UAV positioning navigation and path planning optimization method based on machine learning

Abstract

The present invention discloses a method for optimizing positioning, navigation and path planning of unmanned aerial vehicles based on machine learning, including dividing the unmanned aerial vehicle activity area into three-dimensional grids based on the coverage of sensors, obtaining the obstacle layout of each grid, calculating the environmental complexity of each grid based on the obstacle layout, setting a response time for the grid according to the environmental complexity, constructing an obstacle avoidance model, obtaining an obstacle avoidance strategy according to the obstacle avoidance model, constructing a pathfinding algorithm based on the obstacle avoidance strategy, using the response time as the interval search time of the pathfinding algorithm, obtaining a feasible path set based on the pathfinding algorithm in the unmanned aerial vehicle activity area, constructing a multi-objective function to optimize the feasible path set, and obtaining the optimal path. The method adopts a machine learning method, can adapt to the needs of different environments and tasks, effectively enhances the obstacle avoidance capability and optimizes the path, and at the same time improves the autonomy and intelligence level of the unmanned aerial vehicle, and has good interpretability.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle control technology, and in particular to a method for unmanned aerial vehicle positioning, navigation and path planning optimization based on machine learning. Background Art

[0002] In the field of drone applications, especially in complex environments such as cities, forests or disaster areas, drone positioning, navigation and path planning are crucial to ensuring the efficiency and safety of the mission.

[0003] Traditional UAV positioning, navigation and path planning methods mainly rely on GPS and preset flight routes, but this method has problems such as signal blocking, poor environmental adaptability and inability to handle complex dynamic environments. It is necessary to optimize UAV positioning, navigation and path planning based on intelligent algorithm models.

[0004] At present, although there are some methods for UAV positioning, navigation and path planning based on algorithm models, these methods often optimize the path through a single algorithm model and only focus on optimizing the shortest path. They lack analysis of the global path in complex environments and only optimize the path through a simple combination of constraints and objective functions. In addition, some algorithms have weak response capabilities when facing real-time changing environments and dynamic obstacles, and are unable to quickly adjust the path to avoid collisions. Summary of the invention

[0005] The purpose of the present invention is to provide a method for optimizing UAV positioning, navigation and path planning based on machine learning.

[0006] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0007] The present invention provides a method for optimizing positioning, navigation and path planning of unmanned aerial vehicles based on machine learning, comprising the following steps:

[0008] Divide the drone activity area into three-dimensional grids based on sensor coverage;

[0009] Obtaining the obstacle layout of each grid, calculating the environmental complexity of each grid based on the obstacle layout, and setting a response time for each grid according to the environmental complexity;

[0010] Constructing an obstacle avoidance model, setting an obstacle avoidance strategy according to the obstacle avoidance model, constructing a pathfinding algorithm based on the obstacle avoidance strategy, and using the response time as an interval search time of the pathfinding algorithm;

[0011] A feasible path set is obtained in the drone activity area based on the path-finding algorithm, and the feasible path set is optimized by multi-objective optimization to obtain an optimal path.

[0012] Furthermore, the method for calculating the environmental complexity of each grid based on the obstacle layout includes:

[0013] Calculate the obstacle complexity in each grid, the expression is:

[0014]

[0015] Where N is the total number of obstacles in the grid, S i is the volume of the ith obstacle, P is the sensor coverage of the drone, E is the electromagnetic interference level in the grid, and D i is the natural discontinuity scale of the ith obstacle, λ is the adjustment coefficient, A i is the cross-sectional area of ​​the ith obstacle, A max is the maximum cross-sectional area among all obstacles, dist(O i ,O j ) is the distance between the i-th obstacle and the j-th obstacle;

[0016] The obstacle complexity is normalized to obtain the environment complexity.

