A big data-based unmanned aerial vehicle navigation positioning method and system

Abstract

The application discloses a kind of unmanned plane navigation positioning method and system based on big data, comprising: obtaining the position data of unmanned plane, sensor data and environmental data, constructs real-time map according to position data and sensor data using SLAM algorithm and carries out real-time positioning;Sub-region division is carried out to real-time map, and risk score is obtained using cluster analysis;According to environmental data and risk score, the path time between sub-regions is obtained, the fitness function is established according to risk score and path time, and the path planning model is established based on fitness function;According to sensor data, offset data is obtained, path compensation strategy is established according to offset data and path correction is carried out;Unmanned plane navigation model is obtained by learning path compensation strategy and path planning model, and the unmanned plane navigation model is optimized by sine cosine optimization algorithm.The application is suitable for a variety of unmanned plane application scenarios, and promotes the wide application of unmanned plane technology in complex environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle navigation positioning, and particularly relates to an unmanned aerial vehicle navigation positioning method and system based on big data. BACKGROUND

[0002] With the rapid development of unmanned aerial vehicle technology, it has been widely applied in many fields, such as aerial photography, logistics distribution, agricultural plant protection, etc. However, unmanned aerial vehicles face many challenges during flight, and the accuracy and reliability of navigation and positioning are crucial issues.

[0003] Traditional unmanned aerial vehicle navigation and positioning methods mainly rely on the global positioning system (GPS), but in some complex environments, such as urban high-rise areas, mountainous areas, and areas with strong electromagnetic interference, GPS signals may be blocked, interfered or reflected, resulting in inaccurate positioning or even failure. In addition, existing navigation and positioning methods often do not fully consider environmental factors affecting unmanned aerial vehicle flight, such as weather conditions (wind direction, wind speed, etc.), obstacle distribution, and electromagnetic interference, which may increase the risk of unmanned aerial vehicle flight, such as prolonged flight time, increased energy consumption, and even collision accidents. Therefore, it is necessary to develop a more accurate, reliable and adaptable unmanned aerial vehicle navigation and positioning method and system for complex environments. SUMMARY

[0004] The purpose of the present application is to provide an unmanned aerial vehicle navigation and positioning method and system based on big data.

[0005] To achieve the above purpose, the present application is implemented according to the following technical solutions:

[0006] The present application comprises the following steps:

[0007] Obtain the position data, sensor data and environmental data of the unmanned aerial vehicle, and use the SLAM algorithm to construct a real-time map and perform real-time positioning based on the position data and sensor data;

[0008] Divide the real-time map into sub-regions, and use clustering analysis to obtain a risk score;

[0009] Obtain the path time between the sub-regions based on the environmental data and the risk score, establish a fitness function based on the risk score and the path time, and establish a path planning model based on the fitness function;

[0010] Obtain offset data based on the sensor data, establish a path compensation strategy based on the offset data, and perform path correction;

[0011] The UAV navigation model is obtained by learning the path compensation strategy and the path planning model, and then the UAV navigation model is optimized by the sine and cosine optimization algorithm.

[0012] Furthermore, methods for acquiring the location data, sensor data, and environmental data of the drone include:

[0013] The drone's position, obtained through GPS, GLONASS, and Galileo, is fused using a pseudorange algorithm to obtain position data. The sensor data includes data from a 3D camera, barometer, gyroscope, accelerometer, and inertial measurement unit. The sensor data includes attitude data, velocity data, acceleration data, air pressure data, and image data. The environmental data includes meteorological data and electromagnetic interference data. The meteorological data includes wind direction, wind speed, and air pressure, while the electromagnetic interference data includes the intensity and frequency of electromagnetic interference.

[0014] Furthermore, a method for constructing a real-time map and performing real-time positioning using the SLAM algorithm based on the location data and the sensor data includes:

[0015] A GIS system is used to obtain a GIS map of the area surrounding the UAV based on location data. An obstacle model is constructed based on real-time monitoring data from a 3D camera using the SLAM algorithm. A real-time map is then built on the GIS map to obtain the real-time position of the UAV in the real-time map.

[0016] Furthermore, the real-time map is divided into sub-regions, and a risk score is obtained using cluster analysis, including:

[0017] Based on the real-time map, flyable and non-flyable areas are distinguished, and the flyable area is divided into cube-shaped sub-regions.

[0018] Data points are established based on the surrounding obstacle density, wind speed, electromagnetic interference data, and air pressure of the sub-region. The SOM algorithm is then used to cluster the data points to form risk areas.

[0019] K data points are randomly selected as the initial weight vectors for the neurons in the competitive layer. Each neuron in the competitive layer represents a cluster center. The similarity between the data points and the weight vectors of each neuron in the competitive layer is calculated, and the data points are assigned to the cluster of neurons with the highest similarity.

