connect all key points while ensuring that the total weight of the tree is minimized; the branches in the minimum spanning tree are analyzed, the branches that do not meet the main ridgeline features are removed, and the main path is retained; the point set MST of the main ridgeline processed by the minimum spanning tree is retained as MST = {M1, M2, … M t }, where M i =(lon i ,lat i ).

5. The method for collaborative site selection of multiple drone nests for forest inspection according to claim 1 is characterized in that: The implementation process of preliminarily selecting multiple candidate drone nest points in step (2) is as follows: Geographic environmental factors process digital elevation data to obtain the height and slope information of each point to evaluate the suitability of the terrain; at the same time, analyze the local wind direction and wind speed throughout the year, and simulate the surface wind speed to ensure that the climate conditions in the site selection area are suitable for drone flights; Infrastructure factors obtain power supply data to ensure that the site selection area is within the power supply reachable area to meet the energy needs of the machine nest; at the same time, obtain signal coverage data, consider the signal coverage area, and ensure unimpeded communication between the drone and the control center; Safety factors: Obtain no-fly zone data to avoid setting up aircraft nests in no-fly zones, ensure flight safety and comply with relevant regulations.

6. The method for collaborative site selection of multiple drone nests for forest inspection according to claim 1 is characterized in that: The objective function of the multi-objective optimization site selection model in step (3) includes: The total number of actual machine nests is the smallest: In the formula, x i Indicates the use of the candidate point of the machine nest, which is a 0-1 variable, 1 means building a machine nest at the candidate point i, and 0 means not building a machine nest at the candidate point i; The selected machine nest has the largest total coverage of the terrain: In the formula, Cover i Indicates the coverage of the terrain by candidate nest i.

7. The method for collaborative site selection of multiple drone nests for forest inspection according to claim 1 is characterized in that: The constraints of the multi-objective optimization site selection model in step (3) include: The overlap ratio between any two selected nests is less than β: In the formula, O ij represents the overlap ratio between candidate nests i and j; The proportion of any selected nest area blocked by significant terrain is less than γ: Where D k (2) Indicates the proportion of the candidate machine nest area blocked by the terrain line; The effective coverage area ratio of any selected nest is greater than Where D k (3) Indicates the ratio of the valid area of ​​the candidate machine nest; Each selected nest has at least one other selected nest in its coverage area: The drone starts from any point in the terrain and moves to ηd max There is at least one machine nest within the range: Where, d max is the UAV's cruising range, I{d kp ≤η·d max } is the indicator function, when d kp ≤η·d max The value is 1 when , otherwise it is 0, and p is any point in the terrain.

8. The method for collaborative site selection of multiple drone nests for forest inspection according to claim 1 is characterized in that: The implementation process of step (4) is as follows: (41) Population initialization: Each candidate nest point has only two states: selected to build a nest or not selected to build a nest, and binary coding is used; each nest selection scheme corresponds to an individual in the population, and each gene value of the individual chromosome corresponds to the state of the candidate nest; the chromosome length is set to the number of candidate nest points, that is, N; the chromosome genotype X k =(x1,x2,…,x N ), x i The state corresponding to the i-th candidate point of the machine nest; the population size is M, and the population is initialized by random generation. Each individual represents a machine nest location scheme; (42) Fitness evaluation: The fitness evaluation index is set based on the two objective functions of maximizing coverage and minimizing the number of nests. A penalty term is added to the fitness function. Penalties are accumulated when the coverage overlap between nests exceeds the set threshold. Penalties are increased for nests blocked by terrain. Penalties are accumulated for nests with low effective coverage. Each nest must contain at least one spare nest within its coverage range, otherwise penalties are accumulated: P=P overlap +P blocked +P effective +P redundancy (10) Where P is the sum of penalty terms, P overlap , P blocked , P effective , P redundancy They are overlap rate penalty, terrain blocking penalty, effective coverage penalty, and redundant coverage penalty; I{overlap i,j >β} is an indicator function, which takes 1 when the overlap ratio of nests i and j is greater than β, and takes 0 otherwise; I{blocked i >γ} means that the value is 1 when the area of ​​the machine nest blocked by the terrain is greater than γ, otherwise it is 0; Indicates that the effective coverage rate of the machine nest is less than When it is 1, otherwise it is 0; I{nearbycount i <1} indicates that if there are no other nests within the coverage of the nest, the value is 1, otherwise it is 0; F1 is the fitness value of the target maximum coverage rate, and F2 is the fitness value of the target minimum number of nests; the larger the F1 value and the smaller the F2 value, the better the site selection plan; (43) Environmental selection: First, the individuals are sorted by non-domination, and the individuals with the lowest non-domination level are selected as the priority objects; when the number of optimal solution sets exceeds the preset external archive capacity, selection is made based on the crowding distance of the individuals to ensure diversity; in the process of selecting the solution set, individuals with a larger crowding distance will be retained first to ensure that the solution set is evenly distributed on the Pareto frontier and avoid excessive concentration in local areas; Individual i has a certain objective function f j Crowding distance: In the formula, F j (i) represents the value of the jth fitness function of the ith individual; The total crowding distance of individual i is: (44) Two-point crossover and binary bit mutation: The two-point crossover method is used, which randomly selects two crossover points of the parent chromosome and exchanges the gene fragments between the two points; the crossover operation is followed by a mutation operation, which uses a binary bit mutation operation; that is, one or more bits in the individual chromosome are randomly selected for mutation, changing gene 0 to 1, or gene 1 to 0; (45) Update the population: Select the newly generated offspring together with the parent individuals to construct the next generation population; first, perform non-dominated sorting on all individuals to determine the dominance level of each individual; for individuals within the same dominance level, use the elite retention tournament selection method. In the same dominance level, individuals with greater diversity are retained through tournament selection to ensure the diversity of the solution set, while giving priority to retaining individuals with higher non-dominated levels; (46) Termination of iteration: The algorithm terminates when the preset maximum number of iterations is reached or when the Pareto optimal solution set no longer changes significantly within several consecutive generations. The final output solution set is the Pareto optimal solution set for decision makers to choose the optimal solution.

9. The method for collaborative site selection of multiple drone nests for forest inspection according to claim 1 is characterized in that: The implementation process of step (5) is as follows Calculate the optimal index u(i) of each solution and sort them. The solution with the largest optimal value is the compromise solution to the multi-objective problem: u(i)=u1(i)+u2(i)(21) In the formula, u j (i) is the optimal value of the jth objective function value of the ith solution, f j (i) is the objective function value of the jth objective function value of the ith solution, f jmin and f jmax is the minimum and maximum of the j-th objective function values.

10. The method for collaborative site selection of multiple drone nests for forest inspection according to claim 3 is characterized in that: The small-range outliers and missing values ​​are data where the number of pixels affected is less than 5% of the total number of pixels.

Images