[0017] Furthermore, the method of setting the response time for each grid according to the complexity of the environment includes:

[0018] Obtain the weight of the UAV's floating-point operations per second, communication bandwidth, and sensor resolution indicators on the UAV's computing power;

[0019] Calculate the base time using the expression:

[0020]

[0021] Among them, ω1, ω2 and ω3 are the weights of the number of floating-point operations per second, communication bandwidth and sensor resolution of the drone on the computing power of the drone, respectively. d S is the number of floating point operations per second of the drone. d is the communication bandwidth of the UAV, C m is the sensor resolution of the drone, R is the coverage of the drone sensor, V m is the maximum achievable speed of the drone;

[0022] Set the response time based on the environment complexity and the benchmark time. The expression is:

[0023]

[0024] Among them, T0 is the benchmark time and K is the environment complexity.

[0025] Furthermore, the method for constructing an obstacle avoidance model includes:

[0026] The obstacle avoidance model is constructed based on the potential field force method, specifically:

[0027] Define the repulsive force potential field function, the expression is:

[0028]

[0029] Among them, U r (x,x0) is the repulsive force potential field function generated by the obstacle center x0 at the position x of the UAV, k r is the repulsive force strength coefficient, ∈ is the softening parameter, x is the current position of the UAV, x0 is the center position of the obstacle, ‖x-x0‖ is the Euclidean distance between the UAV position x and the obstacle center x0, and R0 is the safety radius of the obstacle;

[0030] Calculate the repulsive force, the expression is:

[0031]

[0032] in, Expressed as the gradient of the repulsive force potential field function, It is expressed as the direction of repulsive force;

[0033] Update the drone's speed and position according to the repulsive force. The expression is:

[0034] v(t+Δt)=v(t)+Δt·F r (x)

[0035] x(t+Δt)=x(t)+Δt·v(t+Δt)

[0036] Among them, v(t+Δt) is the velocity vector of the UAV at time t+Δt, v(t) is the velocity vector of the UAV at time t, Δt is the time increment, x(t+Δt) is the position vector of the UAV at time t+Δt, and x(t) is the position vector of the UAV at time t.

[0037] Furthermore, the method for setting an obstacle avoidance strategy according to the obstacle avoidance model includes:

[0038] Construct a speed obstacle set for the drone to obtain the speed constraint during the drone's obstacle avoidance process. The speed constraint must be met when the obstacle avoidance model updates the drone's speed and position.

[0039] For stationary obstacles, avoidance is performed based on the obstacle avoidance model and speed constraints;

[0040] For moving obstacles, a trajectory prediction model is constructed to obtain the predicted trajectory of the obstacle, the center of the repulsive force is updated, and obstacle avoidance is performed in real time based on the obstacle avoidance model and speed constraints.

[0041] Furthermore, the method of constructing a speed obstacle set for the UAV and obtaining the speed constraint of the UAV during obstacle avoidance includes:

[0042] Dilate the area of ​​the obstacle to determine the safety radius of the drone and the obstacle;

[0043] For each obstacle, a speed barrier set is constructed, and the expression is:

[0044]

[0045] Where v is the velocity vector of the drone, is a three-dimensional real space, v rel and P rel are the relative velocity vector and relative position vector between the UAV and the obstacle, respectively, |P rel | is the Euclidean distance between the drone and the obstacle, R A is the safety radius of the drone, R B is the safety radius of the obstacle, C is the square of the closest distance when the UAV collides with the obstacle within time τ, that is, the speed obstacle boundary, and τ is the time required for the UAV to detect the obstacle and perform the obstacle avoidance action;

[0046] For each obstacle, the velocity constraint of the drone during obstacle avoidance is obtained, and the expression is:

[0047]

[0048] Where N is the total number of obstacles in the grid, v is the velocity vector of the drone, and V rel,i is the relative velocity vector between the UAV and the ith obstacle, λ i is the adjustment parameter of the safe distance between the UAV and the ith obstacle, n i is the unit vector perpendicular to the surface of the i-th obstacle, i.e., the safety boundary.

[0049] Furthermore, the method for constructing a trajectory prediction model includes:

[0050] Obtain the time series data of the position difference characteristics of the obstacle at continuous time steps, including: speed, acceleration and moving direction;

[0051] A trajectory prediction model is constructed based on the long short-term memory network, including: an input layer, a hidden layer and an output layer, where: the number of neurons in the input layer is set to 3, corresponding to the position difference characteristics of the obstacle in consecutive time steps, the hidden layer includes 128 LSTM units, and the number of neurons in the output layer is set to 3, corresponding to the three-dimensional coordinates of the obstacle in the future time step;

[0052] The time series data is divided into a training set and a test set in a ratio of 7:3, and the trajectory prediction model is trained to obtain a trained trajectory prediction model.

