Lake overhead unmanned aerial vehicle route planning method and system fused with meteorological data

Abstract

The invention discloses a method and a system for planning a route of an unmanned aerial vehicle over a lake by fusing meteorological data, and belongs to the technical field of unmanned aerial vehicle navigation. The method comprises the following steps: acquiring an initial route according to an unmanned aerial vehicle flight starting place and no-fly zone data; obtaining current meteorological data of an area corresponding to the initial route of the unmanned aerial vehicle, inputting the current meteorological data into the climate risk model, outputting a risk entropy value, and judging whether the unmanned aerial vehicle passes through a high-risk field; meanwhile, the flight safety of the unmanned aerial vehicle is judged according to the stability function; if the initial route passes through the high-risk field and the flight safety of the unmanned aerial vehicle is dangerous, taking a cost function to obtain a minimum value as a target, and optimizing the initial route to obtain a target route; and if the flight safety of the unmanned aerial vehicle is safe, taking the initial route as a target route. And safe flight and compliance control in a complex airspace above a lake are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle navigation, and particularly relates to a lake airspace space unmanned aerial vehicle route dynamic planning method and system combining real-time meteorological data and no-fly zone policy data. BACKGROUND

[0002] With the vigorous development of urban low-altitude economy, the application of unmanned aerial vehicles in scenarios such as "last mile" logistics distribution, urban infrastructure inspection, and emergency medical material transportation presents exponential growth. In the urban environment, the complex underlying surface and the "valley effect" caused by building clusters result in dramatic changes in wind fields, frequent local strong winds, and sudden gusts. In terms of airspace management, the no-fly zones in modern smart cities exhibit multi-level and dynamic characteristics. In addition to permanent airport clearance zones and government agency protection zones, there are also temporary no-fly zones triggered by real-time events, such as major event security zones and sudden accident disposal zones. The establishment and removal of these zones often have only a few hours or even tens of minutes of advance, requiring unmanned aerial vehicle systems to have minute-level dynamic response capabilities. Therefore, the dense and dynamic changes of no-fly zones in urban airspace pose high requirements for the real-time obstacle avoidance and compliant flight of unmanned aerial vehicles.

[0003] In terms of ecological monitoring, unmanned aerial vehicles can quickly and efficiently monitor lake water quality, aquatic organism distribution, and surrounding ecological environment, providing real-time and accurate data support for ecological protection. By carrying high-resolution optical cameras and multispectral sensors, unmanned aerial vehicles can clearly capture color changes in lake water, thereby analyzing the reproduction of algae in the water and promptly discovering issues such as water eutrophication. In the field of low-altitude sightseeing, unmanned aerial vehicles provide tourists with a new visual experience, allowing them to appreciate the natural scenery of lakes from unique perspectives through high-definition pictures taken by unmanned aerial vehicles.

[0004] However, unmanned aerial vehicles face unique challenges when flying over lakes. The open terrain of lake areas and the thermal differences between water and land easily form local circulation, leading to dramatic changes in wind fields. Existing unmanned aerial vehicle path planning methods are mostly based on geometric obstacle avoidance, mainly considering static obstacles and improving search efficiency through heuristic functions, but ignoring the impact of meteorological factors on flight safety. Traditional climate models lack temporal and spatial dynamics and cannot accurately predict sudden meteorological changes over lakes. Although wind speed factors are considered, static safety thresholds are used, and three-dimensional risk fields are not constructed, making it difficult to adapt to complex and variable lake meteorological environments. Therefore, a method that can dynamically evaluate meteorological risks and optimize paths is needed.

[0005] The current mainstream industry solutions have two major technical bottlenecks: First, the meteorological risk assessment of drone routes over lakes is static and lagging. Existing drone systems generally use a judgment logic based on a single wind speed threshold. This simplified model cannot reflect the differences in aerodynamic characteristics of different drone models, nor can it respond to the spatiotemporal variation of wind speed.

[0006] Second, the optimization objective is singular and the decision-making process is fragmented. Traditional cascaded architectures treat weather obstacle avoidance, policy compliance, and path optimization as independent sequential modules. This "obstacle avoidance first, compliance second" or "compliance second, obstacle avoidance third" approach cannot achieve a globally optimal solution. In practical applications, the dilemma of "avoiding strong winds but entering a no-fly zone" or "complying with airspace control but flying into a turbulent area" often occurs. Summary of the Invention

[0007] In response to the above-mentioned deficiencies or improvement needs of existing technologies, this invention provides a method and system for planning UAV flight routes over lakes by integrating meteorological data, thereby solving the key technical problem of existing technologies neglecting safe flight and compliance management in complex airspace over lakes.

[0008] To achieve the above objectives, in a first aspect, the present invention provides a method for planning unmanned aerial vehicle (UAV) flight paths over lakes by integrating meteorological data, comprising: S1. Obtain the start and destination points of the drone flight and the policy no-fly zone data, and obtain the initial flight path based on the drone flight start and destination points and no-fly zone data; S2. Obtain the current meteorological data of the area corresponding to the initial flight path of the UAV, input the current meteorological data into the climate risk model, and output the risk entropy value; when the risk entropy value exceeds a preset threshold: determine that the initial flight path crosses a high-risk field; determine the flight safety of the UAV when flying along the initial flight path according to the stability function; if the flight safety of the UAV is dangerous, execute S3; if the flight safety of the UAV is safe, execute S4. S3. The improved A* algorithm is used to optimize the waypoint nodes on the initial route in turn, using the cost function of each waypoint node. The goal is to minimize the value, determine the next waypoint, and optimize to obtain the target route. The improved A* algorithm features a redesigned cost function. n represents waypoint nodes on the route. Let n be the total path length from the starting point to the current waypoint node n. η is the three-dimensional risk field value at the location of node n, and η is the risk weight coefficient. It is the estimated cost of the distance from waypoint node n to the destination, and μ is the heuristic weight coefficient, which is set to 0.3. S4. Use the initial route as the target route.

