Fire disaster unmanned aerial vehicle emergency processing path optimization method

Abstract

The invention discloses a fire unmanned aerial vehicle emergency processing path optimization method, and belongs to the field of unmanned aerial vehicle channel control, and the method comprises the steps: building a three-dimensional map, setting a starting point and an ending point of an unmanned aerial vehicle, and obtaining a reference datum line through the starting point and the ending point; adding a fire intensity function formed by the wind speed, the temperature and the smoke concentration on the reference datum line; constructing a constraint function on the reference datum line added with the fire intensity function; key parameter optimization is carried out by using a multi-objective optimization method; and the optimized parameters are substituted into the improved fast random tree model for solving, and a Bezier curve is used for performing path smoothing optimization operation. According to the invention, factors such as the environment when the unmanned aerial vehicle flies are considered when a fire occurs, and the dynamic adaptability and the flight safety of the flight path of the unmanned aerial vehicle are improved; through the improved fast random tree model, efficient, smooth and safe path planning in a complex terrain environment during a fire disaster is realized, and powerful support is provided for an intelligent navigation technology.

Description

Technical Field

[0001] The present invention relates to the field of UAV flight path control, and particularly to an optimized method for the emergency handling path of a fire-fighting UAV. Background Art

[0002] Due to frequent fire accidents, not only are lives threatened, but serious property losses may also be caused. Therefore, it is necessary to complete fire detection, rescue material delivery, and emergency personnel guidance within a very short time. Traditional emergency response methods have problems such as response delays, information lags, and low evacuation efficiency, making it difficult to meet the emergency requirements of fires. Therefore, the use of UAVs for emergency disposal is considered.

[0003] However, the complexity of the environment poses many challenges to UAV path planning. For example, dense buildings result in a large number of no-fly zones, restricting the feasible flight paths of UAVs; significant terrain undulations require UAVs to frequently adjust their flight altitudes to avoid collisions; and crowded people mean that path planning needs to take safety into account to prevent threats to people. Existing path planning methods have deficiencies when applied to fire emergency tasks: insufficient dynamic adaptability: traditional path planning methods do not fully consider the dynamic changes in the spread of the fire, and cannot adjust the UAV path according to real-time fire data, which may lead to the invalidation or inefficiency of the planned path; security defects: existing algorithms do not fully consider the impact of high-temperature areas and smoke on UAV sensors and flight stability, which may cause the path to cross dangerous areas, increasing the risk of mission failure; limited environmental adaptability: most path planning methods do not design optimization strategies for the complex terrain and dense buildings of the environment, making it difficult to effectively avoid obstacles and achieve stable flight in no-fly zones and low-altitude restricted areas; insufficient multi-objective optimization ability: traditional algorithms mainly focus on the optimal path length or time, and do not comprehensively consider factors such as safety, energy consumption, and flight stability, resulting in a reduction in the flight efficiency and mission success rate of UAVs. Therefore, fires are characterized by strong suddenness, rapid spread, scattered occurrence locations, and high emergency response requirements, posing great challenges to UAV path planning. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides an optimized method for the emergency handling path of a fire-fighting UAV.

[0005] To achieve the above technology, the specific steps are as follows:

[0006] S1. Establish a three-dimensional map of the target area through an open-source dataset, and set the starting position and ending position of the UAV in the obtained three-dimensional map according to a preset ratio. Calculate the Euclidean distance between the starting position and the ending position to obtain a reference baseline;

[0007] Establish the eligible area of the digital elevation model (DEM) of the target area through MATLAB; among them, the eligible area is to set the area where the elevation of the target area in the digital elevation model (DEM) is higher than the lowest sea level height to invalid value (NaN) to obtain the eligible area, and the lowest sea level height is set to 1940 meters; the set resolution of the digital elevation model (DEM) is 2 meters;

[0008] Convert the pixel coordinates to actual geographical coordinates through MATLAB, and establish the eligible area with two-dimensional grid and actual geographical coordinates through MATLAB;

[0009] Obtain the elevation values of the grid points in the eligible area of the two-dimensional grid and actual geographical coordinates through data interpolation method;

[0010] Establish a three-dimensional map of the actual eligible area with elevation values, two-dimensional grid and actual geographical coordinates through MATLAB;

[0011] Specify the location and end position of the drone on the three-dimensional map according to the preset ratio, ensure that the starting point and the end point are within the terrain range, and at the same time increase the location and end position by 1 meter to ensure that the marked points are clearly visible on the three-dimensional map; among them, the preset ratio is: the starting point coordinates take 0.40 of the length of the three-dimensional map and the coordinates take 0.50 of the width of the three-dimensional map; the end point coordinates take 0.62 of the length of the three-dimensional map and the coordinates take 0.33 of the width of the three-dimensional map;

[0012] Obtain the Euclidean distance between the starting point and the end point through MATLAB, that is, the reference baseline.