[0053] Furthermore, the method of constructing a pathfinding algorithm based on the obstacle avoidance strategy and using the response time as the interval search time of the pathfinding algorithm includes:

[0054] Use A * The algorithm is combined with the obstacle avoidance strategy to construct a pathfinding algorithm, specifically:

[0055] In A * The algorithm introduces probability selection when selecting the next node, where the expression of the probability selection function is:

[0056]

[0057] Where n represents the current node, λ is the adjustment factor, f(n) is the cost function of node n, min(f) is the minimum value of the cost function among all neighboring nodes, and N is the set of all neighboring nodes of the current node;

[0058] For each neighbor node, the probability of being selected is calculated through the probability selection function. During the search process, obstacles are ensured not to be considered as feasible neighbor nodes. The next node is selected through the cumulative probability method, and the local path is modified through the obstacle avoidance strategy to obtain the path from the starting point to the target point.

[0059] Whenever a path from the starting point to the target point is found, the path is smoothed using the sliding window smoothing method and then added to the set of feasible paths;

[0060] The response time of the grid where the UAV is located is used as the interval search time of the path-finding algorithm. After each interval search time, the path search is performed through the path-finding algorithm.

[0061] Furthermore, the method for optimizing the feasible path set by multi-objective optimization includes:

[0062] Taking the shortest path length, shortest flight time and minimum obstacle distance as multiple objectives, the non-dominated path is identified by applying the Pareto dominance relation in the feasible path set, where the expression of the Pareto dominance relation is:

[0063]

[0064] Among them, x δ and δ are two different paths in the feasible path set, x δ <y δ Represented as path x δIs the Pareto dominated path y δ ,L(x δ ) and L(y δ ) are the paths x δ and path y δ The path length, T(x δ ) and T(y δ ) are the paths x δ and path y δ The flight time, D(x δ ) and D(y δ ) are the paths x δ and path y δ The minimum obstacle distance, m is the index, {L, T, D} represents the objective function set, f m (x δ ) and f m (y δ ) are the paths x δ and path y δ The objective function value in the objective function set;

[0065] All non-dominated paths are screened out from the feasible path set to form a Pareto front set, and a path with the shortest path length is selected from the Pareto front set as the optimal path.

[0066] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0067] (1) The present invention provides a method for optimizing UAV positioning, navigation and path planning based on machine learning, constructs an obstacle avoidance strategy, and * Combining algorithms with step strategies for path optimization and combining multiple algorithms for pathfinding can improve adaptability and stability when facing different environments and conditions, enhance the robustness of the pathfinding algorithm, and effectively respond to dynamic obstacle avoidance and path optimization needs in complex environments.

[0068] (2) The present invention sets a response time for the path-finding algorithm so that the path-finding algorithm can adjust the search frequency according to the actual complexity of the environment, thereby ensuring the real-time performance of path planning and improving the efficiency and response speed of the algorithm. Grids with high environmental complexity are given a longer response time, which helps the UAV to make obstacle avoidance decisions more calmly in complex environments, thereby improving the obstacle avoidance efficiency.

[0069] (3) The present invention uses multi-objective optimization path finding to simultaneously consider path length, flight time, and minimum obstacle distance in path planning, and can find the optimal path that performs well in multiple dimensions, rather than optimizing a single indicator. This method improves the practicality and economy of path planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 This is a flowchart of the steps of the UAV positioning, navigation and path planning optimization method based on machine learning of the present invention. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0072] Reference Figure 1 As shown, the present invention provides a method for optimizing positioning, navigation and path planning of unmanned aerial vehicles based on machine learning, including:

[0073] Divide the drone activity area into three-dimensional grids based on sensor coverage;

[0074] In the actual evaluation, in the optimization of agricultural plant protection UAV path planning, the farmland area is 200 meters × 150 meters, the UAV operation height limit is 10 meters, and the sensor's optimal accuracy range is 50 meters. A three-dimensional grid size of 50 meters × 50 meters × 10 meters is selected to grid the farmland area, and a total of 4 × 3 × 1 = 12 grids are obtained. A three-dimensional coordinate number is assigned to each grid. For example, (1,1,1) represents the first grid;

