Multi-fan cooperative control method for wind resistance test of small and medium-sized unmanned aerial vehicles

Abstract

The invention discloses a multi-fan cooperative control method for a small and medium-sized unmanned aerial vehicle wind resistance test, and the method comprises the steps: calculating a first wind field deviation and a second wind field deviation, carrying out the multivariate fitting, determining a fan superposition effect and an unmanned aerial vehicle interference effect, building an unmanned aerial vehicle wind field model, and determining a representative measurement point and a working fan. Inputting a target test scene into the unmanned aerial vehicle wind field model to obtain initial prediction fan data, operating an unmanned aerial vehicle to obtain a design wind field, calculating a third wind field deviation according to the design wind field and the target wind field, and optimizing the unmanned aerial vehicle wind field model, and inputting the target test scene into the optimized unmanned aerial vehicle wind field model to determine optimized prediction fan data and carry out multi-fan cooperative control. According to the method, the authenticity of wind field simulation and the reliability of a test result can be improved, a new technical path can be provided for wind resistance evaluation and optimization design of small and medium-sized unmanned aerial vehicles, and the method is of great significance to promotion of upgrading of an unmanned aerial vehicle test technology and industrial standardized development.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle testing, and particularly relates to a multi-fan cooperative control method for wind resistance testing of small and medium-sized unmanned aerial vehicles. BACKGROUND

[0002] With the vigorous development of low-altitude economy, small and medium-sized unmanned aerial vehicles are increasingly widely applied in aerial photography, logistics distribution, and forest plant protection, and the flight safety and stability thereof have become the core problem of the industry, and the wind resistance performance as a key indicator for measuring the adaptability of unmanned aerial vehicles to complex environments directly affects the task execution efficiency and equipment survivability, therefore, precise and efficient wind resistance testing of small and medium-sized unmanned aerial vehicles has become an important prerequisite for guaranteeing their large-scale application, by simulating wind field environments of different intensities and directions, the attitude adjustment, power output and track keeping capacity of the unmanned aerial vehicles can be evaluated, which can provide key data support for the structural optimization and control system improvement of the unmanned aerial vehicles, and promote the development of unmanned aerial vehicle technology in a more reliable and intelligent direction.

[0003] At present, wind resistance testing of small and medium-sized unmanned aerial vehicles mostly relies on single fan or fixed fan array, and there are problems such as low wind field simulation precision, poor dynamic response, and difficulty in reproducing real complex airflow environment; at the same time, the airflow superposition effect between fans and the interference of the unmanned aerial vehicle to the wind field are often ignored in traditional methods, resulting in a large deviation between the test results and actual working conditions, therefore, it is urgent to develop a control method that can precisely model the cooperative effect of fans and the wind field interference of the unmanned aerial vehicle. The present application provides a multi-fan cooperative control method for wind resistance testing of small and medium-sized unmanned aerial vehicles, a three-dimensional fan array model is constructed, the fan superposition effect is quantified by fusing monitoring wind field and simulated wind field data, the interference of the unmanned aerial vehicle to the wind field is analyzed in combination with the induced wind field, and a dynamic optimized wind field model of the unmanned aerial vehicle is established based on the comprehensive wind field set, which realizes the precision and intelligence of multi-fan cooperative control, not only can significantly improve the reality of wind field simulation and the reliability of test results, but also provides a new technical path for efficient evaluation and optimal design of the wind resistance performance of small and medium-sized unmanned aerial vehicles, which has important significance for promoting the upgrading of unmanned aerial vehicle testing technology and the development of industry standardization. SUMMARY

[0004] The present application aims to provide a multi-fan cooperative control method for wind resistance testing of small and medium-sized unmanned aerial vehicles.

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

[0006] The present application comprises the following steps:

[0007] The three-dimensional fan array is arranged, and fan data acquisition and monitoring of the wind field are performed; a three-dimensional fan array model is constructed for finite element simulation to obtain a simulated wind field; a first wind field deviation is calculated according to the monitoring wind field and the simulated wind field; and a fan superposition effect is determined by performing multivariate fitting on the first wind field deviation and neighboring fan data; the fan data includes fan positions and fan operating parameters;

[0008] An unmanned aerial vehicle (UAV) is operated in the three-dimensional fan array to obtain an UAV-induced wind field; a second wind field deviation is calculated according to the monitoring wind field and the UAV-induced wind field; and an UAV interference effect is determined by performing multivariate fitting on the second wind field deviation and UAV states;

[0009] A comprehensive wind field set is obtained; an UAV wind field model is constructed according to the comprehensive wind field set, the fan superposition effect, and the UAV interference effect; and representative measuring points and working fans are determined according to an UAV target path; the comprehensive wind field set includes the monitoring wind field, the UAV-induced wind field, the simulated wind field, UAV states, and corresponding fan data;

[0010] An initial predicted fan data is obtained by inputting a target test scenario into the UAV wind field model; and a design wind field is obtained by arranging a three-dimensional fan array and operating an UAV according to the initial predicted fan data; the target test scenario includes working fans, UAV states, and a target wind field;

[0011] A third wind field deviation is calculated according to the design wind field and the target wind field; the UAV wind field model is optimized according to the third wind field deviation; and optimized predicted fan data is obtained by inputting a target test scenario into the optimized UAV wind field model to perform multifan collaborative control.