[0013] S2. Add the fire intensity function composed of wind speed, temperature and smoke concentration to the reference baseline for the normal flight of the drone in the fire environment;

[0014] The fire intensity function P(x h ,y h ,z h ) is expressed as follows:

[0015] P(x h ,y h ,z h +Δt) = f(P(x h ,y h ,z h ),V wind ,H temp ,ρ smoke )

[0016] In the formula, V wind represents the wind speed, which is used to affect the flame spread direction; H temp represents the temperature, which is used to affect the fire source intensity; ρsmoke represents the smoke concentration; Δt represents the fire spreading time; where, set H temp ≥80 °C, the temperature is too high, marked as a no-fly zone and needs to be avoided; set ρ smoke >0.7, the smoke concentration is too high, marked as a no-fly zone and needs to be avoided;

[0017] The temperature H temp increases linearly with the real-time fire detection model and the fire spreading time Δt;

[0018] The expression of the real-time fire detection model is as follows:

[0019]

[0020] In the formula, T safe represents the maximum temperature that the UAV can withstand, T safe <80 °C; T(x i' ,y i' ) represents the real-time temperature of the UAV, where (x i' ,y i' ) represents the real-time coordinates of the UAV; N represents the total number of coordinates of the UAV in the target area; i' represents the coordinates of the UAV in the target area;

[0021] S3. Construct a constraint function on the reference baseline of the fire intensity function to obtain the safe flight conditions of the UAV;

[0022] In the constraint function, combine the building height in the target area to construct the UAV height constraint condition, and the expression is as follows:

[0023] Z min (x a ,y a )=H building (x,y)+H safe

[0024] In the formula, Z min (x a ,y a ) represents the minimum flight height of the UAV at the position; H building (x,y) represents the height of the actual geographical coordinates of the building at the position (x,y), which is the height of the obstacle that the UAV needs to avoid; H safe represents that the safe height margin of the UAV is 2 meters; the UAV path must satisfy H safe >Z min (x a ,y a );

[0025] The constraints of the actual environment in the target area also include: the constraints of the crowded area and the constraints of the signal coverage area;

[0026] The constraints in crowded areas are as follows: A crowd heat map is established by collecting crowd density information in crowded areas, where the crowd density information includes: past activity data of the crowd, location and time data of the crowd entering and leaving places, and peak data of the crowd's dining locations; by integrating the crowd heat map into the UAV path planning, the UAV can avoid crowded areas and fly safely;

[0027] Integrating the crowd heat map into the UAV path planning is as follows: By constructing a weighted function with past activity data, class time data, and restaurant peak data, a model representing "cost" is obtained through data normalization; when using a sampling-based algorithm (such as RRT) for random sampling, the sampling probability in high-cost areas is reduced to guide the path to avoid crowded areas;

[0028] S4. Use the multi-objective optimization method (MOGA) to optimize the key parameters of the Rapidly-exploring Random Tree (RRT); the steps are as follows:

[0029] S4.1. Set the shortest path and path smoothness during the search of the Rapidly-exploring Random Tree as the optimization objectives, and input the multi-objective optimization method to optimize the UAV path;

[0030] The path smoothness is used to reduce the flight energy consumption and control difficulty of the UAV;

[0031] The optimization method is as follows:

[0032] Construct a fitness function for the UAV with the shortest path and path smoothness, and the expression is:

[0033] F = w1f1 + w2f2 + w3C f + w4C c

[0034] In the formula, F represents the total optimization model of the UAV; f1(x) represents the shortest path optimization objective; f2(x) represents the path smoothness optimization objective; C f (x,y) represents the fire impact cost; C d (x,y) represents the crowd impact cost; w1 represents the weight of the shortest path optimization objective; w2 represents the weight of the path smoothness optimization objective; w3 represents the weight of the fire impact cost; w4 represents the weight of the crowd impact cost; in the present invention, w1 = 0.35, w2 = 0.25, w3 = 0.30, w4 = 0.10;

[0035] The expression of the shortest path optimization objective f1(x) is as follows:

[0036]

[0037] Wherein, n represents the total number of points on the path; x ii represents the ii-th point on the path;

[0038] The expression of the path smoothness optimization objective f2(x) is as follows:

[0039]

[0040] The fire impact cost C f (x, y) is expressed as follows:

[0041] C f (x, y) = w f *P(x h , y h , z h )

[0042] Wherein, w f represents the learning coefficient of the fire intensity P(x h , y h , z h );

[0043] The crowd impact cost C d (x, y) is expressed as follows:

[0044] C c (x, y) = w c *D dentisty (x, y)

[0045] Wherein, w c represents the learning coefficient of the crowd impact cost D dentisty (x, y);

[0046] S4.2. Extract the key parameters of the multi-objective optimization method optimization process and perform optimization processing on the key parameters through the non-dominated sorting genetic algorithm (NSGA-II) to obtain the optimal result;

[0047] The key parameters include: step size; maximum number of iterations; target sampling rate; target threshold; sampling bias;

[0048] The expression for constructing the key parameters for inputting into the non-dominated sorting genetic algorithm is:

[0049] X = [s, N', γ', τ, β]

[0050] Where s represents the step size, with a range of [1, 10], which is used to affect the growth pace of the tree; represents the maximum number of iterations, with a range of [100, 1000], which is used to determine the computational budget for the search; represents the goal sampling rate, with a range of [0.1, 0.9], which is used to determine the probability of the goal point in the sampling; represents the goal area threshold, with a range of [0.5, 5.0], which is used to judge whether the path reaches the goal point; represents the sampling bias, with a range of [0.0, 1.0], which is used to affect the randomness of the sampling point distribution;

[0051] The non-dominated sorting genetic algorithm (NSGA-II) is used to solve the optimization objective in the optimization process, and the expression is as follows:

[0052] min f1(x), f2(x), C f , C c ;

[0053] Pareto front analysis is adopted to optimize the results, and a set of optimal solutions is obtained; the solution with the smallest total objective is selected as the optimal parameter, and the expression is as follows:

[0054] X * = argmin f1(x);

[0055] The optimized parameters are: step size: 1.34; maximum number of iterations: 100; goal sampling rate: 0.86; goal threshold: 1.57; sampling bias: 0.73.