[0075] Obtaining the obstacle layout of each grid, calculating the environmental complexity of each grid based on the obstacle layout, and setting a response time for each grid according to the environmental complexity;

[0076] In the actual evaluation, the performance parameters of the drone are obtained as follows: the number of floating-point operations per second of the drone P d =10 9 Second, the communication bandwidth of the UAV is S d = 100Mbps, the drone’s sensor resolution is C m =5MP, the maximum achievable speed of the drone is V max=20m / s, use the drone to scan the storage center and its surroundings with the laser radar and high-definition camera, and collect the location, shape and height information of obstacles. Taking the (1,1,1) grid as an example, the obstacle information collected is: Obstacle 1: a tree with a trunk radius of 1 meter and a crown radius of 3 meters; Obstacle 2: a telephone pole with a bottom radius of about 0.5 meters and a safe avoidance radius of 1.5m. The environmental complexity of the grid (1,1,1) is calculated to be 0.217, and the response time is further set to 1.72s for the grid (1,1,1);

[0077] Constructing an obstacle avoidance model, setting an obstacle avoidance strategy according to the obstacle avoidance model, constructing a pathfinding algorithm based on the obstacle avoidance strategy, and using the response time as an interval search time of the pathfinding algorithm;

[0078] In the actual evaluation, the repulsive force intensity coefficient was set to 100, the softening parameter was set to 2, and the obstacle safety radius was set to 5 meters. The obstacle avoidance model of the drone was constructed based on the potential field method. During the flight, the drone detected the surrounding environment in real time through the sensors it carried. Once an obstacle was detected, the obstacle avoidance model was activated. The repulsive force potential field function in the model calculated the distance between the drone and the obstacle, and calculated the repulsive force based on this distance to guide the drone to avoid the obstacle.

[0079] In the actual evaluation, for dynamic obstacles, a trajectory prediction model based on the long short-term memory network (LSTM) was introduced. This model predicts the movement trajectory of the obstacle in the future by analyzing the historical position data of the obstacle. In the actual test, the UAV can plan the route in advance according to the predicted trajectory of the obstacle. The speed obstacle set is further constructed to provide the speed constraint for the UAV in the obstacle avoidance process, ensuring that the UAV will not lose control due to excessive speed when avoiding obstacles, nor will it affect the operation efficiency due to excessive speed. In the actual farmland operation, taking the grid (1,1,1) as an example, the path finding algorithm takes the response time of 1.72s as the interval. While maintaining safe obstacle avoidance, the UAV performs spraying operations at a maximum speed of 20m / s;

[0080] Specifically, in the simulated environment, the drone was able to maintain a safe distance when faced with obstacles of different sizes and shapes, and successfully plan a path to bypass the obstacles. Experimental data showed that the drone began to adjust its route when it was 5 meters away from the obstacle, with an average obstacle avoidance time of 2.5 seconds and an obstacle avoidance success rate of 98%;

[0081] A feasible path set is obtained in the drone activity area based on the path-finding algorithm, and the feasible path set is optimized by multi-objective optimization to obtain an optimal path.

[0082] In the actual evaluation, the shortest path length L, flight time T, and minimum obstacle distance D are used as multiple objectives, and the feasible path set is optimized through multi-objective optimization to obtain the optimal path;

[0083] Specifically, obtain the feasible path set for the drone to plan from point A to point B, including:

[0084] Path 1: length L1 = 500 meters, flight time T1 = 300 seconds, minimum obstacle distance D1 = 10 meters;

[0085] Path 2: length L2 = 480 meters, flight time T2 = 320 seconds, minimum obstacle distance D2 = 8 meters;

[0086] Path 2: length L3 = 520 meters, flight time T3 = 280 seconds, minimum obstacle distance D3 = 12 meters;

[0087] According to the Pareto dominance relationship, we can obtain that: Path 2 is not worse than Path 1 in terms of path length and flight time, and is better in flight time, so Path 2 Pareto dominates Path 1; Path 3 is better than Path 2 in terms of minimum obstacle distance, but its path length and flight time are not as good as Path 2, so Path 2 also Pareto dominates Path 3; we can further obtain that the Pareto frontier set only contains Path 2, and Path 2 is selected as the optimal path.