[0012] Further, the method for determining the fan superposition effect includes:

[0013] The action radius of the fan is set as a fan interval; a three-dimensional fan array is arranged on a test area wall surface according to the fan interval; a monitoring point is randomly selected; the monitoring point is projected onto the wall surface; the fan closest to the wall surface projection point is selected as a main fan; the fans within the action radius of the main fan are selected as neighboring fans; a group of working fans corresponding to the monitoring point is formed by the main fan and the neighboring fans on all wall surfaces; and a Monte Carlo method is used to randomly generate a group of fan operating parameters for the working fans; the fan operating parameters include fan angles and fan rotating speeds;

[0014] The working fans are operated according to the fan operating parameters to generate a wind field in the test area; and wind field data of the monitoring point is collected to form a monitoring wind field; the wind field data includes wind speeds, wind directions, wind types, turbulence intensities, and turbulence types;

[0015] According to the fan position of the working fan and the fan operation parameter, a three-dimensional fan array model is constructed to obtain wind field data of the corresponding monitoring point by CFD finite element simulation to form a simulated wind field, a first wind field deviation is calculated according to the monitoring wind field and the simulated wind field, and a fan superposition effect is determined by multiple fitting according to the monitoring point position, the fan position of the working fan, the fan operation parameter and the first wind field deviation; the fan superposition effect includes a wind speed superposition effect, a wind direction superposition effect and a turbulence intensity superposition effect, and the expression is:

[0016]

[0017] Wherein P is the monitoring point position, ΔV1(P) is the wind speed deviation caused by the wind speed superposition effect of the fan, Δθ1(P) is the wind direction angle deviation caused by the wind direction superposition effect of the fan, ΔI t1 (P) is the turbulence intensity deviation caused by the turbulence intensity superposition effect of the fan at t moment, n is the number of working fans, α1 is the wind speed transmission coefficient, V i is the wind speed of the working fan i, κ1, κ2, κ3 are distance attenuation coefficients, d i is the distance from the working fan i to the monitoring point, α2 is the turbulence interference gain coefficient, θ i is the angle of the working fan i, θ targte is the angle of the monitoring wind field, λ1 is the conflict coefficient, is the turbulence intensity of the working fan i at t moment, φ i is the angle deviation of the working fan i, ξ is the conditional coupling coefficient, wherein α1, α2, κ1, κ2, κ3, λ, η are determined by multiple fitting.

[0018] Further, the method for determining the interference effect of the unmanned aerial vehicle comprises:

[0019] A set of unmanned aerial vehicle states are randomly determined in the corresponding monitoring wind field, and the unmanned aerial vehicle is run to obtain an unmanned aerial vehicle sensing wind field; the unmanned aerial vehicle state includes the unmanned aerial vehicle speed, direction, attitude, hovering mode, size and blade speed;

[0020] A second wind field deviation is calculated according to the monitoring wind field and the unmanned aerial vehicle sensing wind field, and a multiple fitting is performed according to the unmanned aerial vehicle state, the unmanned aerial vehicle sensing wind field and the second wind field deviation to determine the interference effect of the unmanned aerial vehicle; the interference effect of the unmanned aerial vehicle includes a wind speed superposition effect, a wind direction superposition effect and a turbulence intensity superposition effect, and the expression is:

[0021]

[0022] Δd P =‖P-P d ‖

[0023] wherein P is the monitoring point position, ΔV2(P) is the wind speed deviation caused by the wind speed superposition effect of the UAV, Δθ2(P) is the wind direction angle deviation caused by the wind direction superposition effect of the UAV, ΔI t2 (P) is the turbulence intensity deviation caused by the turbulence intensity superposition effect of the UAV at time t, β1, β2, β3 are wind speed interference coefficients, ω r is the rotor speed, D is the rotor diameter, Δd P is the Euclidean distance from the monitoring point P to the center P d of the adjacent wind turbine, γ1, γ2 are distance attenuation coefficients, θ p is the pitch angle of the UAV, Δψ is the angle between the monitoring point and the heading direction of the UAV, V d is the flight speed of the UAV, and ‖P-P k ‖ is the Euclidean distance from the monitoring point P to the kth rotor P k , Γ k is the circulation intensity of the kth rotor, φ k is the azimuth angle of the kth rotor, is the roll angle φ r rate, λ2 is a dynamic response coefficient, τ is a vorticity coupling coefficient, is the background flow field curl, is a curl operator, V tip is the rotor tip line speed, η1, η2, η3 are turbulence gain coefficients, ζ v is the air kinematic viscosity, is the rotor speed gradient module, β1, β2, β3, γ1, γ2, λ2, τ, η1, η2, η3, ζ v are determined by multiple fitting.