[0056] S5. Substitute the optimized parameters into the improved rapidly-exploring random tree model to solve the intelligent sampling strategy of the UAV path under safe flight conditions and use the Bezier curve for path smoothing optimization operation; construct a fire source detection and personnel evacuation guidance model for UAV emergency handling through a multi-objective optimization function to complete the path planning of UAV emergency handling for fires;

[0057] The improved rapidly-exploring random tree model is: based on the traditional rapidly-exploring random tree algorithm, an intelligent sampling strategy, path smoothing optimization, and the obtained optimized key parameters are added to improve the feasibility and smoothness of the path;

[0058] The intelligent sampling strategy includes: goal point biasing operation and Gaussian sampling operation;

[0059] The expression of the goal point biasing operation is as follows:

[0060]

[0061] Sampling goal point q goal With probability pg Perform direct sampling or uniform sampling with a probability of 1-(p g +p b ) to obtain the target point-biased sampling result (x rand , y rand ), where p g +p b ≤1;

[0062] The Gaussian sampling operation is as follows: With a probability of p b , perform Gaussian sampling based on the mean point q mean of the current tree. The expression is as follows:

[0063]

[0064] In the formula, Ν”(q mean , Σ) represents the Gaussian distribution (normal distribution) with q mean as the mean and Σ as the covariance matrix;

[0065] The intelligent sampling strategy expression is as follows:

[0066]

[0067] The solution method is: Use nearest neighbor search optimization (accelerated by k-d tree); As the number of coordinates N of the human and machine in the target area increases, the computational cost of brute-force search grows linearly with the data scale, while the cost of nearest neighbor search optimization (accelerated by k-d tree) only grows logarithmically with the data scale, thus bringing the advantage of a significant reduction in computational complexity (from linear to logarithmic), that is, reducing the computational complexity from O(N) to O(logN);

[0068] During the solution process, in order to accelerate the expansion speed in sparse areas and reduce the step size in dense areas, the expression for reducing the step size is:

[0069] η = η max ·e -λd

[0070] In the formula, η represents the reduced step size; η max represents the maximum step size in the sparse area; λ represents the attenuation coefficient; d represents the distance from the current point to other nodes;

[0071] The path smoothing optimization operation is: Use the third-order Bezier curve interpolation method to optimize the solved path. The expression is as follows:

[0072] B(t) = (1 - t) 3 P0 + 3(1 - t) 2 tP1 + 3(1 - t)t 2 P2 + t 3P3, where t ∈ [0, 1]

[0073] In the formula, P0 represents the starting point on the path; P3 represents the ending point on the path; P1 represents the first intermediate control point; P2 represents the second intermediate control point; t represents the normalization parameter; P1 and P2 are taken from the inflection points in the path generated by the improved rapidly-exploring random tree model. To avoid excessive path curvature, it is required that the minimum turning radius R of the UAV min shall not exceed the corresponding maximum allowable curvature, expressed as: K max = 1 / R min ; It is stipulated that R min = 3 m, then the maximum allowable curvature is 0.33; if the maximum allowable curvature exceeds 0.33, then points P1 and P2 need to be adjusted; when the maximum allowable curvature < 0.2, the path is considered smooth;

[0074] The expressions for adjusting points P1 and P2 are as follows:

[0075]

[0076] The constraints are:

[0077] P1 = (1 - α)P0 + αP3, P2 = (1 - β)P0 + β'P3

[0078] In the formula, n' represents the total number of points on the solved path; P iiii represents the i-th point on the path; α represents the smoothness parameter of the first control path; β' represents the smoothness parameter of the second control path; P3 represents the reference point;

[0079] The expression for constructing a fire source detection and personnel evacuation guidance model for UAV emergency response through a multi-objective optimization function is:

[0080] F = α f f fire-detection + β f f evacuation-guide + γ f f shortest-path

[0081] In the formula, f fire-detection represents the data composed of wind speed, temperature, smoke concentration, and flight altitude collected by the UAV for fire; f evacuation-guide represents the safe path for the UAV to guide personnel evacuation; f shortest-path represents the optimal path obtained by the UAV through the improved rapidly-exploring random tree model; α f represents the weight of the data composed of wind speed, temperature, smoke concentration, and flight altitude collected by the UAV for fire; β f represents the weight of the safe path for the UAV to guide personnel evacuation; γ fIndicates the weight of the optimal path obtained by the drone through the improved rapidly-exploring random tree model.

[0082] Advantages of the present invention:

[0083] When a fire occurs, the present invention takes into account factors such as the temperature of the fire, the smoke concentration, and the safe flight altitude during the flight of existing drones, improving the dynamic adaptability and flight safety of the drone flight path;

[0084] After optimizing the parameters of the existing rapidly-exploring random tree model through a multi-objective optimization method, the present invention adds a target point biasing operation and a Gaussian sampling operation to the existing rapidly-exploring random tree model, reducing the sampling calculation complexity and calculation step size of the existing rapidly-exploring random tree model. The flight path of the drone is smoothed and optimized through a Bessel curve, reducing the flight path length of the drone during a fire and improving the flight smoothness; realizing efficient, smooth, and safe path planning in a complex terrain environment during a fire, providing strong support for intelligent navigation technology. Description of the drawings

[0085] Figure 1 Is the step flow chart of the present invention;

[0086] Figure 2 Is the campus digital elevation model diagram of the embodiment of the present invention;

[0087] Figure 3 Is the scene model result diagram of the embodiment of the present invention;

[0088] Figure 4 Is the key part code diagram of the improved rapidly-exploring random tree model of the present invention;

[0089] Figure 5 Is the path planning diagram of the embodiment of the present invention. Detailed implementation manners

[0090] The present invention will be further described in detail below in conjunction with specific embodiments.