[0009] Optionally, the method for calculating the three-dimensional risk field value comprises: constructing a three-dimensional risk field based on a safety wind speed threshold; an expression of the three-dimensional risk field is:

[0010] wherein, is a three-dimensional risk field value at a waypoint node with coordinates ; λ is a wind field gradient weight factor; k is a terrain sheltering attenuation coefficient, is a distance between a waypoint node with coordinates and a terrain sheltering object, W safe is a safety wind speed threshold, W( ) is a wind speed prediction value of a waypoint node with coordinates( ).

[0011] Optionally, the safety wind speed threshold ; wherein, is a reference safety threshold, is a proportion factor, an expression of the proportion factor is: , is a carbon fiber machine system number; is a battery discharge rate coefficient; is an RTK positioning accuracy coefficient; is a weather influence coefficient; , , , is a weight parameter.

[0012] Optionally, the method for judging flight safety of the unmanned aerial vehicle when flying along the initial flight path according to the stability function comprises: calculating a stability quantitative value S(v) of the unmanned aerial vehicle under current meteorological data according to a stability function; if the stability quantitative value S(v) is less than a safety wind speed threshold , it is determined that the flight safety of the unmanned aerial vehicle is dangerous; if the stability quantitative value S(v) is greater than or equal to the safety wind speed threshold , it is determined that the flight safety of the unmanned aerial vehicle is safe; wherein the stability function is , and v is a real-time wind speed.

[0013] Optionally, when taking a minimum value of a cost function of each waypoint node as an objective, determining a next waypoint node, and optimizing to obtain a target flight path, the method further comprises: selecting a point with a lowest three-dimensional risk field value R(n) in an optimization process as a next waypoint node for optimization, so that the flight path bypasses a high risk field.

[0014] Optionally, the initial flight path is obtained according to the starting point of the flight of the unmanned aerial vehicle and the no-fly zone data. Based on the starting point, the destination point and the no-fly zone of the flight of the unmanned aerial vehicle, an initial flight path that does not pass through the no-fly zone is constructed using an A* algorithm.

[0015] Optionally, the climate risk model is a spatio-temporal convolution climate risk model using a ConvLSTM-Transformer hybrid architecture. The climate risk model includes a five-channel input tensor, wherein channel 0 stores real-time wind field data, channel 1 stores historical wind field sequences, channel 2 stores wind field gradient modulus, channel 3 stores terrain elevation data, and channel 4 stores unmanned aerial vehicle parameters.

[0016] Optionally, the system further comprises a verification operation performed after S3, including: obtaining each waypoint on the optimized flight path , calculating the effective distance from the no-fly zone

[0017] wherein, is the effective distance; (x, y) is the waypoint coordinate, and (cx, cy) is the no-fly zone center coordinate; is a coefficient for converting grid resolution to actual distance; r is the radius of the no-fly zone; is a buffer distance; determining whether the target flight path is compliant according to the effective distance ; when , it is determined as potential intrusion, the optimized flight path is non-compliant, the risk weight coefficient in the improved A* algorithm is updated to 1.1η, and 1.1η is less than 1, and the system returns to S3 to optimize the flight path again and verify its compliance.

[0018] In a second aspect, the present application also provides a lake over unmanned aerial vehicle flight path planning system integrating meteorological data, which is used to implement the lake over unmanned aerial vehicle flight path planning method integrating meteorological data according to any one of the first aspect, and includes: An initial flight path planning module is configured to obtain a starting point and a destination point of the flight of the unmanned aerial vehicle, and obtain an initial flight path according to the starting point and the destination point of the flight of the unmanned aerial vehicle. The risk judgment module is configured to acquire current meteorological data of a region corresponding to an initial flight route of the UAV, input the current meteorological data into a climate risk model, and output a risk entropy value; when the risk entropy value exceeds 0.85, it is determined that the initial flight route passes through a high-risk field; at the same time, the flight safety of the UAV flying along the initial flight route is determined according to a stability function; if the initial flight route passes through a high-risk field and the flight safety of the UAV is dangerous, the route optimization module is entered; if the flight safety of the UAV is safe, the route determination module is entered. The route optimization module is configured to optimize the waypoint nodes on the initial flight route in sequence by using an improved A* algorithm, and determine a target flight route by taking the minimum value of a cost function of each waypoint node as a target. The improved A* algorithm is designed by redesigning the cost function, and the cost function is n is a waypoint node on the flight route, is a total length of a path from a starting point to the current waypoint node n, is a three-dimensional risk field value of a position of the node n, and η is a risk weight coefficient, is an estimated cost of a distance from the waypoint node n to a target terminal point, and μ is a heuristic weight coefficient. The route determination module is configured to take the initial flight route as a target flight route.