[0024] Further, the method for constructing the UAV wind field model comprises:

[0025] composing a comprehensive wind field set by combining the monitoring wind field, the UAV induced wind field, the simulation wind field, the UAV state, and the corresponding wind turbine data, dividing the comprehensive wind field set into a training set and a test set according to a ratio of 6:4 by using a random forest algorithm, training the UAV wind field model by using the training set, and evaluating the performance of the UAV wind field model by using the test set;

[0026] The UAV wind field model comprises an input layer, a first adjustment module, a second prediction adjustment module, and an output layer.

[0027] The first adjustment module is embedded with unmanned aerial vehicle interference effect constraints, and an MDCN multi-modal interference compensation network is used to calculate the unmanned aerial vehicle interference effect according to the unmanned aerial vehicle state and the target wind field, and the target wind field is restored to the initial guess monitoring wind field without considering the unmanned aerial vehicle interference effect according to the unmanned aerial vehicle interference effect. The MDCN multi-modal interference compensation network includes a bottom layer and a fusion layer. The bottom layer processes the unmanned aerial vehicle state feature through three parallel physical constraint convolution channels. The fusion layer uses a vorticity attention mechanism to obtain a vorticity weight to adjust the unmanned aerial vehicle state feature. The unmanned aerial vehicle interference effect is calculated according to the adjusted unmanned aerial vehicle state feature and the corresponding monitoring position. The initial guess monitoring wind field is calculated according to the unmanned aerial vehicle interference effect and the target wind field.

[0028] The second prediction adjustment module is embedded with wind turbine superposition effect constraints. The wind turbine data of the target wind field is generated by iteratively eliminating the wind turbine superposition effect. The specific steps are as follows: receiving the initial guess monitoring wind field output by the first adjustment module, inputting the initial guess monitoring wind field and the corresponding wind turbine position into the wind field graph neural network to obtain the initial guess wind turbine data, calculating the wind turbine superposition effect according to the wind turbine data and the monitoring wind field, calculating the initial guess simulation wind field according to the wind turbine superposition effect and the monitoring wind field, updating the initial guess simulation wind field by taking the initial guess simulation wind field as the initial guess monitoring wind field input in the second iteration, calculating the simulation wind field deviation of the first iteration and the second iteration, inputting the simulation wind field deviation into the gradient inverse propagator to obtain the wind turbine data correction amount, and directly updating the initial guess simulation wind field by using the wind turbine data corrected in the last iteration in the subsequent iteration and calculating the simulation wind field deviation until the simulation wind field deviation is less than the set threshold, and then outputting the wind turbine data of the target wind field. The gradient inverse propagator is a regularized Jacobian pseudo-inverse solver.

[0029] Further, the method for determining the representative measuring points and the working wind turbines comprises:

[0030] The curvature change of the unmanned aerial vehicle three-dimensional target path is calculated, the points where the curvature change is greater than the curvature change threshold are taken as the first measuring points, the straight line distance between each first measuring point is calculated as the measuring point distance, when the measuring point distance is greater than the measuring point distance threshold, the nearest point of the equal division point of the connecting line of the two measuring points to the target path distance is taken as the second measuring point, and the representative measuring points of the target test scene are composed of the first measuring points and the second measuring points.

[0031] The representative measuring points of the target test scene are projected onto each wall surface, the wind turbine closest to each wall surface projection point is selected as the main wind turbine, the wind turbines within the action radius of the main wind turbine are the neighborhood wind turbines, and the union of the main wind turbines and the neighborhood wind turbines of each wall surface projection point of all representative measuring points is taken as the working wind turbines of the target test scene.

[0032] Further, the method for calculating the third wind field deviation comprises:

[0033] The target wind field, the unmanned aerial vehicle state, the representative measuring point and the representative measuring point corresponding working wind turbine are input into the unmanned aerial vehicle wind field model to obtain the initial predicted wind turbine data corresponding to each representative measuring point. The representative measuring point position of the unmanned aerial vehicle is taken as a hovering position on the target path of the unmanned aerial vehicle, and a hovering mode is selected. The design wind field of each representative measuring point is obtained by running the unmanned aerial vehicle according to the initial predicted wind turbine data, the unmanned aerial vehicle state and the target path. The third wind field deviation is obtained by calculating the deviation between the design wind field of each representative measuring point and the target wind field.

[0034] Further, the method for optimizing the unmanned aerial vehicle wind field model according to the third wind field deviation comprises:

[0035] A multi-index objective function is determined according to the third wind field deviation, and the expression is:

[0036]

[0037] Wherein is the multi-index objective function χ k is the weight of the wind field index k, R m is the representative measuring point set, ΔW k is the third wind field deviation of the representative measuring point i, including the wind speed deviation, the wind direction deviation, the turbulence intensity deviation, the spatial gradient deviation and the time stability, τ k is the wind field deviation threshold value;

[0038] The unmanned aerial vehicle wind field model hyperparameter set is defined as a particle swarm, and the MPSO improved particle swarm optimization algorithm is used to optimize the unmanned aerial vehicle wind field model hyperparameters. The specific steps are:

[0039] The population position and particle velocity are initialized, the individual optimal particle and global optimal particle are determined, and the initial value of the multi-index objective function is calculated The particle velocity and position are updated, and the expression is:

[0040]

[0041] Wherein is the velocity update of the particle i in the t+1 iteration, is the position update of the particle i in the t+1 iteration, c1, c2, c3 are learning factors, r1, r2, r3 are random factors in (0, 1), P i is the individual historical optimal position of the particle i, and G is the global historical optimal position, is the gradient of the objective function in the t iteration norm, w t is the dynamic inertia weight, w min is the minimum inertia weight, w max is the maximum inertia weight, is the decay strength, is the rate of change intensity, T is the maximum number of iterations, is the rate of change of the target function for the tth iteration, is the initial value of the target function, is the current parameter vector of the UAV wind field model, and is the difference step size, is the positive disturbance target value, is the negative disturbance target value.