[0091] As Figure 1 shown, a method for optimizing the emergency handling path of a fire drone includes the following steps:

[0092] S1. Establish a three-dimensional map of the target area through an open-source dataset, and set the starting position and ending position of the drone in the obtained three-dimensional map according to a preset ratio. Calculate the Euclidean distance between the starting position and the ending position to obtain a reference baseline;

[0093] Establish the eligible area of the digital elevation model (DEM) of the target area through MATLAB; among them, the eligible area is to set the area where the elevation of the target area in the digital elevation model (DEM) is higher than the lowest sea level height to invalid value (NaN) to obtain the eligible area, and the lowest sea level height is set to 1940 meters; the set resolution of the digital elevation model (DEM) is 2 meters;

[0094] Convert the pixel coordinates to actual geographic coordinates through MATLAB, and establish the eligible area with two-dimensional grid and actual geographic coordinates through MATLAB;

[0095] Obtain the elevation values of the grid points in the eligible area of the two-dimensional grid and actual geographic coordinates through data interpolation method;

[0096] Establish a three-dimensional map of the actual eligible area with elevation values, two-dimensional grid and actual geographic coordinates through MATLAB;

[0097] Specify the location and end point positions of the UAV in the three-dimensional map according to the preset ratio, ensure that the starting point and the end point are within the terrain range, and at the same time increase the location and end point positions by 1 meter to ensure that the marked points are clearly visible in the three-dimensional map; among them, the preset ratio is: the starting point x q coordinate takes 0.40 of the length of the three-dimensional map, y q coordinate takes 0.50 of the width of the three-dimensional map; the end point x z coordinate takes 0.62 of the length of the three-dimensional map, y z coordinate takes 0.33 of the width of the three-dimensional map;

[0098] Obtain the Euclidean distance between the starting point and the end point through MATLAB, that is, the reference baseline;

[0099] In this embodiment, a certain university is selected as the target area, and its digital elevation model (DEM) is as Figure 2 shown;

[0100] The specific establishment process of the three-dimensional map of the target area is as follows:

[0101] Use the geotiffread function in MATLAB to read the digital elevation model (DEM) data and geographic reference information of the target area, and convert the read information into double type; among them, the resolution accuracy of the elevation model (DEM) data set is 2 meters;

[0102] Determine the number of rows and columns of the elevation model (DEM) data matrix from the data converted into double type through the size function in MATLAB;

[0103] Set the areas in the digital elevation model (DEM) where the elevation of the target area is higher than the lowest sea level height to invalid values (NaN) to obtain the qualified areas; where the lowest sea level height is 1940 meters;

[0104] Use the intrinsicToWorld function in MATLAB to convert the pixel coordinates of the qualified areas into actual geographical coordinates, and use the meshgrid function in MATLAB to construct a two-dimensional grid;

[0105] The expression for converting pixel coordinates to actual geographical coordinates is as follows:

[0106] x = x min +(col - 1)×Δx

[0107] y = y max -(row - 1)×Δy

[0108] In the formula, x and y represent the horizontal and vertical coordinates in the actual geographical coordinates (world coordinates); x min represents the minimum actual coordinate value of the image in the x direction; y max represents the maximum actual coordinate value of the image in the y direction; col represents the column index of the pixel; row represents the row index of the pixel; Δx represents the resolution in the x direction; Δy represents the resolution in the y direction;

[0109] Use data interpolation method for the actual qualified areas with geographical coordinates after constructing the two-dimensional grid, so that each grid point has a corresponding elevation value;

[0110] The data interpolation method is: use the interp2 function in MATLAB to perform two-dimensional linear interpolation on the DEM data, that is, for a point (x', y') to be interpolated, according to its four known actual geographical coordinates (x1, y1), (x2, y1), (x1, y2), (x2, y2) near its position, the interpolation formula is as follows:

[0111]

[0112] In the formula, f ij represents the elevation values of the four nearby points, where i, j ∈ 1 or 2;

[0113] The actual eligible area with a two-dimensional grid, geographical coordinates, and elevation values is plotted as a three-dimensional surface using the surf function in MATLAB; apply smooth shading (shading interp) and a color map (such as turbo) to display the elevation changes; add lighting and material effects, and use the Phong lighting model to enhance the three-dimensional sense; establish views, coordinate axes, titles, and color bars to make the graphical information more intuitive; limit the display range of the z-axis using the zlim function in MATLAB to only show the part within 1940 meters, obtaining a three-dimensional map;

[0114] By selecting positions at a preset ratio within the terrain range, ensure that the starting point and the ending point are within the terrain range. The x q coordinate of the starting point takes 0.40 of the length of the three-dimensional map, and the y q coordinate takes 0.50 of the width of the three-dimensional map; the x z coordinate of the ending point takes 0.62 of the length of the three-dimensional map, and the y z coordinate takes 0.33 of the width of the three-dimensional map;

[0115] Use the interp2 function to obtain the elevation values of the starting point and the ending point in the interpolated terrain data, and increase the elevation values of the starting point and the ending point by 1 meter to ensure that the marked points are clearly visible in the three-dimensional graph;

[0116] Calculate the actual distance from the starting point to the ending point. Use the norm function to calculate the Euclidean distance between two points in three-dimensional space. The expression is as follows:

[0117]

[0118] In the formula, (x z , y z , z z ) represents the coordinates of the three-dimensional space position of the ending point, where z z represents the elevation value of the ending point; (x q , y q , z q ) represents the coordinates of the three-dimensional space position of the starting point, where z q represents the elevation value of the starting point; obtain a reference baseline through the Euclidean distance for comparison with the path distance planned by the present invention to evaluate the rationality and accuracy of the planning result;

[0119] As Figure 3 shown, use the MarkerSize parameter to control the size of the marker, and use the MarkerFaceColor parameter to set the filling color of the marker; Figure 3 Among them, use a green solid circle to represent the starting point and a red solid circle to represent the ending point.