[0088] In this embodiment, the method for calculating the environmental complexity of each grid based on the obstacle layout includes:

[0089] Calculate the obstacle complexity in each grid, the expression is:

[0090]

[0091] Where N is the total number of obstacles in the grid, S i is the volume of the ith obstacle, P is the sensor coverage of the drone, E is the electromagnetic interference level in the grid, and D i is the natural discontinuity scale of the ith obstacle, λ is the adjustment coefficient, A i is the cross-sectional area of ​​the ith obstacle, A max is the maximum cross-sectional area among all obstacles, dist(O i ,O j ) is the distance between the i-th obstacle and the j-th obstacle;

[0092] The obstacle complexity is normalized to obtain the environment complexity.

[0093] In this embodiment, the method for setting the response time for each grid according to the complexity of the environment includes:

[0094] Obtain the weight of the UAV's floating-point operations per second, communication bandwidth, and sensor resolution indicators on the UAV's computing power;

[0095] Calculate the base time using the expression:

[0096]

[0097] Among them, ω1, ω2 and ω3 are the weights of the number of floating-point operations per second, communication bandwidth and sensor resolution of the drone on the computing power of the drone, respectively. d S is the number of floating point operations per second of the drone. d is the communication bandwidth of the UAV, C m is the sensor resolution of the drone, R is the coverage of the drone sensor, V m is the maximum achievable speed of the drone;

[0098] Set the response time based on the environment complexity and the benchmark time. The expression is:

[0099]

[0100] Among them, T0 is the benchmark time and K is the environment complexity.

[0101] In this embodiment, the method for constructing an obstacle avoidance model includes:

[0102] The obstacle avoidance model is constructed based on the potential field force method, specifically:

[0103] Define the repulsive force potential field function, the expression is:

[0104]

[0105] Among them, U r (x,x0) is the repulsive force potential field function generated by the obstacle center x0 at the position x of the UAV, k r is the repulsive force strength coefficient, ∈ is the softening parameter, x is the current position of the UAV, x0 is the center position of the obstacle, ‖x-x0‖ is the Euclidean distance between the UAV position x and the obstacle center x0, and R0 is the safety radius of the obstacle;

[0106] Calculate the repulsive force, the expression is:

[0107]

[0108] in, Expressed as the gradient of the repulsive force potential field function, It is expressed as the direction of repulsive force;

[0109] Update the drone's speed and position according to the repulsive force. The expression is:

[0110] v(t+Δt)=v(t)+Δt·F r (x)

[0111] x(t+Δt)=x(t)+Δt·v(t+Δt)

[0112] Among them, v(t+Δt) is the velocity vector of the UAV at time t+Δt, v(t) is the velocity vector of the UAV at time t, Δt is the time increment, x(t+Δt) is the position vector of the UAV at time t+Δt, and x(t) is the position vector of the UAV at time t.

[0113] In this embodiment, the method for setting an obstacle avoidance strategy according to the obstacle avoidance model includes:

[0114] Construct a speed obstacle set for the drone to obtain the speed constraint during the drone's obstacle avoidance process. The speed constraint must be met when the obstacle avoidance model updates the drone's speed and position.

[0115] For stationary obstacles, avoidance is performed based on the obstacle avoidance model and speed constraints;

[0116] For moving obstacles, a trajectory prediction model is constructed to obtain the predicted trajectory of the obstacle, the center of the repulsive force is updated, and obstacle avoidance is performed in real time based on the obstacle avoidance model and speed constraints.