[0019] Compared with the prior art, the above technical scheme of the present application can achieve the following beneficial effects: 1. The lake overflight UAV route planning method fusing meteorological data provided by the present application can determine whether to perform route optimization by judging whether the initial flight route passes through a high-risk field through a climate risk model and judging the safety and stability of flight in the region by comprehensively considering the safety wind speed threshold. The three-dimensional risk field is constructed, and the three-dimensional risk value of the high-risk field is calculated. The initial flight route is optimized based on the three-dimensional risk value by taking the cost function

[0020] 2. The lake over unmanned aerial vehicle route planning method fusing meteorological data provided by the application constructs a spatio-temporal convolution climate risk model fusing multi-source data, and the system can determine whether the initial route passes through a high-risk field. Real-time wind field, historical wind field sequence, wind field gradient, terrain elevation and unmanned aerial vehicle body parameters are analyzed simultaneously through five-channel input tensors, and a ConvLSTM-Transformer hybrid architecture is used to capture the risk law. The stability function and the safety wind speed threshold are used to evaluate the real-time stability of flight, and the risk assessment is dynamic and timely. The stability function calculates the stability quantitative value of the unmanned aerial vehicle under the current meteorological data, and the safety wind speed threshold calculates the safety bottom line quantitative value under the body factor and the environmental factor. This makes the risk response strategy improve from "passive response" to "active foresight".

[0021] 3. The lake over unmanned aerial vehicle route planning method fusing meteorological data provided by the application uses a dynamic risk field and a cost function to optimize the initial route, which significantly improves the flight safety performance in complex meteorological environments. A three-dimensional risk field is constructed to calculate the three-dimensional risk value of the high-risk field, and a cost function is used as a constraint condition to optimize the initial route based on the risk value, so that the route bypasses the high-risk field. An effective distance algorithm based on a buffer distance is used for online compliance verification, and the optimized route has compliance and safety, ensuring that the optimized route is strictly located outside the buffer zone of the no-fly zone, forming a full-process safety closed loop from risk assessment to route optimization and then to compliance control. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is a whole architecture diagram of the lake over unmanned aerial vehicle route planning method fusing meteorological data provided by the embodiment of the application; Figure 2 is a schematic diagram of the climate risk model used by the embodiment of the application; Figure 3 is a safety wind speed threshold calculation logic diagram in the embodiment of the application; Figure 4 is a path result diagram of the lake over unmanned aerial vehicle route planning method fusing meteorological data provided by the embodiment of the application. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application and do not limit the application. In addition, the technical features involved in each embodiment of the application described below can be combined with each other as long as they do not conflict with each other.

[0024] This invention provides a method for planning UAV flight paths over lakes by integrating meteorological data, comprising: S1. Obtain the start and destination points of the drone flight and the policy no-fly zone data, and obtain the initial flight path based on the drone flight start and destination points and no-fly zone data; S2. Obtain the current meteorological data of the area corresponding to the initial flight path of the UAV, input the current meteorological data into the climate risk model, and output the risk entropy value; when the risk entropy value exceeds a preset threshold: determine that the initial flight path crosses a high-risk field; determine the flight safety of the UAV when flying along the initial flight path according to the stability function; if the flight safety of the UAV is dangerous, execute S3; if the flight safety of the UAV is safe, execute S4. S3. The improved A* algorithm is used to optimize the waypoints on the initial route in turn, using the cost function of each waypoint. The goal is to find the minimum value, determine the next waypoint node, and optimize to obtain the target route. The improved A* algorithm features a redesigned cost function. n represents a node on the flight path. Let n be the total path length from the starting point to the current waypoint node n. η is the three-dimensional risk field value at the location of node n, and η is the risk weight coefficient. It is the estimated cost of the distance from waypoint node n to the destination, and μ is the heuristic weight coefficient, which is set to 0.3. S4. Use the initial route as the target route.

[0025] The specific content includes: initial route generation, climate risk assessment, three-dimensional risk field construction, A* algorithm optimization of routes, and compliance verification.

[0026] (1) Initial route generation The system acquires data on the start and destination points of the drone flight, as well as policy-mandated no-fly zones. The A* algorithm is used for initial flight path planning, with its core being the design of the Manhattan distance heuristic function. The cost function comprehensively considers path length and no-fly zone avoidance, ensuring that the generated flight path meets basic safety requirements at the geometric level.

[0027]

[0028] Where g(n) is the actual cost from the starting point to the current node, and h(n) is the estimated cost of the Manhattan distance from the current node to the target point.

[0029] (2) Climate risk assessment The climate risk model is a spatio-temporal convolutional climate risk model, and the inference architecture adopts a ConvLSTM-Transformer hybrid architecture.

[0030] Further, the mathematical definition of the input tensor is:

[0031] where X: input tensor, represents the multi-dimensional data structure of the entire field; R: real set, represents that the tensor element is a real number; B (BatchSize): batch size, represents the number of samples processed at the same time; T (TimeSteps): time step, reflects the current time state; H (Height): spatial height dimension, represents the number of grid rows; W (Width): spatial width dimension, represents the number of grid columns; C (Channels): channel number, represents the number of different physical characteristics.

[0032] Further, the climate risk model adopts a five-channel input tensor structure, wherein each channel carries specific physical quantity information: Channel 0: stores real-time wind field data, and considers the gust acceleration effect and gust coefficient.