[0020] The formula for calculating similarity is:

[0021]

[0022] d(h,e) represents the similarity between h and e, where h and e are two r-dimensional vectors, ||h(w κ )|| and ||e(w κ )|| are the weighted vector lengths of vectors h and e, respectively, where h is the weighted vector length of vector h and e.κ denotes the component of vector h in the k-th dimension, w κ is the weight coefficient in the k-th dimension, and σ(t) denotes the neighborhood radius centered at neuron i, initially with σ(0) as the neighborhood radius,

[0023] Dynamic topological neighborhood function in the competition layer:

[0024]

[0025] where h gi (t) is the dynamic topological neighborhood function of neuron i and neuron g in the t-th iteration, d ig (t) is the distance between neuron i and neuron g in the t-th iteration, and σ(t) is the neighborhood radius in the t-th iteration, t max is the maximum number of iterations, and δ(t) is the influence factor that changes over time, ξ is the initial value of the influence factor, and ζ is the adjustment coefficient of the influence factor,

[0026] Update of the weight vector of the neuron:

[0027] w ij (t+1) = w ij (t) + η(t)·h gi (t)·(x j -w ij (t))-γ·w ij (t)

[0028] where γ is the strength of the regularization term, w ij (t) and w ij (t+1) are the j-th dimensional components of the weight vector of the i-th neuron in the t-th and t+1-th iterations, respectively, η(t) is the learning rate in the t-th iteration, x j is the j-th dimensional component of the data point,

[0029] The clustering result is output after reaching the maximum number of iterations, and the risk score of each risk area is the average of the weighted vector lengths of the data points in it.

[0030] Further, the method for obtaining the path time between the sub-areas according to the environmental data and the risk score, comprising:

[0031] Path time function:

[0032]

[0033] where T cb is the path time between the center of sub-area c and the center of sub-area b, w PP is the influence weight of the risk score, P c and P b are the risk scores of the risk areas corresponding to the sub-area c and the sub-area b respectively, S c and S b are the parts of the path length between the centers of the sub-area c and the sub-area b in the sub-area c and the sub-area b respectively, V is the speed of the UAV, V f (c) and V f (b) are the wind speeds of the sub-area c and the sub-area b respectively, and are the clockwise angles between the wind directions in the sub-area c and the sub-area b and the advancing direction of the UAV respectively, ω f is the influence weight of the wind speed on the speed of the UAV.

[0034] Further, a method for establishing a fitness function based on the risk score and the path time comprises:

[0035] The fitness function:

[0036]

[0037] wherein f(L) is the fitness function of the path L, c∈L represents that the sub-area c is in the sub-area set of the path L, the sub-area c and the sub-area b are adjacent, cb∈L represents that the sub-area c to b is a sub-path in the path L, W cb is the energy consumption of the sub-path cb, w T and w W are the weights of the path time and the energy consumption respectively,

[0038] The calculation formula of the energy consumption:

[0039]

[0040] wherein m and m l are the weights of the UAV and the carried load respectively, φ is the gravitational acceleration, ΔH is the height change, and are the air pressures of the sub-area c and the sub-area b respectively, is the gas constant, and are the temperatures of the sub-area c and the sub-area b respectively, A is the windward area of the UAV.

[0041] Further, a method for establishing a path planning model based on the fitness function comprises:

[0042] Each ant moves on a directed graph composed of nodes formed by the sub-areas and leaves pheromone, the parameters of the ant colony algorithm are initialized, and the ant transfer probability between nodes is:

[0043]

[0044] wherein is the probability that the kth ant moves from node u to node v at time s, and are the weights of pheromone and heuristic function respectively, is the index of nodes in N(u), N(u) is the set of neighbor nodes of node u, is the heuristic function of node u to node , is the distance of node u to node , is the path length between the center of sub-region u and the center of sub-region , is the risk score of sub-region

[0045] The pheromone update formula is:

[0046]

[0047] wherein, μ uv (s) and μ uv (s+1) represent the amount of pheromone on path uv at time s and time s+1 respectively, α is the pheromone evaporation coefficient, Z uv is the fitness of path uv, and is the fitness function value of path uv,

[0048] When the amount of pheromone update is lower than the update threshold, the best path corresponding to the strongest pheromone signal is output as the planning path,

[0049] When the real-time map and environmental data change, the path planning is re-performed.

[0050] Further, the method for obtaining offset data from the sensor data, establishing a path compensation strategy based on the offset data, and performing path correction, comprises:

[0051] The deviation amount between the actual flight path of the UAV and the planning path is obtained as offset data from the real-time position, attitude data and speed data, including position deviation, attitude deviation and speed deviation, the state equation and observation equation of the UAV are established based on the offset data through the Kalman filtering algorithm, the state estimation of the UAV is continuously performed, the path compensation strategy is reversely applied based on the state estimation and the offset amount of the planning path, the flight direction and speed of the UAV are adjusted, the path correction is performed, and the offset amount of the state estimation and the planning path is reduced.