[0042] The individual optimal particle and the global optimal particle are updated, the multi-index target function value is calculated, and the iteration is repeated until the maximum number of iterations is reached or the target function rate of change is less than 0.05% for three consecutive iterations, the optimal UAV wind field model hyperparameter is output, and the UAV wind field model is updated. The optimal prediction wind turbine data is determined by inputting the target test scene into the optimized UAV wind field model to perform multi-turbine collaborative control.

[0043] The beneficial effects of the present application are:

[0044] The present application is a multi-turbine collaborative control method for small and medium-sized UAV wind resistance testing, which has the following technical effects compared with the prior art:

[0045] The present application can improve the data preprocessing capability and enhance the model adaptability in the multi-turbine collaborative control of small and medium-sized UAV wind resistance testing through test experiments, finite element simulation, multiple fitting, model construction and parameter optimization steps, thereby improving the efficiency and precision of the multi-turbine collaborative control of small and medium-sized UAV wind resistance testing. The multi-turbine collaborative control technology of small and medium-sized UAV wind resistance testing is optimized, which can significantly improve the authenticity of wind field simulation and the reliability of test results, provide key data support for the structure optimization and control system improvement of UAV, and provide a new technical path for efficient evaluation and optimization design of small and medium-sized UAV wind resistance performance. It has important significance for promoting the upgrading of UAV testing technology and the development of industry standardization. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 The present application is a multi-turbine collaborative control method for small and medium-sized UAV wind resistance testing, which has the following technical effects compared with the prior art: DETAILED DESCRIPTION

[0047] 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.

[0048] The multi-turbine collaborative control method for small and medium-sized UAV wind resistance testing of the present application includes the following steps:

[0049] As shown in Figure 1 In the present embodiment, the following steps are included:

[0050] The three-dimensional fan array is arranged, and fan data acquisition and monitoring of the wind field are performed; a three-dimensional fan array model is constructed for finite element simulation to obtain a simulated wind field; a first wind field deviation is calculated according to the monitoring wind field and the simulated wind field; and a fan superposition effect is determined by performing multivariate fitting on the first wind field deviation and neighboring fan data; the fan data includes fan positions and fan operation parameters;

[0051] An unmanned aerial vehicle (UAV) is operated in the three-dimensional fan array to obtain an UAV-induced wind field; a second wind field deviation is calculated according to the monitoring wind field and the UAV-induced wind field; and an UAV interference effect is determined by performing multivariate fitting on the second wind field deviation and UAV states;

[0052] A comprehensive wind field set is obtained; an UAV wind field model is constructed according to the comprehensive wind field set, the fan superposition effect, and the UAV interference effect; and representative measuring points and working fans are determined according to an UAV target path; the comprehensive wind field set includes the monitoring wind field, the UAV-induced wind field, the simulated wind field, UAV states, and corresponding fan data;

[0053] An initial predicted fan data is obtained by inputting a target test scenario into the UAV wind field model; and a three-dimensional fan array is arranged and an UAV is operated to obtain a design wind field according to the initial predicted fan data; the target test scenario includes working fans, UAV states, and a target wind field;

[0054] A third wind field deviation is calculated according to the design wind field and the target wind field; the UAV wind field model is optimized according to the third wind field deviation; and an optimized predicted fan data is determined by inputting a target test scenario into the optimized UAV wind field model to perform multifan cooperative control.

[0055] In this embodiment, the method for determining the fan superposition effect includes:

[0056] The action radius of the fan is set as a fan interval; a three-dimensional fan array is arranged on the wall surface of the test area according to the fan interval; a monitoring point is randomly selected; the monitoring point is projected onto the wall surface; the fan closest to the wall surface projection point is selected as a main fan; the fans within the action radius of the main fan are selected as neighboring fans; a group of working fans corresponding to the monitoring point is formed by the main fan and the neighboring fans on all wall surfaces; and a group of fan operation parameters is randomly generated for the working fans by using the Monte Carlo method; the fan operation parameters include fan angles and fan rotation speeds;

[0057] The working fans are operated according to the fan operation parameters to generate a wind field in the test area; and the wind field data of the monitoring point is collected to form a monitoring wind field; the wind field data includes wind speed, wind direction, wind type, turbulence intensity, and turbulence type;