[0117] In this embodiment, the method of constructing a speed obstacle set for the drone and obtaining the speed constraint during the obstacle avoidance process of the drone includes:

[0118] Dilate the area of ​​the obstacle to determine the safety radius of the drone and the obstacle;

[0119] For each obstacle, a speed barrier set is constructed, and the expression is:

[0120]

[0121] Where v is the velocity vector of the drone, is a three-dimensional real space, v rel and P rel are the relative velocity vector and relative position vector between the UAV and the obstacle, respectively, |P rel | is the Euclidean distance between the drone and the obstacle, R A is the safety radius of the drone, R B is the safety radius of the obstacle, C is the square of the closest distance when the UAV collides with the obstacle within time τ, that is, the speed obstacle boundary, and τ is the time required for the UAV to detect the obstacle and perform the obstacle avoidance action;

[0122] For each obstacle, the velocity constraint of the drone during obstacle avoidance is obtained, and the expression is:

[0123]

[0124] Where N is the total number of obstacles in the grid, v is the velocity vector of the drone, and V rel,i is the relative velocity vector between the UAV and the ith obstacle, λ i is the adjustment parameter of the safe distance between the UAV and the ith obstacle, n i is the unit vector perpendicular to the surface of the i-th obstacle, i.e., the safety boundary.

[0125] In this embodiment, the method for constructing a trajectory prediction model includes:

[0126] Obtain the time series data of the position difference characteristics of the obstacle at continuous time steps, including: speed, acceleration and moving direction;

[0127] A trajectory prediction model is constructed based on the long short-term memory network, including: an input layer, a hidden layer and an output layer, where: the number of neurons in the input layer is set to 3, corresponding to the position difference characteristics of the obstacle in consecutive time steps, the hidden layer includes 128 LSTM units, and the number of neurons in the output layer is set to 3, corresponding to the three-dimensional coordinates of the obstacle in the future time step;

[0128] The time series data is divided into a training set and a test set in a ratio of 7:3, and the trajectory prediction model is trained to obtain a trained trajectory prediction model.

[0129] In this embodiment, the method of constructing a pathfinding algorithm based on the obstacle avoidance strategy and using the response time as the interval search time of the pathfinding algorithm includes:

[0130] Use A * The algorithm is combined with the obstacle avoidance strategy to construct a pathfinding algorithm, specifically:

[0131] In A * The algorithm introduces probability selection when selecting the next node, where the expression of the probability selection function is:

[0132]

[0133] Where n represents the current node, λ is the adjustment factor, f(n) is the cost function of node n, min(f) is the minimum value of the cost function among all neighboring nodes, and N is the set of all neighboring nodes of the current node;

[0134] For each neighbor node, the probability of being selected is calculated through the probability selection function. During the search process, obstacles are ensured not to be considered as feasible neighbor nodes. The next node is selected through the cumulative probability method, and the local path is modified through the obstacle avoidance strategy to obtain the path from the starting point to the target point.

[0135] Whenever a path from the starting point to the target point is found, the path is smoothed using the sliding window smoothing method and then added to the set of feasible paths;

[0136] The response time of the grid where the UAV is located is used as the interval search time of the path-finding algorithm. After each interval search time, the path search is performed through the path-finding algorithm.

[0137] In this embodiment, the method for optimizing the feasible path set by multi-objective optimization includes:

[0138] Taking the shortest path length, shortest flight time and minimum obstacle distance as multiple objectives, the non-dominated path is identified by applying the Pareto dominance relation in the feasible path set, where the expression of the Pareto dominance relation is:

[0139]

[0140] Among them, x δ and δ are two different paths in the feasible path set, x δ <y δ Represented as path x δ Is the Pareto dominated path y δ ,L(x δ ) and L(y δ ) are the paths x δ and path y δ The path length, T(x δ ) and T(y δ ) are the paths x δ and path y δ The flight time, D(x δ ) and D(y δ ) are the paths x δ and path y δ The minimum obstacle distance, m is the index, {L, T, D} represents the objective function set, f m (x δ ) and f m (y δ ) are the paths x δ and path y δ The objective function value in the objective function set;

[0141] All non-dominated paths are screened out from the feasible path set to form a Pareto front set, and a path with the shortest path length is selected from the Pareto front set as the optimal path.