[0052] Further, the UAV navigation model is obtained by learning the path compensation strategy and the path planning model, and the method for optimizing the UAV navigation model through the sine-cosine optimization algorithm, comprises:

[0053] The unmanned aerial vehicle navigation model is trained by using a reinforcement learning algorithm, the state space is real-time position, attitude data, risk score, path time, speed data and environment data of the unmanned aerial vehicle, the action space is adjusting flight direction and speed of the unmanned aerial vehicle, and the reward function is:

[0054]

[0055] Wherein, k P , k T , k W and k θ are risk score reward coefficient, path time reward coefficient, energy consumption reward coefficient and path stability reward coefficient respectively, T expected and T actrual are path time of the planned path of the path planning model and actual path time of the reinforcement learning algorithm respectively, W expected and W actrual are energy consumption of the planned path and actual energy consumption of the reinforcement learning algorithm respectively, Δθ is the deviation rate of the path direction from the path planning direction of the reinforcement learning algorithm, and θ stabel is the deviation rate threshold of the path direction.

[0056] The reciprocal of the objective function value of the unmanned aerial vehicle navigation model is taken as the fitness value, the position of each search individual represents a set of hyperparameter values, and the search individual position updating formula is:

[0057]

[0058] Wherein, and are the positions of the i-th search individual at the t+1th and tth iterations respectively, X * is the global optimal position, is the historical optimal position searched by the i-th search individual, ω t is the nonlinear weight factor of the tth iteration, is the control parameter of the amplitude of the sine function, t is the current iteration number, t max is the maximum iteration number, r1 is a control parameter, r1=a(1-t / t max ), a is a constant greater than 1, is the adjustment parameter of the frequency of the sine function and the cosine function, r2∈(0,2π), is a random weight obeying uniform distribution, r3∈(0,1), is a random number obeying uniform distribution, used for selecting individual position updating strategy,

[0059] When the average fitness value of the historical optimal position searched by all search individuals does not change for 4 consecutive iterations, it is considered that the search is stagnant, at which time the Levy flight random walk strategy is used to update the position of the search individual, and the random walk step length formula of the Levy flight is:

[0060]

[0061] Where s is the random walk step length, u and v are both parameters subject to normal distribution, Gamma (·) is the gamma function, beta is the exponential parameter controlling the shape of the Levy distribution, beta is in (0, 2), beta is taken as 1.5, sigma u is the standard deviation of u, sigma v is the standard deviation of v;

[0062] The random walk strategy based on the Levy flight is:

[0063]

[0064] Where S randn is a random quantity subject to standard normal distribution, alpha is in [-1, 1], and alpha is taken as 0.9,

[0065] The hyperparameter value corresponding to the global optimal position is output when the maximum number of iterations is reached.

[0066] A big data-based unmanned aerial vehicle navigation positioning system, characterized by comprising:

[0067] A collection module acquires position data, sensor data and environment data of the unmanned aerial vehicle, divides a flight area, and acquires spatial information of the flight area by using a GIS system;

[0068] A calculation module constructs a real-time map and performs real-time positioning according to the position data and the sensor data by using a SLAM algorithm, divides a sub-area of the real-time map, obtains a risk score by using cluster analysis, obtains a path time between the sub-areas according to the environment data and the risk score, establishes a fitness function according to the risk score and the path time, establishes a path planning model based on the fitness function, obtains offset data according to the sensor data, and learns an unmanned aerial vehicle navigation model by using a path compensation strategy and the path planning model;

[0069] A correction module establishes a path compensation strategy and performs path correction according to the offset data;

[0070] An optimization module optimizes the unmanned aerial vehicle navigation model by using a sine-cosine optimization algorithm.

[0071] The present application has the following advantages:

[0072] The present application realizes accurate navigation positioning by comprehensively using multiple technologies. Data is acquired by multiple systems and fused to improve positioning accuracy; environmental risks are analyzed based on big data to optimize path planning; reinforcement learning and optimization algorithms are used to improve the performance of the navigation model, and the adaptability, safety and efficiency of the unmanned aerial vehicle in complex environments are enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0073] Figure 1 A flowchart of the steps of the present application, a big data-based unmanned aerial vehicle navigation positioning method and system. DETAILED DESCRIPTION

[0074] The present application will be further described below through specific embodiments, and the illustrative embodiments of the present application and the description are used to explain the present application, but not as a limitation of the present application.