[0120] S2. Add a fire intensity function composed of wind speed, temperature, and smoke concentration to the reference baseline for the normal flight of the drone in a fire environment;

[0121] The fire intensity function P(x h ,y h ,z h ) is expressed as follows:

[0122] P(x h ,y h ,z h +Δt) = f(P(x h ,y h ,z h ),V wind ,H temp ,ρ smoke )

[0123] In the formula, V wind represents the wind speed, which is used to affect the direction of flame spread; H temp represents the temperature, which is used to affect the intensity of the fire source; ρ smoke represents the smoke concentration; Δt represents the fire spreading time; among them, when H temp ≥80°C, the temperature is too high and is marked as a no-fly zone and needs to be avoided; when ρ smoke >0.7, the smoke concentration is too high and is marked as a no-fly zone and needs to be avoided;

[0124] The temperature H temp increases linearly with the real-time fire detection model and the fire spreading time Δt;

[0125] The expression of the real-time fire detection model is as follows:

[0126]

[0127] In the formula, T safe represents the maximum temperature that the drone can withstand, T safe <80°C; T(x i' ,y i' ) represents the real-time temperature of the drone, where (x i' ,y i' ) represents the real-time coordinates of the drone; N represents the number of coordinates of the drone in the target area.

[0128] S3. Construct a constraint function on the reference baseline with the fire intensity function added to obtain the safe flight conditions of the drone;

[0129] In the constraint function, combine the building height of the target area to construct the height constraint condition of the drone, and the expression is as follows:

[0130] Z min (xa , y a ) = H building (x, y) + H safe

[0131] Wherein, Z min (x a , y a ) represents the lowest flight altitude of the UAV at the position; H building (x, y) represents the height of the actual geographical coordinates of the building at the position (x, y), which is the height of the obstacle that the UAV needs to avoid; H safe represents that the safety altitude margin of the UAV is 2 meters; the UAV path must satisfy H safe > Z min (x a , y a );

[0132] The constraints of the actual environment of the target area also include: the constraints of the crowded area and the constraints of the signal coverage area;

[0133] The constraint of the crowded area is: establish a crowd heat map by collecting the crowd density information of the crowded area. Among them, the crowd density information includes: the past activity data of the crowd, the location and time data of the crowd entering and leaving the venue, and the peak data of the dining places of the crowd; by integrating the crowd heat map into the UAV path planning, the UAV gets to avoid the crowded area and fly safely;

[0134] The integration is: construct a weighted function by using the past activity data, class time data, and restaurant peak data, and obtain a model representing "cost" through data normalization; when randomly sampling by a sampling-based algorithm (such as RRT), reduce the sampling probability of high-cost areas and guide the path to avoid the crowded area;

[0135] In this embodiment, the target area is a certain university area. For the crowded area, use the cameras, infrared sensors or WiFi access points on the campus to collect the crowd density information in real time to establish a crowd heat map. Among them, the crowd density information includes the past activity data, the location and time data of the class, and the peak data of the restaurant. When planning the path, integrate the heat map data into the UAV path planning, assign a "cost" value to each area, set a higher "cost" for the crowded area, and avoid these high "cost" areas or increase the flight altitude when searching for the path;

[0136] The constraint of the signal coverage area is: considering the communication requirements between the UAV and the ground control center, set a higher "cost" value for the weak signal area. When the path planning algorithm searches for the optimal path, it will automatically avoid the signal blind area to ensure stable data transmission.

[0137] S4. Optimize the key parameters of the Rapidly-exploring Random Tree (RRT) using the multi-objective optimization method (MOGA); the steps are as follows:

[0138] S4.1. Set the shortest path and path smoothness during the RRT search as the optimization objectives, and input the multi-objective optimization method to optimize the UAV path;

[0139] Path smoothness is used to reduce the flight energy consumption and control difficulty of the UAV;

[0140] The optimization method is:

[0141] Construct the fitness function of the UAV with the shortest path and path smoothness, and the expression is:

[0142] F = w1f1 + w2f2 + w3C f + w4C c

[0143] In the formula, F represents the total optimization model of the UAV; f1(x) represents the shortest path optimization objective; f2(x) represents the path smoothness optimization objective; C f (x, y) represents the fire impact cost; C d (x, y) represents the crowd impact cost; w1 represents the weight of the shortest path optimization objective; w2 represents the weight of the path smoothness optimization objective; w3 represents the weight of the fire impact cost; w4 represents the weight of the crowd impact cost; in the present invention, w1 = 0.35, w2 = 0.25, w3 = 0.30, w4 = 0.10;

[0144] The expression of the shortest path optimization objective f1(x) is as follows:

[0145]

[0146] In the formula, n represents the total number of points on the path; x ii represents the ii-th point on the path;

[0147] The expression of the path smoothness optimization objective f2(x) is as follows:

[0148]

[0149] The fire impact cost C f (x, y) has the following expression:

[0150] C f (x, y) = w f *P(x h , y h , z h )

[0151] Where, w f represents the learning coefficient of the fire intensity P(x h , y h , z h );

[0152] The crowd influence cost C d (x, y) is expressed as follows:

[0153] C c (x, y) = w c *D dentisty (x, y)

[0154] Where, w c represents the learning coefficient of the crowd influence cost D dentisty (x, y);

[0155] S4.2. Extract the key parameters of the multi-objective optimization method optimization process and optimize the key parameters through the non-dominated sorting genetic algorithm (NSGA-II) to obtain the optimal result;

[0156] The key parameters include: step size; maximum number of iterations; target sampling rate; target threshold; sampling bias;

[0157] The expression for constructing the key parameters for inputting into the non-dominated sorting genetic algorithm is:

[0158] X = [s, N', γ', τ, β]

[0159] Where, s represents the step size (Step size), with a range of [1, 10], which is used to affect the growth pace of the tree; represents the maximum number of iterations (Maxiterations), with a range of [100, 1000], which is used to determine the computational budget of the search; represents the target point sampling rate (Goal samplerate), with a range of [0.1, 0.9], which is used to determine the probability of the target point in the sampling; represents the target area threshold (Goalthreshold), with a range of [0.5, 5.0], which is used to judge whether the path reaches the target point; represents the sampling bias (Samplingbias), with a range of [0.0, 1.0], which is used to affect the randomness of the sampling point distribution;

[0160] Use the non-dominated sorting genetic algorithm (NSGA-II) to solve the optimization objective in the optimization process, and the expression is as follows:

[0161] min f1(x), f2(x), C f , C c ;

[0162] Optimize the results using Pareto front analysis to obtain a set of optimal solutions; select the solution with the minimum total objective as the optimal parameter, and the expression is as follows:

[0163] X * = argmin f1(x);

[0164] The optimized parameters are: step size: 1.34; maximum number of iterations: 100; target sampling rate: 0.86; target threshold: 1.57; sampling bias: 0.73.

[0165] S5. Substitute the optimized parameters into the improved rapidly-exploring random tree model to solve the intelligent sampling strategy for the UAV path under safe flight conditions and perform path smoothing optimization operations using Bessel curves; construct a fire source detection and personnel evacuation guidance model for UAV emergency fire handling through a multi-objective optimization function to complete the path planning for UAV emergency fire handling;

[0166] The improved rapidly-exploring random tree model is: based on the traditional rapidly-exploring random tree algorithm, add an intelligent sampling strategy, path smoothing optimization, and the obtained optimized key parameters to improve the feasibility and smoothness of the path. Its key code is as Figure 4 shown;

[0167] The intelligent sampling strategy includes: target point bias operation and Gaussian sampling operation;

[0168] The expression of the target point bias operation is as follows:

[0169]

[0170] The sampling target point q goal With probability p g Perform direct sampling or through 1 - (p g + p b ) probability to perform uniform sampling to obtain the target point bias sampling result (x rand , y rand ), where p g + p b ≤ 1;

[0171] The Gaussian sampling operation is: with probability p b Based on the mean point q of the current tree mean Perform Gaussian sampling, and the expression is as follows:

[0172]

[0173] In the formula, Ν”(q mean , Σ) represents using a Gaussian distribution (normal distribution) with q mean as the mean and Σ as the covariance matrix;

[0174] The intelligent sampling strategy expression is as follows:

[0175]

[0176] The solution method is: using the nearest neighbor search optimization (accelerated by k-d tree); as the number N of the coordinates of the human and machine in the target area increases, the computational cost of brute-force search grows linearly with the data scale, while the cost of the nearest neighbor search optimization (accelerated by k-d tree) only grows logarithmically with the data scale, thus bringing the advantage of a significant decrease in computational complexity (from linear to logarithmic), that is, reducing the computational complexity from O(N) to O(logN);

[0177] In this embodiment, the computational complexity is reduced from O(1000) = 1000 to O(log1000) = 10 (100 times faster);

[0178] During the solution process, in order to accelerate the expansion speed in the sparse area and reduce the step size in the dense area, the expression for reducing the step size is:

[0179] η = η max ·e -λd

[0180] In the formula, η represents the reduced step size; η max represents the maximum step size in the sparse area; λ represents the attenuation coefficient; d represents the distance from the current point to other nodes;

[0181] The path smoothing optimization operation is: using the cubic Bezier curve interpolation method to optimize the solved path, and the expression is as follows:

[0182] B(t) = (1 - t) 3 P0 + 3(1 - t) 2 tP1 + 3(1 - t)t 2 P2 + t 3 P3, t ∈ [0, 1]

[0183] In the formula, P0 represents the starting point on the path; P3 represents the ending point on the path; P1 represents the first intermediate control point; P2 represents the second intermediate control point; t represents the normalization parameter; P1 and P2 are taken from the inflection points in the path generated by the improved rapidly-exploring random tree model. In order to avoid excessive path bending, it is required that the minimum turning radius R min of the unmanned aerial vehicle cannot exceed the corresponding maximum allowable curvature, which is expressed as: K max = 1 / R min ; it is stipulated that R min = 3 meters, then the maximum allowable curvature is 0.33; if the maximum allowable curvature exceeds 0.33, then the points P1 and P2 need to be adjusted; when the maximum allowable curvature < 0.2, the path is considered smooth;

[0184] The expressions for adjusting points P1 and P2 are as follows:

[0185]

[0186] The constraints are:

[0187] P1 = (1 - α)P0 + αP3, P2 = (1 - β)P0 + β'P3

[0188] In the formula, n' represents the total number of points on the solved path; P iiii represents the iii-th point on the path; α represents the smoothness parameter of the first control path; β' represents the smoothness parameter of the second control path; P3 represents the reference point;

[0189] The expression for constructing a fire source detection and personnel evacuation guidance model for fire drones through a multi-objective optimization function is:

[0190] F = α f f fire-detection + β f f evacuation-guide + γ f f shortest-path

[0191] In the formula, f fire-detection represents the data composed of wind speed, temperature, smoke concentration, and flight altitude collected by the drone for the fire; f evacuation-guide represents the safe path for the drone to guide personnel evacuation; f shortest-path represents the optimal path obtained by the drone through the improved rapidly-exploring random tree model; α f represents the weight of the data composed of wind speed, temperature, smoke concentration, and flight altitude collected by the drone for the fire; β f represents the weight of the safe path for the drone to guide personnel evacuation; γ f represents the weight of the optimal path obtained by the drone through the improved rapidly-exploring random tree model.

[0192] As Figure 5 shown, the total length of the fire drone flight path finally obtained by the present invention is reduced by 18%, the path smoothness is improved by 25%, the bypass rate of high-fire-risk areas reaches 95%, and the crowd avoidance rate is increased by 60%, ensuring that the drone can efficiently and safely perform rescue and guidance tasks in fire emergency missions.

Claims

1. An optimization method for the emergency handling path of a fire-fighting drone, characterized in that, It includes the following steps: S1. Establish a 3D map of the target area through an open-source dataset, set the starting position and ending position of the drone in the obtained 3D map according to a preset ratio, and obtain a reference baseline by calculating the Euclidean distance between the starting position and the ending position; S2. Add a fire intensity function composed of wind speed, temperature, and smoke concentration to the reference baseline for the normal flight of the drone in a fire environment; S3. Construct a constraint function on the reference baseline with the fire intensity function added to obtain the safe flight conditions of the drone; S4. Use a multi-objective optimization method to optimize the key parameters of the rapidly-exploring random tree; S5. Substitute the optimized parameters into the improved rapidly-exploring random tree model to solve the intelligent sampling strategy for the drone path under safe flight conditions and perform path smoothing optimization operations using Bezier curves; construct a fire source detection and personnel evacuation guidance model for fire drone emergency handling through a multi-objective optimization function to complete the path planning for fire drone emergency handling.

2. The optimized method for the emergency handling path of a fire drone according to claim 1, wherein The step of establishing a 3D map of the target area through an open-source dataset, setting the starting position and ending position of the drone in the obtained 3D map according to a preset ratio, and obtaining a reference baseline by calculating the Euclidean distance between the starting position and the ending position is as follows: Use MATLAB to establish a qualified area of the digital elevation model of the target area; among them, the qualified area is to set the area where the elevation of the target area in the digital elevation model is higher than the lowest sea level height to an invalid value to obtain the qualified area, and the lowest sea level height is set to 1940 meters; the setting resolution of the digital elevation model is 2 meters; Use MATLAB to convert pixel coordinates into actual geographical coordinates and establish a qualified area with a two-dimensional grid and actual geographical coordinates through MATLAB; Obtain the elevation values of the grid points in the qualified area of the two-dimensional grid and actual geographical coordinates through data interpolation; Use MATLAB to establish a 3D map of the actual qualified area with elevation values, a two-dimensional grid, and actual geographical coordinates; Specify the location and end point of the drone on the 3D map according to a preset ratio, ensuring that the starting point and the end point are within the terrain range. At the same time, increase the location and end point by 1 meter to ensure that the marked points are clearly visible in the 3D map. Among them, the preset ratio is: the starting point x q coordinate takes 0.40 of the length of the 3D map, y q coordinate takes 0.50 of the width of the 3D map; the end point x z coordinate takes 0.62 of the length of the 3D map, y z coordinate takes 0.33 of the width of the 3D map; Use MATLAB to obtain the Euclidean distance between the starting point and the ending point, that is, the reference baseline.

3. The method for optimizing the emergency handling path of a fire-fighting drone according to claim 1, wherein The expression of the fire intensity function for adding a fire intensity function composed of wind speed, temperature, and smoke concentration to the reference baseline for the normal flight of the drone in a fire environment is as follows: P(x h ,y h ,z h +Δt) = f(P(x h ,y h ,z h ),V wind ,H temp ,ρ smoke ) Wherein, V wind represents the wind speed; H temp represents the temperature; ρ smoke represents the smoke concentration; Δt represents the fire spreading time; wherein, when H temp ≥ 80°C, the temperature is too high and is marked as a no-fly zone and needs to be avoided; when ρ smoke > 0.7, the smoke concentration is too high and is marked as a no-fly zone and needs to be avoided.

4. The method for optimizing the emergency handling path of a fire drone according to claim 1, characterized in that, Construct a constraint function on the reference baseline with the fire intensity function added to obtain the safe flight conditions of the drone; In the constraint function, combine the building height of the target area to construct the height constraint condition of the drone, and the expression is as follows: Z min (x a ,y a ) = H building (x, y) + H safe where Z min (x a , y a ) represents the minimum flight altitude of the UAV at the position; H building (x, y) represents the height of the actual geographical coordinates of the building at the position (x, y), which is the height of the obstacle that the UAV needs to avoid; H safe represents that the safety altitude margin of the UAV is 2 meters; the UAV path must satisfy H safe > Z min (x a , y a ).