[0142] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A method for optimizing UAV positioning, navigation and path planning based on machine learning, characterized in that: The following steps are involved: Divide the drone activity area into three-dimensional grids based on sensor coverage; Obtaining the obstacle layout of each grid, calculating the environmental complexity of each grid based on the obstacle layout, and setting a response time for each grid according to the environmental complexity; Constructing an obstacle avoidance model, setting an obstacle avoidance strategy according to the obstacle avoidance model, constructing a pathfinding algorithm based on the obstacle avoidance strategy, and using the response time as an interval search time of the pathfinding algorithm; Obtaining a feasible path set in the drone activity area based on the path-finding algorithm, optimizing the feasible path set by multi-objective optimization, and obtaining an optimal path; The method of constructing a pathfinding algorithm based on the obstacle avoidance strategy and using the response time as the interval search time of the pathfinding algorithm includes: Use A * The algorithm is combined with the obstacle avoidance strategy to construct a pathfinding algorithm, specifically: In A * The algorithm introduces probability selection when selecting the next node, where the expression of the probability selection function is: Where n represents the current node, λ is the adjustment factor, f(n) is the cost function of node n, min(f) is the minimum value of the cost function among all neighboring nodes, and N is the set of all neighboring nodes of the current node; For each neighbor node, the probability of being selected is calculated through the probability selection function. During the search process, obstacles are ensured not to be considered as feasible neighbor nodes. The next node is selected through the cumulative probability method, and the local path is modified through the obstacle avoidance strategy to obtain the path from the starting point to the target point. Whenever a path from the starting point to the target point is found, the path is smoothed using the sliding window smoothing method and then added to the set of feasible paths; The response time of the grid where the UAV is located is used as the interval search time of the path-finding algorithm. After each interval search time, the path search is performed through the path-finding algorithm.

2. The method for optimizing unmanned aerial vehicle positioning, navigation and path planning based on machine learning according to claim 1, characterized in that: The method for calculating the environmental complexity of each grid based on the obstacle layout comprises: Calculate the obstacle complexity in each grid, the expression is: Where N is the total number of obstacles in the grid, S i is the volume of the ith obstacle, P is the sensor coverage of the drone, E is the electromagnetic interference level in the grid, and D i is the natural discontinuity scale of the ith obstacle, λ is the adjustment coefficient, A i is the cross-sectional area of ​​the ith obstacle, A max is the maximum cross-sectional area among all obstacles, dist(O i , O j ) is the distance between the i-th obstacle and the j-th obstacle; The obstacle complexity is normalized to obtain the environment complexity.

3. The method for optimizing unmanned aerial vehicle positioning, navigation and path planning based on machine learning according to claim 1, characterized in that: The method of setting a response time for each grid according to the complexity of the environment comprises: Obtain the weight of the UAV's floating-point operations per second, communication bandwidth, and sensor resolution indicators on the UAV's computing power; Calculate the base time using the expression: Among them, ω1, ω2 and ω3 are the weights of the number of floating-point operations per second, communication bandwidth and sensor resolution of the drone on the computing power of the drone, respectively. d S is the number of floating point operations per second of the drone. d is the communication bandwidth of the UAV, C m is the sensor resolution of the drone, R is the coverage of the drone sensor, V m is the maximum achievable speed of the drone; Set the response time based on the environment complexity and the benchmark time. The expression is: Among them, T0 is the benchmark time and K is the environment complexity.

4. The method for optimizing unmanned aerial vehicle positioning, navigation and path planning based on machine learning according to claim 1, characterized in that: The method for constructing an obstacle avoidance model comprises: The obstacle avoidance model is constructed based on the potential field force method, specifically: Define the repulsive force potential field function, the expression is: Among them, U r (x, x0) is the repulsive force potential field function generated by the obstacle center x0 at the position x of the UAV, k r is the repulsion strength coefficient, ∈ is the softening parameter, x is the current position of the drone, x0 is the center position of the obstacle, ||x-x0|| is the Euclidean distance between the drone position x and the obstacle center x0, and R0 is the safety radius of the obstacle; Calculate the repulsive force, the expression is: in, Expressed as the gradient of the repulsive force potential field function, It is expressed as the direction of repulsive force; Update the drone's speed and position according to the repulsive force. The expression is: v(t+Δt)=v(t)+Δt·F r (x) x(t+Δt)=x(t)+Δt·v(t+Δt) Among them, v(t+Δt) is the velocity vector of the UAV at time t+Δt, v(t) is the velocity vector of the UAV at time t, Δt is the time increment, x(t+Δt) is the position vector of the UAV at time t+Δt, and x(t) is the position vector of the UAV at time t.