[0075] The present application, a big data-based unmanned aerial vehicle navigation positioning method and system, comprises the following steps:

[0076] As shown in the figure, in the present embodiment, the following steps are included: Figure 1

[0077] The position data, sensor data and environmental data of the unmanned aerial vehicle are acquired, and a real-time map is constructed and real-time positioning is performed according to the position data and the sensor data using a SLAM algorithm;

[0078] In actual evaluation, in a flight, the initial position data of the unmanned aerial vehicle is longitude 116.404°, latitude 39.915°, and height 50 meters, the height of the unmanned aerial vehicle is calculated by the difference between the altitude of the unmanned aerial vehicle and the altitude of the ground; the unmanned aerial vehicle advances towards the north direction, a three-dimensional camera captures a 100-meter-high and 30-meter-wide obstacle 50 meters in front in real time; the barometer shows that the current air pressure is 1013.25 hPa; the gyroscope and accelerometer provide the attitude and acceleration information of the unmanned aerial vehicle, the pitch angle is 5°, the roll angle is 0°, the speed is 20 m / s, and the acceleration is 0.5 m / s 2 ; the environmental data: the current wind direction is north wind, the wind speed is 3 m / s; the electromagnetic interference intensity is 5 V / m, and the frequency is 1 MHz; after the real-time map is established, it is found that the length of the obstacle is 50 meters, and the initial real-time position of the unmanned aerial vehicle is actually longitude 116.444°, latitude 30.015°, and height 50 meters;

[0079] The real-time map is divided into sub-regions, and a risk score is obtained by cluster analysis;

[0080] ​In actual assessment, the size of each sub-region can be different or the same when dividing the area. The 50×50×50 meter cube area between the drone and the obstacle is divided into sub-region 1, the 50×50×50 meter cube area to the left of the obstacle is divided into sub-region 2, the 50×50×50 meter cube area to the right of the obstacle is divided into sub-region 3, and the 50×50×50 meter cube area behind the obstacle is divided into sub-region 4. The surrounding obstacle densities are 0.4, 0.5, 0.4 and 0.5 for sub-regions 1, 2, 3 and 4, respectively, and the risk scores are 0.3, 0.4, 0.3 and 0.3 for sub-regions 1, 2, 3 and 4, respectively.

[0081] The path time between the sub-regions is obtained based on the environmental data and the risk score. A fitness function is established based on the risk score and the path time. A path planning model is established based on the fitness function.

[0082] In the actual evaluation, the path time was: 8.5s for sub-region 1 to 2, 4.8s for sub-region 1 to 3, 5.3s for sub-region 2 to 4, and 8s for sub-region 3 to 4; the fitness function values ​​were: 0.53 for path 1-2, 0.8 for path 1-3, 0.73 for path 2-4, 0.6 for path 3-4, 1.36 for path 1-2-4, and 1.5 for path 1-3-4; the planned path of the path planning model was path 1-3-4.

[0083] Offset data is obtained based on the sensor data, and a path compensation strategy is established and path correction is performed based on the offset data.

[0084] In the actual assessment, the actual flight path of the drone deviated from the planned path by 2 meters to the west and the speed deviation was -2 m / s. The path compensation strategy was to adjust the drone's flight direction to the northeast and the acceleration to 1 m / s for 2 seconds.

[0085] The UAV navigation model is obtained by learning the path compensation strategy and the path planning model, and then the UAV navigation model is optimized by the sine and cosine optimization algorithm.

[0086] In this embodiment, the method for acquiring the location data, sensor data, and environmental data of a UAV includes:

[0087] The drone's position, obtained through GPS, GLONASS, and Galileo, is fused using a pseudorange algorithm to obtain position data. The sensor data includes data from a 3D camera, barometer, gyroscope, accelerometer, and inertial measurement unit. The sensor data includes attitude data, velocity data, acceleration data, air pressure data, and image data. The environmental data includes meteorological data and electromagnetic interference data. The meteorological data includes wind direction, wind speed, and air pressure, while the electromagnetic interference data includes the intensity and frequency of electromagnetic interference.

[0088] In the embodiment, the method for constructing a real-time map and performing real-time positioning according to the position data and the sensor data by using the SLAM algorithm comprises the following steps:

[0089] The GIS system is used to obtain a GIS map of the surrounding area of the UAV according to the position data, the SLAM algorithm is used to construct an obstacle model based on the real-time monitoring data of the three-dimensional camera, and a real-time map is constructed based on the GIS map to obtain the real-time position of the UAV in the real-time map.

[0090] In the embodiment, the method for obtaining a risk score by using cluster analysis on the basis of the division of the real-time map into sub-regions comprises the following steps:

[0091] The real-time map is used to distinguish between flyable regions and non-flyable regions, the non-flyable regions include no-fly zones, regions exceeding the flight height limit of the UAV, and obstacles, the flyable regions are divided into sub-regions in the form of cubes, the length of each side of each sub-region is greater than or equal to 3 times the maximum length of the UAV and less than or equal to 500 meters, and the parts of the flyable regions that are smaller than the minimum sub-region are removed and regarded as non-flyable regions,

[0092] Data points are established according to the surrounding obstacle density, wind speed, electromagnetic interference data, and air pressure of the sub-regions, the surrounding obstacle density is the volume proportion of obstacles within the influence area of 3 times the maximum length of the UAV around the sub-region, and the SOM algorithm is used to cluster the data points to form risk regions:

[0093] k data points are randomly selected as the initial weight vectors of the competitive layer neurons, one competitive layer neuron represents one cluster center, the similarity between the data points and the weight vectors of each competitive layer neuron is calculated, and the data points are assigned to the cluster of the neuron with the highest similarity,

[0094] The calculation formula of the similarity is as follows:

[0095]

[0096] d(h, e) is the similarity between h and e, h and e are two r-dimensional vectors, ||h(w κ )|| and ||e(w κ )|| are the weighted vector lengths of the vectors h and e respectively, h κ represents the component of the vector h in the κth dimension, w κ is the weight coefficient of the κth dimension, and a neighborhood with a neuron as the center and an initial neighborhood radius σ(0) is drawn,

[0097] Dynamic topological neighborhood function in the competitive layer:

[0098]

[0099] where h gi (t) is the dynamic topological neighborhood function of neuron i and neuron g at the tth iteration, d ig (t) is the distance between neuron i and neuron g at the tth iteration, s(t) is the similarity between neuron i and neuron g, and σ(t) is the neighborhood radius at the tth iteration, t max is the maximum number of iterations, and δ(t) is the influence factor that changes over time, ξ is the initial value of the influence factor, and ζ is the adjustment coefficient of the influence factor,

[0100] Update of the weight vector of the neuron:

[0101] w ij (t+1) = w ij (t) + η(t)·h gi (t)·(x j -w ij (t))-γ·w ij (t)

[0102] where γ is the strength of the regularization term, w ij (t) and w ij (t+1) are the jth component of the weight vector of the ith neuron at the tth and t+1th iteration, respectively, η(t) is the learning rate at the tth iteration, x j is the jth component of the data point,

[0103] The clustering result is output after the maximum number of iterations is reached, and the risk score of each risk area is the average value of the weighted vector length of the data points in it.

[0104] In this embodiment, the method for obtaining the path time between the sub-areas according to the environmental data and the risk score comprises:

[0105] Path time function:

[0106]

[0107] where T cb is the path time between the center of sub-area c and the center of sub-area b, w P is the influence weight of the risk score, P c and P b are the risk scores of the risk areas corresponding to sub-area c and sub-area b, respectively, S c and S b are the parts of the path length between the centers of sub-area c and sub-area b in sub-area c and sub-area b, respectively, and V is the speed of the UAV, V f (c) and V f(b) the wind speed of sub-region c and sub-region b respectively, and the clockwise angle between the wind direction and the advancing direction of the UAV in sub-region c and sub-region b respectively, ω f is the weight of the wind speed on the UAV speed.

[0108] In the embodiment, the method for establishing a fitness function according to the risk score and the path time comprises:

[0109] The fitness function:

[0110]

[0111] wherein f(L) is the fitness function of path L, c∈L represents that sub-region c is in the sub-region set of path L, sub-region c and sub-region b are adjacent, cb∈L represents that sub-region c to b is a sub-path in path L, W cb is the energy consumption of sub-path cb, w T and w W are the weights of the path time and the energy consumption respectively, and the calculation formula of the energy consumption is:

[0112]

[0113] wherein m and m l are the weights of the UAV and the carried load respectively, φ is the gravitational acceleration, and ΔH is the height change, and are the air pressures of sub-region c and b respectively, is the gas constant, and are the temperatures of sub-region c and b respectively, and A is the windward area of the UAV.

[0114] In the embodiment, the method for establishing a path planning model based on the fitness function comprises:

[0115] Each ant moves on a directed graph composed of nodes formed by sub-regions, and leaves pheromone, and the parameters of the ant colony algorithm are initialized, and the ant transfer probability between nodes is:

[0116]

[0117] wherein is the probability that the kth ant transfers from node u to node v at time s, and are the weights of the pheromone and the heuristic function respectively, is the index of the node in N(u), and N(u) is the neighbor node set of node u, is the distance between node u and node heuristic function, is the distance from node u to node is the path length between the center of sub-region u and the center of sub-region is the risk score of sub-region is the risk score of sub-region

[0118] The pheromone update formula is:

[0119]

[0120] wherein μ uv (s) and μ uv (s+1) represent the amount of pheromone on path uv at time s and time s+1 respectively, α is the pheromone evaporation coefficient, Z uv is the fitness of path uv, and f is the fitness function value of path uv.

[0121] When the amount of pheromone update is lower than the update threshold, the best path corresponding to the strongest pheromone signal is output as the planning path.

[0122] When the real-time map and environmental data change, path planning is re-performed.

[0123] In the embodiment, the method for obtaining offset data according to the sensor data, establishing a path compensation strategy according to the offset data, and performing path correction includes the following steps.