[0033]

[0034] wherein, is the value of the 0th channel at position (i, j); is the basic wind speed (unit: m / s); is the terrain acceleration coefficient, reflecting the modulation effect of terrain on wind speed; is the gust coefficient, reflecting the ratio of instantaneous wind speed to average wind speed, = 1.9; is the turbulence modulation matrix, describing the local turbulence intensity.

[0035] Channel 1: stores historical wind field time series, used to capture the evolution law of the wind field.

[0036]

[0037] wherein, is the historical time window size, representing the number of historical time considered; is the time decay factor, embodying the decay of historical data on the current influence; is the wind speed value of the kth historical time at position (i, j).

[0038] Channel 2: stores the wind field gradient modulus calculated by the central difference method, representing the spatial wind shear intensity, calculated by the central difference method.

[0039]

[0040]

[0041]

[0042] where, is the wind field gradient modulus, representing the rate of change of wind speed in space (unit: s -1 ); is the partial derivative of wind speed in the x direction (east direction); is the partial derivative of wind speed in the y direction (north direction); , is the grid spacing (unit: meters).

[0043] Channel 3: Store terrain elevation data.

[0044]

[0045] where, is the digital elevation model value, representing the elevation of position (i,j) (unit: m); is the measurement error, following a normal distribution ; is the standard deviation of the elevation measurement error Channel 4: Store normalized UAV body parameters.

[0046]

[0047] where, is the carbon fiber machine system number, reflecting the structural strength of the body; is the battery discharge rate coefficient, reflecting the battery load state; is the TK positioning accuracy coefficient, reflecting the accuracy of the navigation system; is the weight coefficient of each parameter, satisfying the normalization condition .

[0048] Further, the inference architecture of the climate risk model adopts a ConvLSTM-Transformer hybrid model:

[0049] where: : Feature map output by the first convolution layer; : ReLU activation function, introducing nonlinearity; : Weight tensor of the first convolution kernel; , , : The size of the convolution kernel in the time, height, and width directions; c: the input channel index (0-4); k: the output channel index; : The bias vector of the first layer of convolution; , , : The size of the convolution kernel in the three directions.

[0050] Further, the training of the climate risk model is based on a dataset of flight data and meteorological data. After training is complete, the model receives the five-channel tensor and sequentially extracts multi-scale spatial features through a 3D convolution layer and captures time dependencies through a ConvLSTM layer.

[0051] Self-attention calculation:

[0052] Where: Q is the Query Matrix; K is the Key Matrix; V is the Value Matrix; is the dot product of the query and the key, which calculates the similarity; is the dimension of the key vector, used to scale the dot product; : The normalization exponential function converts the score into a probability distribution.

[0053] The query key value matrix is obtained by linear transformation of the feature map:

[0054] Where: , , is the learnable linear transformation weight matrix.

[0055] Multi-head attention:

[0056] Where: : The output of the i-th attention head; h: the number of attention heads; Concat: tensor concatenation operation; : The output projection matrix; h: the number of attention heads.

[0057] Feature fusion and compression:

[0058] Where, g is the feature vector after global average pooling; is the point (i, j) risk entropy value of the Transformer output; is the total number of spatial positions.

[0059] Full connection layer transformation:

[0060]

[0061]

[0062] wherein, , is the intermediate layer feature vector; , is the fully connected layer weight matrix; , is the fully connected layer bias vector; is the layer normalization operation; is the final feature vector.

[0063] Further, the meteorological data of the area corresponding to the initial flight path of the UAV is input into the climate risk model, and a risk entropy value continuously distributed between 0 and 1 is output, and the larger the value represents the higher the comprehensive flight risk at the location. The risk entropy is defined as:

[0064]

[0065] wherein, is the risk entropy value, ranging from [0, 1]; is the feature vector; z is the comprehensive risk index; is the risk weight vector; is the risk bias term; exp(x) is the exponential function.

[0066] Further, the risk stability of the key area is confirmed. The stability function of the UAV is obtained by least squares fitting based on virtual wind tunnel test data. S(v) is monotonically decreasing with the increase of v, indicating that the higher the wind speed, the worse the stability. The safety trigger condition is when S(v) < S min , the risk control protocol is triggered. Wherein, S min = 65 is a static safety threshold, and S min can be adjusted through a dynamic model in subsequent calculations.

[0067]

[0068] wherein, S(v) is the stability quantization value of the UAV at wind speed v, and the lower the value represents the worse the stability. v is the real-time wind speed, unit meter / second (m / s). a is a zero wind speed stability benchmark constant, representing the inherent stability of the UAV under ideal conditions. b is a linear decay coefficient, reflecting the degree of linear influence of wind speed on stability. c is a nonlinear disturbance coefficient, capturing nonlinear effects such as turbulence at high wind speeds.

[0069] The parameters are determined by minimizing the sum of squared residuals:

[0070] Normal equations:

[0071] According to the goodness-of-fit indicators, including the coefficient of determination = 0.986, the adjusted = 0.983, the root mean square error RMSE = 2.34, the final stability function is obtained:

[0072] Based on the basic wind speed, the lake center acceleration coefficient and the gust coefficient, the wind speed (V) model coupling model is constructed:

[0073] Wherein, is the basic wind speed; is the lake center acceleration coefficient: is the gust coefficient.