[0058] The wind field data of the corresponding monitoring points is obtained by constructing a three-dimensional fan array model according to the fan positions and fan operation parameters of the working fan, and performing CFD finite element simulation, a first wind field deviation is calculated according to the monitoring wind field and the simulation wind field, and a fan superposition effect is determined by multiple fitting according to the monitoring point position, the fan position of the working fan, the fan operation parameter and the first wind field deviation; the fan superposition effect includes a wind speed superposition effect, a wind direction superposition effect and a turbulence intensity superposition effect, and the expression is:

[0059]

[0060] Wherein P is the monitoring point position, ΔV1(P) is the wind speed deviation caused by the wind speed superposition effect of the fan, Δθ1(P) is the wind direction angle deviation caused by the wind direction superposition effect of the fan, ΔI t1 (P) is the turbulence intensity deviation caused by the turbulence intensity superposition effect of the fan at t moment, n is the number of working fans, α1 is the wind speed transmission coefficient, V i is the wind speed of the working fan i, κ1, κ2, κ3 are distance attenuation coefficients, d i is the distance from the working fan i to the monitoring point, α2 is the turbulence interference gain coefficient, θ i is the angle of the working fan i, θ target is the angle of the monitoring wind field, λ1 is the conflict coefficient, is the turbulence intensity of the working fan i at t moment, φ i is the angle deviation of the working fan i, ξ is the conditional coupling coefficient, wherein α1, α2, κ1, κ2, κ3, λ, η are determined by multiple fitting;

[0061] In actual evaluation, the fan array is formed by uniformly arranging the fans on the basis of the action radius of the fan as the fan interval on the 6 walls of the test area, a monitoring point A is randomly selected, the monitoring point is projected onto 4 walls to determine 4 main fans and 4*4=16 adjacent fans as a group of working fans of the monitoring point A, and a group of fan operation parameters is generated for the 20 working fans in the fan angle range and the fan speed range by using the Monte Carlo method;

[0062] The fan operation parameters of the 25 working fans are set, the wind field is generated in the test area by running the working fans, the wind field data of the monitoring points is collected to form a monitoring wind field, a simulation wind field corresponding to the monitoring points is obtained by constructing a CFD finite element model according to the fan positions and fan operation parameters of the 25 working fans, the first wind field deviation is calculated, the fan superposition effect is determined by multiple fitting according to the monitoring point position, the fan position of the working fan, the fan operation parameter and the first wind field deviation by using Gaussian process regression, and the wind speed transmission coefficient α1, the distance attenuation coefficient κ1 / κ2 / κ3, the conflict coefficient λ1 and the conditional coupling coefficient ξ are determined.

[0063] In the embodiment, the method for determining the UAV interference effect comprises:

[0064] A set of UAV states are randomly determined according to the corresponding monitoring wind field, and the UAV is operated to obtain a UAV-induced wind field; the UAV states comprise UAV speed, direction, attitude, hovering mode, size and blade rotating speed;

[0065] A second wind field deviation is calculated according to the monitoring wind field and the UAV-induced wind field, and the UAV interference effect is determined by multiple fitting according to the UAV states, the UAV-induced wind field and the second wind field deviation; the UAV interference effect comprises wind speed superposition effect, wind direction superposition effect and turbulence intensity superposition effect, and the expression is:

[0066]

[0067] Δd P =‖P-P d ‖

[0068] wherein P is the monitoring point position, ΔV2(P) is the wind speed deviation caused by the UAV wind speed superposition effect, Δθ2(P) is the wind direction angle deviation caused by the UAV wind direction superposition effect, ΔI t2 (P) is the turbulence intensity deviation caused by the UAV turbulence intensity superposition effect at t moment, β1, β2, β3 are wind speed interference coefficients, ω r is the rotor rotating speed, D is the rotor diameter, Δd P is the Euclidean distance from the monitoring point P to the center P d of the adjacent wind turbine, γ1, γ2 are distance attenuation coefficients, θ p is the UAV pitch angle, Δψ is the angle between the monitoring point and the UAV heading direction, V d is the UAV flight speed, ‖P-P k ‖ is the Euclidean distance from the monitoring point P to the kth rotor P k , Γ k is the kth rotor circulation intensity, φ k is the kth rotor azimuth angle, is the roll angle φ r change rate, λ2 is a dynamic response coefficient, τ is a vorticity coupling coefficient, is the background flow field curl, is the curl operator, V tip is the rotor tip speed, η1, η2, η3 are turbulence gain coefficients, ζ v is the air kinematic viscosity, is the rotor rotating speed gradient module, β1, β2, β3, γ1, γ2, λ2, τ, η1, η2, η3, ζ v are determined by multiple fitting;

[0069] In the actual evaluation, the fan operation parameters of the working fan 25 are set, the working fan generates a wind field in the test area, the wind field data of the monitoring points are collected to form a monitoring wind field, appropriate sizes of the unmanned aerial vehicle are selected, a group of unmanned aerial vehicle states are obtained by randomly combining the unmanned aerial vehicle speed, direction, attitude, hovering mode and blade speed, and the unmanned aerial vehicle is operated in the corresponding monitoring wind field to obtain an unmanned aerial vehicle induced wind field; and the second wind field deviation is calculated through the unmanned aerial vehicle induced wind field and the monitoring wind field.