[0124] The deviation amount between the actual flight path of the UAV and the planning path is obtained as offset data according to real-time position, attitude data and speed data, including position deviation, attitude deviation and speed deviation. The state equation and observation equation of the UAV are established based on the offset data through the Kalman filtering algorithm, the state of the UAV is continuously estimated, the path compensation strategy is reversely applied based on the state estimation and the offset amount of the planning path, the flight direction and speed of the UAV are adjusted, path correction is performed, and the offset amount of the state estimation and the planning path is reduced.

[0125] In the embodiment, the UAV navigation model is obtained by learning the path compensation strategy and the path planning model, and the method for optimizing the UAV navigation model through the sine-cosine optimization algorithm includes the following steps.

[0126] The UAV navigation model is trained by using a reinforcement learning algorithm. The state space is the real-time position, attitude data, risk score, path time, speed data and environmental data of the UAV, the action space is the adjustment of the flight direction and speed of the UAV, and the reward function is:

[0127]

[0128] wherein k P , k​T , k W , and k θ are a risk score reward coefficient, a path time reward coefficient, an energy consumption reward coefficient, and a path stability reward coefficient, respectively, T expected and T actrual are the path time of the planning path of the path planning model and the actual path time of the reinforcement learning algorithm, respectively, W expected and W actrual are the energy consumption of the planning path and the actual energy consumption of the reinforcement learning algorithm, respectively, and Δθ is the deviation rate of the path direction from the path planning direction of the reinforcement learning algorithm, θ stabel is the deviation rate of the path direction, and θ * is the deviation rate threshold value.

[0129] The reciprocal of the objective function value of the unmanned aerial vehicle navigation model is taken as the fitness value, and the position of each search individual represents a set of hyperparameter values. The search individual position update formula is:

[0130]

[0131] wherein, and are the positions of the i-th search individual at the t+1-th and t-th iterations, respectively, X * is the global optimal position, is the historical optimal position searched by the i-th search individual, ω t is a nonlinear weight factor at the t-th iteration, used to adjust the dependence of the search individual on the current own position information in the population iteration process, is a control parameter of the amplitude of the sine function, t is the current iteration number, t max is the maximum iteration number, r1 is a control parameter, r1=a(1-t / t max ), a is a constant greater than 1, is an adjustment parameter of the frequency of the sine and cosine functions, r2∈(0,2π), is a random weight subject to uniform distribution, r3∈(0,1), is a random number subject to uniform distribution, used to select an individual position update strategy,

[0132] When the average of the fitness values of the historical optimal positions searched by all search individuals does not change for 4 consecutive iterations, it is considered that the search has fallen into stagnation, at which time the position of the search individual is updated using the Levy flight random walk strategy. The random walk step length formula of the Levy flight is:

[0133]

[0134] wherein s is the random walk step length, u and v are both parameters subject to normal distribution, Gamma function, beta is an exponential parameter controlling the shape of the Levy distribution, beta is in (0, 2), take beta = 1.5, sigma u is the standard deviation of u, sigma v is the standard deviation of v.

[0135] The random walk strategy based on Levy flight is:

[0136]

[0137] Where S randn is a random quantity obeying standard normal distribution, alpha is in [-1, 1], is a proportional factor, take alpha = 0.9,

[0138] Output the hyperparameter value corresponding to the global optimal position when the maximum iteration number is reached.

[0139] A big data-based unmanned aerial vehicle navigation positioning system, characterized in that it comprises:

[0140] A collection module acquires position data, sensor data and environment data of the unmanned aerial vehicle, divides a flight area, and acquires spatial information of the flight area using a GIS system.

[0141] A calculation module constructs a real-time map and performs real-time positioning according to the position data and the sensor data using a SLAM algorithm, divides a sub-area of the real-time map, obtains a risk score using cluster analysis, obtains path time between the sub-areas according to the environment data and the risk score, establishes a fitness function according to the risk score and the path time, establishes a path planning model based on the fitness function, obtains offset data according to the sensor data, learns an unmanned aerial vehicle navigation model by learning a path compensation strategy and the path planning model;

[0142] A correction module establishes a path compensation strategy and performs path correction according to the offset data.

[0143] An optimization module optimizes the unmanned aerial vehicle navigation model through a sine-cosine optimization algorithm.

[0144] The above only describes the preferred embodiments of the present application and is not intended to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for navigation and positioning of a UAV based on big data, characterized in that, The method comprises the following steps: Obtain the position data, sensor data and environment data of the UAV, construct a real-time map and perform real-time positioning according to the position data and the sensor data by using a SLAM algorithm, divide the real-time map into sub-regions, and obtain a risk score by using cluster analysis; Obtain the path time between the sub-regions according to the environment data and the risk score, establish a fitness function according to the risk score and the path time, and establish a path planning model based on the fitness function; Obtain offset data according to the sensor data, establish a path compensation strategy according to the offset data, and perform path correction; Learn the path compensation strategy and the path planning model to obtain a UAV navigation model, and optimize the UAV navigation model by using a sine-cosine optimization algorithm.