5. The method for optimizing the emergency handling path of a fire drone according to claim 1, wherein The steps of using a multi-objective optimization method to optimize the key parameters of the rapidly-exploring random tree are as follows: S4.

1. Set the shortest path and path smoothness during the search of the rapidly-exploring random tree as the optimization objectives, and input the multi-objective optimization method to optimize the drone path; The optimization method is: Construct a fitness function for the drone with the shortest path and path smoothness, and the expression is: F = w1f1 + w2f2 + w3C f + w4C c In the formula, F represents the overall optimization model of the UAV; f1(x) represents the optimization objective of the shortest path; f2(x) represents the optimization objective of path smoothness; C f (x, y) represents the fire impact cost; C d (x, y) represents the crowd impact cost; w1 represents the weight of the shortest path optimization objective; w2 represents the weight of the path smoothness optimization objective; w3 represents the fire impact cost weight; w4 represents the crowd impact cost weight; The expression of the shortest path optimization objective f1(x) is as follows: where n represents the total number of points on the path; x ii represents the i-th point on the path; The expression of the path smoothness optimization objective f2(x) is as follows: Fire impact cost C f (x,y) is expressed as follows: C f (x,y) = w f *P(x h ,y h ,z h ) where w f represents the learning coefficient of the fire intensity P(x h , y h , z h ); Crowd impact cost C d (x,y) is expressed as follows: C c (x,y) = w c *D dentisty (x,y) where w c represents the crowd influence cost D dentisty (x, y) learning coefficient; S4.

2. Extract the key parameters in the optimization process of the multi-objective optimization method and optimize the key parameters through the non-dominated sorting genetic algorithm to obtain the optimal result; The key parameters include: step size; maximum number of iterations; target sampling rate; target threshold; sampling bias; The expression for constructing the key parameters for inputting into the non-dominated sorting genetic algorithm is: X = [s, N', γ', τ, β] In the formula, s represents the step size, with a range of [1, 10]; represents the maximum number of iterations, with a range of [100, 1000]; represents the target point sampling rate, with a range of [0.1, 0.9]; represents the target area threshold, with a range of [0.5, 5.0]; represents the sampling deviation, with a range of [0.0, 1.0]; Use the non-dominated sorting genetic algorithm to solve the optimization objective in the optimization process; Adopt the Pareto front analysis for the optimization result to obtain a set of optimal solutions; select the solution with the minimum total objective as the optimal parameter; The optimized parameters are: step size: 1.34; maximum number of iterations: 100; target sampling rate: 0.86; target threshold: 1.57; sampling bias: 0.

73.

6. The method for optimizing the emergency handling path of a fire drone according to claim 1, wherein Bring the optimized parameters into the improved rapidly-exploring random tree model to solve the intelligent sampling strategy for the UAV path under safe flight conditions and use the Bezier curve for path smoothness optimization operation. Among them, the improved rapidly-exploring random tree model is: on the basis of the traditional rapidly-exploring random tree algorithm, add the intelligent sampling strategy, path smoothness optimization, and the obtained optimized key parameters; The intelligent sampling strategy includes: target point bias operation and Gaussian sampling operation; The solution method for bringing the optimized parameters into the improved rapidly-exploring random tree model to solve the intelligent sampling strategy for the UAV path under safe flight conditions is: use the nearest neighbor search optimization; The path smoothness optimization operation is: use the third-order Bezier curve interpolation method to optimize the solved path, and the expression is as follows: B(t) = (1 - t) 3 P0 + 3(1 - t) 2 tP1 + 3(1 - t)t 2 P2 + t 3 P3, t ∈ [0, 1] In the formula, P0 represents the starting point on the path; P3 represents the ending point on the path; P1 represents the first intermediate control point; P2 represents the second intermediate control point; t represents the normalized parameter; P1 and P2 are taken from the inflection points in the path generated by the improved rapidly-exploring random tree model; The minimum turning radius R of the drone is required min not to exceed the corresponding maximum allowable curvature, expressed as: K max = 1 / R min ; It is stipulated that R min = 3 m, then the maximum allowable curvature is 0.33; If the maximum allowable curvature exceeds 0.33, then points P1 and P2 need to be adjusted; The expressions for adjusting points P1 and P2 are as follows: The constraint is: P1 = (1 - α)P0 + αP3, P2 = (1 - β)P0 + β'P3 where n' represents the total number of points on the solved path; P iiii represents the i-th point on the path; α represents the smoothness parameter of the first control path; β' represents the smoothness parameter of the second control path; P3 represents the reference point.

7. A method for optimizing the emergency handling path of a fire-fighting drone according to claim 1, characterized in that, The expression for constructing the fire source detection and personnel evacuation guidance model for UAV emergency handling through the multi-objective optimization function is: F = α f f fire-detection + β f f evacuation-guide + γ f f shortest-path where f fire-detection represents the data composed of wind speed, temperature, smoke concentration, and flight altitude collected by the drone fire; f evacuation-guide represents the safe path for the drone to guide personnel evacuation; f shortest-path represents the optimal path obtained by the improved rapidly-exploring random tree model for the UAV; α f represents the weights of the data composed of wind speed, temperature, smoke concentration, and flight altitude collected by the UAV for the fire; β f represents the weight of the safe path for the UAV to guide personnel evacuation; γ f represents the weight of the optimal path obtained by the UAV through the improved rapidly-exploring random tree model.

Images