[0070] The multi-task learning neural network is adopted to determine the unmanned aerial vehicle interference effect according to the unmanned aerial vehicle state, the unmanned aerial vehicle induced wind field and the second wind field deviation; the multi-task learning neural network includes a shared bottom feature extraction layer (processing the unmanned aerial vehicle state parameters) and a parallel output layer; the parallel output layer includes a wind speed interference branch (fitting and determining wind speed interference coefficients β1 / β2 / β3, distance attenuation coefficients γ1 / γ2), a wind direction interference branch (fitting and determining dynamic response coefficients λ2, vorticity coupling coefficients τ), and a turbulence intensity interference branch (fitting and determining turbulence gain coefficients η1 / η2 / η3, air movement viscosity ζ v ); the parallel output layer adopts an activation function to constrain the wind speed interference (interference intensity > 0) and the wind direction interference (deviation angle ± 90°), and a loss function adds a fluid continuity equation residual term:

[0071]

[0072] wherein is the loss function of the parallel output layer, y pred is the predicted fitting coefficient, y true is the actual fitting coefficient, λ3=0.15 is the residual coupling strength, is the continuous residual of the wind speed deviation ΔV2 at the spatial position (x, y, z).

[0073] In this embodiment, the method for constructing the unmanned aerial vehicle wind field model comprises:

[0074] The monitoring wind field, the unmanned aerial vehicle induced wind field, the simulation wind field, the unmanned aerial vehicle state and the corresponding fan data are combined to form a comprehensive wind field set, the comprehensive wind field set is divided into a training set and a test set according to a ratio of 6:4 by using a random forest algorithm, the unmanned aerial vehicle wind field model is trained by using the training set, and the performance of the unmanned aerial vehicle wind field model is evaluated by using the test set;

[0075] The unmanned aerial vehicle wind field model includes an input layer, a first adjustment module, a second prediction adjustment module and an output layer.

[0076] The first adjustment module is embedded with unmanned aerial vehicle interference effect constraints, and an MDCN multi-modal interference compensation network is used to calculate the unmanned aerial vehicle interference effect according to the unmanned aerial vehicle state and the target wind field, and to restore the target wind field to the initial guess monitoring wind field without considering the unmanned aerial vehicle interference effect according to the unmanned aerial vehicle interference effect. The MDCN multi-modal interference compensation network comprises a bottom layer and a fusion layer. The bottom layer processes the unmanned aerial vehicle state features through three parallel physical constraint convolution channels. The fusion layer uses a vorticity attention mechanism to obtain vorticity weights to adjust the unmanned aerial vehicle state features. The unmanned aerial vehicle interference effect is calculated according to the adjusted unmanned aerial vehicle state features and the corresponding monitoring position. The initial guess monitoring wind field is calculated according to the unmanned aerial vehicle interference effect and the target wind field.

[0077] The second prediction adjustment module is embedded with wind turbine superposition effect constraints. The wind turbine data of the target wind field is generated by iteratively eliminating the wind turbine superposition effect. The specific steps are as follows: receiving the initial guess monitoring wind field output by the first adjustment module, inputting the initial guess monitoring wind field and the corresponding wind turbine position into a wind field graph neural network to obtain initial guess wind turbine data, calculating the wind turbine superposition effect according to the wind turbine data and the monitoring wind field, calculating the initial guess simulation wind field according to the wind turbine superposition effect and the monitoring wind field, updating the initial guess simulation wind field by taking the initial guess simulation wind field as the initial guess monitoring wind field input in the second iteration, calculating the simulation wind field deviation of the first iteration and the second iteration, inputting the simulation wind field deviation into a gradient inverse propagator to obtain the wind turbine data correction amount, and directly updating the initial guess simulation wind field by using the wind turbine data corrected in the last iteration in subsequent iterations and calculating the simulation wind field deviation until the simulation wind field deviation is less than a set threshold, and then outputting the wind turbine data of the target wind field. The gradient inverse propagator is a regularized Jacobian pseudo-inverse solver.

[0078] In actual evaluation, the three physical constraint convolution channels of the bottom layer of the MDCN multi-modal interference compensation network in the first adjustment module are respectively used for rotor momentum feature extraction, attitude dynamic feature extraction, and motion-induced feature extraction. The vorticity attention mechanism is as follows:

[0079]

[0080] Δ drone =∑δ i f i

[0081] wherein δ i is the attention weight of the i-th constraint convolution channel, Γ is a learnable vorticity coupling matrix, is a curl operator, f i is the unmanned aerial vehicle state feature of the i-th constraint convolution channel, and Δ drone is the unmanned aerial vehicle interference effect compensation field.