5. The method for optimizing unmanned aerial vehicle positioning, navigation and path planning based on machine learning according to claim 1, characterized in that: The method for setting an obstacle avoidance strategy according to the obstacle avoidance model comprises: Construct a speed obstacle set for the drone to obtain the speed constraint during the drone's obstacle avoidance process. The speed constraint must be met when the obstacle avoidance model updates the drone's speed and position. For stationary obstacles, avoidance is performed based on the obstacle avoidance model and speed constraints; For moving obstacles, a trajectory prediction model is constructed to obtain the predicted trajectory of the obstacle, the center of the repulsive force is updated, and obstacle avoidance is performed in real time based on the obstacle avoidance model and speed constraints.

6. The method for optimizing unmanned aerial vehicle positioning, navigation and path planning based on machine learning according to claim 5, characterized in that: The method of constructing a speed obstacle set for a UAV and obtaining a speed constraint in an obstacle avoidance process of the UAV includes: Dilate the area of ​​the obstacle to determine the safety radius of the drone and the obstacle; For each obstacle, a speed barrier set is constructed, and the expression is: Where v is the velocity vector of the drone, is a three-dimensional real space, v rel and P rel are the relative velocity vector and relative position vector between the UAV and the obstacle, respectively, |P rel | is the Euclidean distance between the drone and the obstacle, R A is the safety radius of the drone, R B is the safety radius of the obstacle, C is the square of the closest distance when the UAV collides with the obstacle within time τ, that is, the speed obstacle boundary, and τ is the time required for the UAV to detect the obstacle and perform the obstacle avoidance action; For each obstacle, the velocity constraint of the drone during obstacle avoidance is obtained, and the expression is: Where N is the total number of obstacles in the grid, v is the velocity vector of the drone, and V rel，i is the relative velocity vector between the UAV and the ith obstacle, λ i is the adjustment parameter of the safe distance between the UAV and the ith obstacle, n i is the unit vector perpendicular to the surface of the i-th obstacle, i.e., the safety boundary.

7. The method for optimizing unmanned aerial vehicle positioning, navigation and path planning based on machine learning according to claim 5, characterized in that: The method for constructing a trajectory prediction model comprises: Obtain the time series data of the position difference characteristics of the obstacle at continuous time steps, including: speed, acceleration and moving direction; A trajectory prediction model is constructed based on the long short-term memory network, including: an input layer, a hidden layer and an output layer, where: the number of neurons in the input layer is set to 3, corresponding to the position difference characteristics of the obstacle in consecutive time steps, the hidden layer includes 128 LSTM units, and the number of neurons in the output layer is set to 3, corresponding to the three-dimensional coordinates of the obstacle in the future time step; The time series data is divided into a training set and a test set in a ratio of 7:3, and the trajectory prediction model is trained to obtain a trained trajectory prediction model.

8. The method for optimizing unmanned aerial vehicle positioning, navigation and path planning based on machine learning according to claim 1, characterized in that: The method for optimizing the feasible path set by multi-objective optimization includes: Taking the shortest path length, shortest flight time and minimum obstacle distance as multiple objectives, the non-dominated path is identified by applying the Pareto dominance relation in the feasible path set, where the expression of the Pareto dominance relation is: Among them, x δ and δ are two different paths in the feasible path set, x δ <y δ Represented as path x δ Is the Pareto dominated path y δ ,L(x δ ) and L(y δ ) are the paths x δ and path y δ The path length, T(x δ ) and T(y δ ) are the paths x δ and path y δ The flight time, D(x δ ) and D(y δ ) are the paths x δ and path y δ The minimum obstacle distance, m is the index, {L, T, D} represents the objective function set, f m (x δ ) and f m (y δ ) are the paths x δ and path y δ The objective function value in the objective function set; All non-dominated paths are screened out from the feasible path set to form a Pareto front set, and a path with the shortest path length is selected from the Pareto front set as the optimal path.

Images