[0074] Further, the safety wind speed threshold is calculated. The safety wind speed threshold is not a fixed value, but a dynamic adjustment of the benchmark threshold by a multi-factor weighted model. The calculation formula of the proportion factor p is:

[0075]

[0076] Wherein, p is a dynamic proportion factor, which comprehensively reflects the adjustment coefficient of the unmanned aerial vehicle body state and environmental influence. is the safety wind speed threshold, the safety boundary value adjusted based on the proportion factor. S base is the benchmark safety threshold, taking the value of 65. K1, K2, K3, K4 are calibration weight parameters, K1 is the machine body coefficient weight, indicating that the machine body structural integrity is the primary determinant of safety, taking the value of 0.95; K2 is the battery load weight, reflecting the potential impact of high power output on stability, taking the value of 0.36; K3 is the positioning accuracy weight, indicating the risk brought by the uncertainty of fine-tuning environment, taking the value of 0.21; K4 is the weather influence weight, indicating the risk brought by the uncertainty of fine-tuning environment, taking the value of 0.15. is the machine body coefficient, reflecting the machine body structural strength, taking a high value (0.90-0.98) for carbon fiber machine body. is the battery discharge rate coefficient, reflecting the battery load state, taking 1.25-1.50 for high-rate discharge. For positioning accuracy coefficient, RTK positioning takes 0.95-0.98, and GNSS positioning takes 0.80-0.85. For weather influence factor, sunny weather takes 0.9, and rainy weather takes 1.1. The influence of UAV parameters on dynamic scale factor is shown in Table 1.

[0077] Table 1 Risk Adaptation

[0078] (3) Construction of three-dimensional risk field Further, based on the principles of fluid mechanics and aerodynamics, a three-dimensional risk field is constructed according to the safety wind speed threshold:

[0079]

[0080]

[0081]

[0082]

[0083]

[0084] wherein, is the grid point, i.e. the waypoint node. R is the three-dimensional risk field value at point , the higher the value, the greater the risk. is the wind field gradient weight factor, controlling the influence of wind shear on risk, taking the value 0.65. is the wind field gradient modulus, unit (s -1 ), reflecting the rate of change of wind speed in the spatial direction, calculated by central difference method. k is the terrain shielding attenuation coefficient, representing the blocking effect of buildings or terrain on the wind field, taking 0.18 for building groups and 0.12 for natural terrain. is the distance from the nearest terrain shelter, unit meter (m). W is the wind speed prediction value at point , unit meter / second (m / s). is the safety wind speed threshold, unit meter / second (m / s). Q is the nearest shelter point.

[0085] (4) Route optimization In route optimization, an improved A* algorithm is used, the core of which is to reconstruct the traditional cost function and incorporate the output results of the three-dimensional risk field model of the previous step into the path cost calculation. The cost function f(n) is redesigned as:

[0086]

[0087]

[0088]

[0089]

[0090] Where f(n) is the total cost of node n, used as the sorting criterion for the priority queue. g(n) is the actual path cost from the starting node to node n, usually based on path length or energy consumption, c( , ) is from node arrive The cost. R(n) is the risk field value at node n, derived from the three-dimensional risk field model. The heuristic weight μ is set to 0.3, which aims to weaken the influence of the heuristic function to avoid premature convergence of the heuristic search to a local optimum in complex risk field environments.

[0091] Furthermore, the risk weighting coefficient η is used to balance path length and flight safety, and its value ranges from [0.5, 1.0]. The optimal solution is determined through Monte Carlo simulation optimization.

[0092]

[0093]

[0094] in, It is a path security indicator; This is a path efficiency metric. , These are the weighting coefficients. + =1. Setting The search interval is [0.5, 1.0], the search step size is 0.05, and the number of Monte Carlo simulations is 1000 per parameter point. The resulting utility function is: The optimal solution η is 0.7.

[0095] Furthermore, during the optimization process, when the A* algorithm encounters a high-risk node, the sharp increase in the value of R(n) causes the total value of f(n) to become extremely large. This node therefore ranks low in the priority queue and will not be prioritized for expansion. Instead, the algorithm will expand nodes that, although slightly farther from the target point and with longer paths, have much lower risk values ​​R(n), so that the route deviates from the three-dimensional risk field to obtain the target route.

[0096] Further, the compliance check ensures that the optimized route meets the requirements.

[0097] An effective distance algorithm based on the buffer distance is adopted to calculate the effective distance of each waypoint on the target route to the restricted area ; according to the effective distance , it is determined whether the target route is compliant; when , it is determined as potential intrusion, the target route is non-compliant, and the target route is adjusted.

[0098] When there is a non-compliant waypoint in the route, the actual distance of the waypoint is calculated by backstepping the effective distance, the grid distance is calculated to determine the position, and the improved A* algorithm is fed back to impose a penalty on the non-compliant waypoint and move in the gradient direction. Specifically, the risk weight coefficient in the improved A* algorithm is updated to 1.1η, and 1.1η is less than 1, and the route optimization is performed again in S3 and the compliance is verified. If the compliance check fails after multiple parameter adjustments, the constraint condition can be relaxed, that is, the buffer distance is reduced. If the automatic adjustment method is still ineffective after the number of adjustments exceeds the preset number of times, the manual path planning is started based on the non-compliant analysis and risk assessment provided by the algorithm.