[0082] In the second prediction adjustment module, the wind field graph neural network is specifically an FC-GNN fluid constraint graph network, which is constructed by a simulated wind field simulated by CFD finite element simulation and corresponding wind turbine data, and includes an aggregation layer, an N-S equation embedding layer and a residual constraint layer. The N-S equation embedding layer is used to perform N-S equation physical constraint on the feature tensor representing the local flow field state input by the aggregation layer, so that the neural network learns the wind field characteristics conforming to the basic law of fluid mechanics. The wind field graph neural network adopts alternating optimization of data loss and physical loss as a training strategy, and the expression is as follows:

[0083]

[0084] Wherein is the N-S equation residual, that is, the N-S equation physical constraint condition, v is the wind speed vector field, is a gradient operator, is a Laplace operator, ζ v is the kinematic viscosity of air, S pred is the network predicted wind turbine data, S true is the real wind turbine data;

[0085] The second prediction adjustment module adopts a regularized Jacobian pseudo-inverse solver to solve the wind turbine parameter correction amount (that is, the wind turbine parameter adjustment direction) that minimizes the wind field residual according to the wind field residual and the current wind turbine data, and the expression is as follows:

[0086]

[0087] Wherein ΔF is the wind turbine parameter correction amount, J is the Jacobian matrix, μ is the regularization strength, R is the regularization matrix, R ij is the regularization matrix element, L i is the position coordinate of the wind turbine i, θ i is the yaw angle of the wind turbine i, d0 is the characteristic distance scale, is the initial guess simulation wind field of the tth iteration.

[0088] In the embodiment, the method for determining the representative measurement points and the working wind turbine includes:

[0089] The curvature change of the three-dimensional target path of the unmanned aerial vehicle is calculated, points at which the curvature change is greater than a curvature change threshold are taken as first measurement points, straight line distances between the first measurement points are calculated as measurement point distances, when the measurement point distance is greater than a measurement point distance threshold, the nearest point of the target path distance between two measurement points to the bisection point of the line connecting the two measurement points is taken as a second measurement point, and the first measurement points and the second measurement points constitute the representative measurement points of the target test scene;

[0090] The representative measuring points of the target test scene are projected on each wall surface, the fan closest to each wall surface projection point is selected as the main fan, the fans within the action radius of the main fan are the neighborhood fans, and the union of the main fan and the neighborhood fans of each wall surface projection point of all representative measuring points is taken to obtain the working fan of the target test scene;

[0091] In actual evaluation, taking a certain wind-resistant test three-dimensional target path of an X brand Y model unmanned aerial vehicle as an example, the curvature change rate threshold is 0.15 m -1 / rad, the sampling interval is 0.1 rad, the first measuring point number is determined to be 10, the measuring point distance threshold is taken to be 1.2 m, the straight line distance between each measuring point is calculated, the second measuring point number is determined to be 8 (between the first measuring points 5 and 6, the straight line distance is 2.8 m, so the two points closest to the 3 equal points are taken as the second measuring points), and 18 representative measuring points are formed.

[0092] In the test area with a wall surface size of 8*16 m (an 8*8*16 m cuboid test bin), the fans with an action radius of 1.6 m are uniformly arranged (the number of single-wall surface test fans is 36), and the 18 representative measuring points are projected on each wall surface to determine 96 working fans.

[0093] In this embodiment, the method for calculating the third wind field deviation comprises:

[0094] The target wind field, the unmanned aerial vehicle state, the representative measuring points and the representative measuring point corresponding working fan are input into the unmanned aerial vehicle wind field model to obtain the initial predicted fan data corresponding to each representative measuring point, the representative measuring point position of the unmanned aerial vehicle is taken as the hovering position on the target path of the unmanned aerial vehicle and the hovering mode is selected, the design wind field of each representative measuring point is obtained according to the initial predicted fan data, the unmanned aerial vehicle state and the target path, the deviation between the design wind field of each representative measuring point and the target wind field is calculated to obtain the third wind field deviation.

[0095] In the actual evaluation, taking a certain anti-wind test three-dimensional target path of an X brand Y model unmanned aerial vehicle as an example, the target test scene (wind speed / wind direction / turbulence intensity) is input into the unmanned aerial vehicle wind field model to obtain the initial predicted fan data (fan angle and fan speed) corresponding to each representative measuring point, the hovering mode (hovering mode is infrared dynamic capture hovering, and hovering time is 30s) is determined, the three-dimensional fan array is set according to the initial predicted fan data, and the unmanned aerial vehicle is run to obtain the design wind field of each representative measuring point. The working fan operation process is as follows: according to the flight speed of the unmanned aerial vehicle, the target path and the hovering time, the time when the unmanned aerial vehicle reaches and leaves each representative measuring point is determined, when there is no common working fan between adjacent representative measuring points, the working fan of the next measuring point is started and adjusted to the preset fan angle and fan speed when the unmanned aerial vehicle leaves the last representative measuring point, when there are some common working fans between adjacent representative measuring points, the fan change rate (fan angle rotation speed and direction, fan speed change rate) of each flight section is calculated according to the flight time (the difference between the arrival time of the next representative measuring point and the departure time of the last representative measuring point) of the unmanned aerial vehicle between the two representative measuring points and the predicted fan data change (fan angle change and fan speed change) of the same working fan between adjacent representative measuring points, and the working fan of the corresponding part is adjusted according to the fan change rate of each flight section, and the non-common fan of the last measuring point is closed when the unmanned aerial vehicle reaches the next measuring point.