2. The method of claim 1, wherein, The method for obtaining the position data, sensor data and environment data of the UAV comprises the following steps: The position data of the UAV is obtained by fusing the positions of the UAV obtained by GPS, GLONASS and Galileo through a pseudo-range algorithm, the sensor data comprises the data of a three-dimensional camera, a barometer, a gyroscope, an accelerometer and an inertial measurement unit, the sensor data comprises attitude data, speed data, acceleration data, air pressure data and image data, the environment data comprises meteorological data and electromagnetic interference data, the meteorological data comprises wind direction, wind speed and air pressure, and the electromagnetic interference data comprises the intensity and frequency of electromagnetic interference. 3.The method of claim 1, wherein, The method for constructing a real-time map and performing real-time positioning according to the position data and the sensor data by using a SLAM algorithm comprises the following steps: Obtain a GIS map of the surrounding area of the UAV according to the position data by using a GIS system, construct an obstacle model based on the real-time monitoring data of the three-dimensional camera by using a SLAM algorithm, construct a real-time map based on the GIS map, and obtain the real-time position of the UAV in the real-time map. 4.The method of claim 1, wherein, The method for dividing the real-time map into sub-regions and obtaining a risk score by using cluster analysis comprises the following steps: Distinguish the flyable region and the non-flyable region from the real-time map, divide the flyable region into sub-regions in the form of cubes, Establish data points according to the surrounding obstacle density, wind speed, electromagnetic interference data and air pressure of the sub-regions, and form risk regions by clustering the data points by using a SOM algorithm: Randomly select k data points as the initial weight vector of the competitive layer neurons, one competitive layer neuron represents one cluster center, calculate the similarity between the data points and the weight vector of each competitive layer neuron, and distribute the data points to the cluster of the neuron with the highest similarity, The calculation formula of the similarity is as follows: d(h, e) is a similarity of h and e, h and e are two r-dimensional vectors, ||h(w κ )|| and ||e(w κ )|| are weighted vector lengths of vector h and vector e respectively, h κ represents a component of vector h in the kth dimension, w κ is a weight coefficient of the kth dimension, and a neighborhood centered on a neuron with an initial neighborhood radius of σ(0) is drawn. The dynamic topological neighborhood function in the competitive layer is as follows: where h gi (t) is the dynamic topological neighborhood function of neuron i and neuron g at the tth iteration, d ig (t) is the distance between neuron i and neuron g at the tth iteration, is the similarity between neuron i and g, and σ(t) is the neighborhood radius at the tth iteration, t max is the maximum number of iterations, and δ(t) is the influence factor that changes over time, ξ is the initial value of the influence factor, and ζ is the adjustment coefficient of the influence factor, The update of the weight vector of the neuron is as follows: w ij (t+1) = w ij (t) + η(t) · h gi (t) · (x j -w ij (t) - γ · w ij (t) where γ is the strength of the regularization term, w ij (t) and w ij (t+1) are the jth component of the weight vector of the ith neuron at the tth and t+1th iteration, respectively, η(t) is the learning rate at the tth iteration, x j is the jth component of the data point, Output the cluster result after reaching the maximum number of iterations, and the risk score of each risk region is the average value of the weighted vector length of the data points therein.

5. The method of claim 1, wherein, The method for obtaining the path time between the sub-regions according to the environment data and the risk score comprises the following steps: The path time function is as follows: wherein T cb is the path time between the center of sub-region c and the center of sub-region b, w P is the influence weight of the wind speed, P c and P b are the risk scores of the risk region corresponding to sub-region c and sub-region b respectively, S c and S b are the portions of the path length between the center of sub-region c and sub-region b in sub-region c and sub-region b respectively, V is the speed of the UAV, V f (c) and V f (b) are the wind speeds of sub-region c and sub-region b respectively, φ c and φ b are the clockwise angles between the wind direction and the advancing direction of the UAV in sub-region c and sub-region b respectively, ω f is the influence weight of the wind speed on the speed of the UAV.