[0099]

[0100]

[0101]

[0102] wherein, is the effective distance, unit: meter (m), positive value represents safety, and negative value represents potential intrusion. is the grid distance. is the actual distance conversion. , are the waypoint coordinates, grid units. , are the restricted area center coordinates, grid units. is the grid resolution, unit: meter per grid (m / grid). r is the radius of the restricted area, unit: meter (m). is the buffer distance, unit: meter (m), used for compatibility with positioning errors, and 50 m is taken in the embodiment.

[0103] In a specific embodiment, as shown in Figure 1 , a city unmanned aerial vehicle route planning method fusing meteorological data can be summarized as the following steps: S100, environment modeling and meteorological data access; S200, initial route generation;​ S300, multi-modal risk assessment and dynamic decision mechanism; S400, three-dimensional risk field construction and multi-objective route optimization; S500, compliance automatic verification.

[0104] In step S100, environmental data modeling and multi-source meteorological data access are performed.

[0105] S110, geographic space modeling The system constructs a digital model containing a 5km*5km verification airspace, uses the UTM50N coordinate system (projected coordinate system), and discretizes the airspace into a 100*100 grid system with a grid resolution of 50m*50m. The no-fly zone data uses semantic annotation, the center of the lake is G(40, 60), the ecological protection area is centered at coordinates (60, 80) with a radius of 550m, and the urban core area is centered at coordinates (60, 60) with a radius of 350m. Each no-fly zone is set with a 50m buffer distance to accommodate GPS positioning errors and map data errors.

[0106] S120, basic data field construction In the embodiment, the meteorological data part: the basic wind speed is set to 8.2m / s, the gust coefficient is 1.9, and the lake center wind acceleration coefficient is set to 1.3 considering the small surface friction coefficient of the lake center and other open water areas. The underlying surface data part: the average value of the elevation value is 23.8m, the lake center area elevation is-12.5m, the ecological protection area surrounding elevation is 32.4m, and the urban core area surrounding elevation is 18.7m. The unmanned aerial vehicle body performance parameters: the moment of inertia is 0.95, the battery discharge rate is 1.25, and the positioning accuracy is 0.97.

[0107] In step S200, initial route generation based on classical search algorithm.

[0108] The initial route generation uses the classical A* search algorithm, which is guided by the heuristic function. An open set is maintained to gradually expand the search space, and the node with the smallest comprehensive cost f(n) is selected for exploration in each iteration. The algorithm also maintains a CameFrom table and a GScore table to ensure the reconstructability and efficiency of the path. When the algorithm successfully reaches the target point (90, 90), the complete flight route can be reconstructed by backtracking the CameFrom table. The performance calculation result shows that the total length of the initial route is 11.31km, containing 226 waypoints, which provides a qualified initial solution for subsequent optimization.

[0109] Step S300 specifically includes: S310, structured risk model data preprocessing The spatio-temporal convolution climate risk model adopts a five-channel input tensor structure, each channel carrying specific physical quantity information. The input tensor designed by the system has a dimension of 1x1x100x100x5, where the first two dimensions correspond to batch size and time step, respectively. The middle two 100x100 dimensions correspond to the spatial grid, and the last dimension carries five different physical feature channels. According to the known data, the input feature vector is This design not only meets the requirements of convolutional neural networks for input data format, but also maintains the physical integrity of meteorological field data.

[0110] S320, risk inference and decision-making based on deep learning The risk inference process adopts a ConvLSTM-Transformer hybrid architecture. The first step extracts multi-scale spatial features through a 3D convolution layer, resulting in The second step uses a ConvLSTM layer to capture the temporal dependence of wind field time series data, and the hidden state is The third step, through the attention mechanism of Transformer, adaptively focuses on high-risk areas in the spatial domain. The calculation includes all points in the field, where the grid G(40, 60) in the lake center area is taken as an example. According to , the calculation results in =0.35, and the final feature vector is derived as The calculated risk entropy value is as high as 0.923, far exceeding the preset risk threshold of 0.85. Finally, by statistically analyzing all the results, it is found that the maximum risk value is 0.923, the average risk value is 0.621, and the risk values of 58 flight points exceed the alarm threshold of 0.85. Further comparison with the initial route shows that it passes through a high-risk field with high risk entropy.

[0111] S330, safety wind speed threshold analysis considering the state of the unmanned aerial vehicle body According to the analysis results of risk entropy, the lake area is the key analysis object, and its safety threshold needs to be dynamically observed. First, calculate the instantaneous wind speed. In the lake area, the instantaneous wind speed calculation considers three factors: basic wind speed, lake center wind acceleration coefficient, and gust coefficient. The product of the three gives the predicted value of the instantaneous wind speed in this area, which is 20.25 m / s. Second, calculate the stability quantization value. Substitute the instantaneous wind speed into the stability function S(v) to calculate the stability quantization value of the unmanned aerial vehicle, which is -32.01. This value indicates that the unmanned aerial vehicle is in a dynamic instability state at this wind speed. Third, calculate the system safety wind speed threshold. According to the weighted model, calculate the proportion factor ρ=1.7212. Then, based on the baseline safety threshold of 65, the safety wind speed threshold is =111.88. Fourth step, compare the stability quantification value of this point with the safety wind speed threshold, get the lake center point position is at risk, need to optimize the route to avoid unstable risk area.

[0112] In the three-dimensional risk field construction and multi-objective route optimization of step S400 includes: S410, three-dimensional risk field calculation The calculation of constructing three-dimensional risk field adopts global grid traversal method, respectively calculates gradient term, terrain shielding term and standardized wind speed term for each 50m*50m*50m grid unit in airspace. Through exponential function and square operation, effectively enlarge the difference between high-risk area and safe area. The three-dimensional risk field generated by calculation clearly shows that there is a significant risk core in the lake center area, and its maximum risk value reaches 4.21, and its influence range radius is about 300m.

[0113] S420, optimization process and performance balance Path optimization includes calculating each component mathematical result, evaluating nodes, and managing priority queue.

[0114] First step, function component calculation. Cost function reconstruction and node evaluation, actual path cost Use Euclidean distance to determine cumulative path length, heuristic cost Use Manhattan distance to control the influence of heuristic function in total cost, determine target point.

[0115] Second step, high-risk node punishment mechanism. Arrange in ascending order according to the value of f(n), preferentially expand low-cost nodes, and its risk avoidance strategy is to greatly increase the value of f(n) when encountering >2.0 high-risk nodes to guide the algorithm to bypass.

[0116] Third step, path iteration optimization. Preliminary risk avoidance identifies and avoids >3.0 extreme risk area, medium risk processing adjusts the processing strategy for =1.5-3.0 area, path smoothing introduces turning angle constraint, limits maximum turning angle to 45°, reduces greedy search behavior, and increases path smoothness.

[0117] Fourth step, fine-tune safety and efficiency weight, small range optimization, after 5 iterations, converge, verify convergence condition to generate final path.

[0118] Finally, the total length of the generated optimized route is 12.02 km, which is 6.3% longer than the initial route. However, a breakthrough improvement in safety performance is achieved: the maximum risk value decreases from 3.82 to 1.23, a decrease of 67.8%; the average risk value decreases from 2.34 to 0.87, a decrease of 62.8%; the number of high-risk waypoints decreases from 58 to 3, and the avoidance rate reaches 94.8%.

[0119] The compliance automatic verification of step S500 specifically includes: The system performs a full-line scan on the optimized route and applies the effective distance algorithm based on the buffer distance to verify the 226 waypoints. The calculation result shows that the effective distance d eff of all waypoints is positive, which means that all waypoints are located outside the restricted area and its buffer zone. This verifies that the verification method based on the buffer distance can effectively solve the false alarm problem caused by the accumulation of positioning errors.

[0120] The embodiment of the present application provides a kind of city unmanned aerial vehicle route planning method of fusion meteorological data, whether the initial route crosses high-risk field is judged by climate risk model, the safety stability of the flight in this area is judged based on safety wind speed threshold value. Three-dimensional risk field is constructed, and the three-dimensional risk value of high-risk field is calculated. With cost function As constraint condition, the initial route is optimized based on three-dimensional risk value, to make the route bypass high-risk field. Through constraint model, path redundancy is reduced, and calculation efficiency is improved, which supports real-time dynamic re-planning. Safe flight and compliance control in complex airspace in city are realized.

[0121] Embodiment two A kind of lake unmanned aerial vehicle route planning system of fusion meteorological data above, the system is used to realize the method of one of the kind of fusion meteorological data of lake unmanned aerial vehicle route planning above as described in any one of embodiment one, it includes: An initial route planning module is configured to obtain a starting location and a destination location of a flight of the unmanned aerial vehicle, and obtain an initial route based on the starting location and the destination location of the flight of the unmanned aerial vehicle; A risk judgment module is configured to obtain current meteorological data of a region corresponding to the initial route of the unmanned aerial vehicle, input the current meteorological data into a climate risk model, and output a risk entropy value; when the risk entropy value exceeds 0.85, it is determined that the initial route crosses a high-risk field; at the same time, the flight safety of the unmanned aerial vehicle when flying according to the initial route is judged according to a stability function; if the initial route crosses a high-risk field and the flight safety of the unmanned aerial vehicle is dangerous, the route optimization module is turned to; if the flight safety of the unmanned aerial vehicle is safe, the route determination module is turned to; A route optimization module is configured to optimize the initial route based on a cost function The initial route is optimized to obtain a target route with the minimum value as the target, wherein the cost function ; n is a node on the route, is the actual path cost as the basic efficiency constraint, is the three-dimensional risk field value of the position of the node n, and η is a risk weight coefficient, is the distance estimation cost from the node n to the target point, and μ is a heuristic weight coefficient, and the heuristic weight μ is set to 0.3; The route determination module is configured to take the initial route as the target route.

[0122] The lake space unmanned aerial vehicle route planning system fusing meteorological data provided by the embodiments of the present application is used to implement the lake space unmanned aerial vehicle route planning method fusing meteorological data, and has the same beneficial effects.