6. The method of claim 5, wherein, The method for establishing a fitness function according to the risk score and the path time comprises the following steps: The fitness function is as follows: where f(L) is a fitness function of path L, c e L denotes that sub-region c is in the sub-region set of path L, sub-region c and sub-region b are adjacent, cb e L denotes that sub-region c to b is a sub-path in path L, W cb is the energy consumption of sub-path cb, w T and w W are the weights of path time and energy consumption, respectively, The calculation formula of the energy consumption is as follows: where m and m l are the weight of the UAV and the carried load, respectively, Φ is the gravitational acceleration, ΔH is the height variation, and are the air pressures of the sub-areas c and b, respectively, is the gas constant, and are the temperatures of the sub-areas c and b, respectively, and A is the wind-facing area of the UAV. 7.The method of claim 1, wherein, The method for establishing a path planning model based on the fitness function comprises: each ant moving on a directed graph composed of nodes formed by sub-regions and leaving pheromones, initializing parameters of the ant colony algorithm, and the ant transfer probability between nodes being: Among them is What is the probability that the k-th ant moves from node u to node v at time s? and These are the weights of the pheromone and the heuristic function, respectively. Let be the index of a node in N(u), where N(u) is the set of neighboring nodes of node u. For node u to node heuristic function, For node u to node The distance between the center of subregion u and the subregion is... Path length between centers sub-region Risk score, The pheromone update formula is: where μ uv (s) and μ uv (s+1) represent the amount of pheromone on the path uv at time s and time s+1, respectively, a is the pheromone evaporation coefficient, Z uv is the fitness of the path uv, and is the fitness function value of the path uv. When the pheromone update amount is lower than the update threshold, output the best path corresponding to the strongest pheromone signal as the planning path, When the real-time map and environmental data change, re-perform path planning. 8.The method of claim 1, wherein, The method for obtaining offset data from the sensor data, establishing a path compensation strategy based on the offset data, and performing path correction comprises: The deviation amount between the actual flight path of the UAV and the planning path is obtained as offset data from real-time position, attitude data and speed data, including position deviation, attitude deviation and speed deviation, the state equation and observation equation of the UAV are established based on the offset data through the Kalman filtering algorithm, the state of the UAV is estimated constantly, the path compensation strategy is applied reversely based on the state estimation and the offset amount of the planning path, the flight direction and speed of the UAV are adjusted, path correction is performed, and the offset amount of the state estimation and the planning path is reduced. 9.The method of claim 1, wherein, The UAV navigation model is obtained by learning the path compensation strategy and the path planning model, and the method for optimizing the UAV navigation model through the sine-cosine optimization algorithm comprises: The UAV navigation model is trained by using a reinforcement learning algorithm, the state space is real-time position, attitude data, risk score, path time, speed data and environmental data of the UAV, the action space is the adjustment of the flight direction and speed of the UAV, and the reward function is: wherein k P , k T , k W and k θ are a risk score reward coefficient, a path time reward coefficient, an energy consumption reward coefficient and a path stability reward coefficient, respectively, T expected and T actrual are a path time of the planning path of the path planning model and an actual path time of the reinforcement learning algorithm, respectively, W expected and W actrual are an energy consumption of the planning path and an actual energy consumption of the reinforcement learning algorithm, respectively, Δθ is an offset rate of the path direction and the path planning direction of the reinforcement learning algorithm, and θ stabel is an offset rate threshold of the path direction. The reciprocal of the objective function value of the UAV navigation model is taken as the fitness value, the position of each search individual represents a set of hyperparameter values, and the search individual position update formula is: wherein, and Xi(t+1) and Xi(t) are the positions of the ith search individual at the (t+1)th and tth iteration, respectively, X * Xi(t) is the global optimal position, Xi(t) is the historical optimal position searched by the ith search individual, ω t wtis the nonlinear weight factor at the tth iteration, is a control parameter of the amplitude of the sine function, t is the current iteration number, t max is the maximum iteration number, r1is a control parameter, r1= a(1-t / t max ), a is a constant greater than 1, is an adjustment parameter of the frequency of the sine and cosine functions, r2∈(0, 2π), is a random weight subject to a uniform distribution, r3∈(0, 1), is a random number subject to a uniform distribution, used for selecting an individual position update strategy, When the mean of the fitness values of the historical optimal positions searched by all search individuals does not change for 4 consecutive iterations, it is considered that the search is in stagnation, at this time, the position of the search individual is updated by using the Levy flight random walk strategy, and the Levy flight random walk step formula is: where s is the random walk step size, u and v are both parameters following a normal distribution, Γ(·) is the gamma function, β is an exponential parameter that controls the shape of the Levy distribution, β ∈ (0, 2), taken β = 1.5, σ u is the standard deviation of u, σ v is the standard deviation of v; The random walk strategy based on the Levy flight is: where S randn is a random quantity that obeys the standard normal distribution, a' e [-1, 1], is a scale factor, and a' = 0.9 is taken, When the maximum number of iterations is reached, the hyperparameter value corresponding to the global optimal position is output. 10.A big data based UAV navigation positioning system for performing the big data based UAV navigation positioning method of any one of claims 1-9. It comprises: The acquisition module acquires position data, sensor data and environmental data of the UAV, divides the flight area, and acquires spatial information of the flight area by using a GIS system; The calculation module constructs a real-time map and performs real-time positioning according to the position data and the sensor data by using a SLAM algorithm, divides the real-time map into sub-regions, obtains a risk score by using cluster analysis, obtains path time between the sub-regions according to the environmental data and the risk score, establishes a fitness function according to the risk score and the path time, establishes a path planning model based on the fitness function, obtains offset data from the sensor data, and learns a path compensation strategy and a path planning model to obtain a UAV navigation model; The correction module establishes a path compensation strategy based on the offset data and performs path correction; The optimization module optimizes the UAV navigation model through the sine-cosine optimization algorithm.

Images