[0123] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present application, and is not intended to limit the present application, and any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for planning UAV flight paths over lakes by integrating meteorological data, characterized in that, include: S1. Obtain the start and destination points of the drone flight and the policy no-fly zone data, and obtain the initial flight path based on the drone flight start and destination points and no-fly zone data; S2. Obtain the current meteorological data of the area corresponding to the initial flight path of the UAV, input the current meteorological data into the climate risk model, and output the risk entropy value. When the risk entropy value exceeds a preset threshold: it is determined that the initial flight path crosses a high-risk field; the flight safety of the UAV flying along the initial flight path is determined according to the stability function; if the flight safety of the UAV is dangerous, S3 is executed; if the flight safety of the UAV is safe, S4 is executed. S3. The improved A* algorithm is used to optimize the waypoint nodes on the initial route in turn, using the cost function of each waypoint node. The goal is to minimize the value, determine the next waypoint, and optimize to obtain the target route. The improved A* algorithm features a redesigned cost function. n represents waypoint nodes on the route. Let n be the total path length from the starting point to the current waypoint node n. η is the three-dimensional risk field value at the location of node n, and η is the risk weight coefficient. It is the distance estimation cost from waypoint node n to the target destination, and μ is the heuristic weighting coefficient; S4. Use the initial route as the target route.

2. The method as described in claim 1, characterized in that, The expression for the three-dimensional risk field is: in, Coordinates are The three-dimensional risk field value at the waypoint node; λ is the wind field gradient weighting factor; k is the terrain shading attenuation coefficient. coordinates The distance between waypoint nodes and terrain shelters, W safe For the safe wind speed threshold, W( ) is the coordinate ( The predicted wind speed at waypoint nodes.

3. The method as described in claim 2, characterized in that, The safe wind speed threshold ;in, As a baseline safety threshold, The scaling factor is expressed as follows: , For the carbon fiber system number; This refers to the battery discharge rate coefficient. This refers to the RTK positioning accuracy coefficient. This refers to the weather impact coefficient. , , , These are the weight parameters.

4. The method as described in claim 3, characterized in that, The step of determining the flight safety of the UAV when flying along the initial route based on the stability function includes: The stability quantization value S(v) of the UAV under the current meteorological data is calculated based on the stability function; if the stability quantization value S(v) is less than the safe wind speed threshold... If the stability quantification value S(v) is greater than or equal to the safe wind speed threshold, then the flight safety of the drone is determined to be dangerous; If the stability function is found to be true, then the flight safety of the drone is determined to be safe; whereby the stability function is... v represents the real-time wind speed.

5. The method as described in claim 1, characterized in that, In the cost function of each waypoint node The goal is to minimize the target value. When determining the next waypoint and optimizing to obtain the target route, the following steps are also included: During the optimization process, the point with the lowest three-dimensional risk field value R(n) is selected as the next waypoint node for optimization, so that the route detours around the high-risk field.

6. The method as described in claim 1, characterized in that, The process of obtaining the initial flight path based on the drone's flight origin and policy no-fly zone data includes: Based on the starting point, destination, and no-fly zone of the UAV flight, the A* algorithm is used to construct an initial flight path that does not pass through the no-fly zone.

7. The method as described in claim 1, characterized in that, The climate risk model is a spatiotemporal convolutional climate risk model, which adopts a ConvLSTM-Transformer hybrid architecture. The climate risk model includes a five-channel input tensor, where channel 0 stores real-time wind field data, channel 1 stores historical wind field sequences, channel 2 stores wind field gradient modulus, channel 3 stores terrain elevation data, and channel 4 stores UAV parameters.

8. The method as described in claim 1, characterized in that, This also includes the verification operations performed after S3, including: Obtain each waypoint on the target route. Calculate its distance to the no-fly zone Effective distance ;in, (x,y) are the waypoint coordinates. These are the coordinates of the center of the no-fly zone. This is the coefficient used to convert grid resolution to actual distance, where r is the radius of the no-fly zone. For buffer distance; Based on effective distance Determine whether the target route is compliant; when If the condition is met, it is determined to be a potential intrusion, and the optimized route is non-compliant. The risk weight coefficient in the updated and improved A* algorithm is 1.1η, and 1.1η is less than 1. Then, the process is switched to S3 to optimize the route again and verify its compliance.

9. A drone flight path planning system for lakes integrating meteorological data, the system being used to implement the urban drone flight path planning method integrating meteorological data as described in any one of claims 1-8, characterized in that, include: The initial flight path planning module is used to obtain the start and destination points of the UAV flight and the policy no-fly zone data, and to obtain the initial flight path based on the UAV flight start and destination points and no-fly zone data; The risk assessment module is used to acquire the current meteorological data of the area corresponding to the initial flight path of the UAV, input the current meteorological data into the climate risk model, and output the risk entropy value. When the risk entropy value exceeds a preset threshold: it is determined that the initial flight path crosses a high-risk field; the flight safety of the UAV flying along the initial flight path is determined according to the stability function; if the flight safety of the UAV is dangerous, the process is transferred to the flight path optimization module; if the flight safety of the UAV is safe, the process is transferred to the flight path determination module. The route optimization module uses an improved A* algorithm to optimize the waypoint nodes on the initial route sequentially, with a cost function for each waypoint node. The goal is to minimize the value, determine the next waypoint, and optimize to obtain the target route. The improved A* algorithm features a redesigned cost function. n represents waypoint nodes on the route. Let n be the total path length from the starting point to the current waypoint node n. η is the three-dimensional risk field value at the location of node n, and η is the risk weight coefficient. It is the distance estimation cost from waypoint node n to the target destination, and μ is the heuristic weighting coefficient; The route determination module is used to use the initial route as the target route.

Images