[0096] The deviation of each representative measuring point design wind field and target wind field is calculated to obtain the third wind field deviation.

[0097] In this embodiment, the method for optimizing the unmanned aerial vehicle wind field model according to the third wind field deviation comprises:

[0098] A multi-index objective function is determined according to the third wind field deviation, and the expression is:

[0099]

[0100] Wherein is the multi-index objective function χ k is the weight of the wind field index k, R m is the representative measuring point set, ΔW k is the third wind field deviation of the representative measuring point i, including wind speed deviation, wind direction deviation, turbulence intensity deviation, spatial gradient deviation and time stability, τ k is the wind field deviation threshold value;

[0101] The unmanned aerial vehicle wind field model hyperparameter set is defined as a particle swarm, and the MPSO improved particle swarm optimization algorithm is used to optimize the unmanned aerial vehicle wind field model hyperparameters, and the specific steps are as follows:

[0102] Initialize the population position and particle velocity, determine the individual optimal particle and global optimal particle, and calculate the initial value of the multi-index objective function The particle velocity and position are updated, and the expression is:

[0103]

[0104] wherein is the velocity update of the particle i in the t+1 iteration, is the position update of the particle i in the t+1 iteration, c1, c2, c3 are learning factors, r1, r2, r3 are random factors in (0, 1), P i is the individual historical optimal position of the particle i, and G is the global historical optimal position, is the gradient of the target function in the t iteration norm, w t is the dynamic inertia weight, w min is the minimum inertia weight, w max is the maximum inertia weight, is the decay strength, is the change rate strength, and T is the maximum iteration number, is the change rate of the target function in the t iteration, is the initial value of the target function, Θ is the current parameter vector of the UAV wind field model, and δ is the difference step, is the positive disturbance target value, is the negative disturbance target value;

[0105] The individual optimal particle and the global optimal particle are updated, the multi-index target function value is calculated, the iteration is repeated until the maximum iteration number is reached or the change rate of the target function in three consecutive iterations is less than 0.05%, the optimal UAV wind field model hyperparameter is output, and the UAV wind field model is updated, the target test scene is input into the optimized UAV wind field model to determine the optimized prediction wind turbine data for multi-wind turbine cooperative control.

[0106] In actual evaluation, the wind field index weight (wind speed / wind direction / turbulence intensity / space gradient / time stability) is taken as 0.35, 0.25, 0.20, 0.15, 0.05, the wind field deviation threshold is 2.0 m / s, 15°, 8%, 3 (m / s) / m, 0.5 Hz, and the initial value of the target function is calculated is 11.232;

[0107] In the particle swarm optimization, the learning factors c1 / c2 / c3 are taken as 1.7 / 1.5 / 0.8, the inertia weight range is (0.3, 0.6), the decay strength is 0.05, and the change rate strength For 0.2, the maximum number of iterations T is 50, the difference step size δ is 0.01, the particle swarm search is performed, when the iteration is to the 23th time, the target function change rates of 21 / 22 / 23 consecutive three iterations are 0.046, 0.0441, 0.042 respectively, which are less than the target function change rate threshold 0.05%, at this time, the optimal unmanned aerial vehicle wind field model super parameter is output according to the latest particle position and the unmanned aerial vehicle wind field model is updated, the target test scene (target wind field wind speed / direction / turbulence intensity, unmanned aerial vehicle path) is input into the optimized unmanned aerial vehicle wind field model to determine the optimized prediction wind machine data for multi-wind machine cooperative control.

[0108] The above merely 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 principle of the present application shall be included in the protection scope of the present application.

Claims

1. A multi-wind turbine collaborative control method for wind resistance testing of small and medium-sized UAVs, characterized in that, Includes the following steps: S1. Arrange a three-dimensional wind turbine array and set up wind turbine data acquisition to monitor the wind field. Construct a three-dimensional wind turbine array model and perform finite element simulation to obtain a simulated wind field. Calculate the first wind field deviation based on the monitored wind field and the simulated wind field. Perform multivariate fitting on the first wind field deviation and the neighboring wind turbine data to determine the wind turbine superposition effect. The wind turbine data includes wind turbine location and wind turbine operating parameters. S2. Run a drone in a three-dimensional wind turbine array to obtain the drone-sensed wind field. Calculate the second wind field deviation based on the monitored wind field and the drone-sensed wind field. Perform multivariate fitting on the second wind field deviation and the drone state to determine the drone interference effect. S3. Obtain a comprehensive wind field set, construct a UAV wind field model based on the comprehensive wind field set, the wind turbine superposition effect, and the UAV interference effect, and determine representative measurement points and working wind turbines based on the UAV target path; the comprehensive wind field set includes the monitored wind field, the UAV-sensed wind field, the simulated wind field, the UAV status, and the corresponding wind turbine data. S4. Input the target test scenario into the UAV wind field model to obtain initial predicted wind turbine data, set up a three-dimensional wind turbine array based on the initial predicted wind turbine data, and run the UAV to obtain the designed wind field; the target test scenario includes the working wind turbine, UAV status, and target wind field. S5. Calculate the third wind field deviation based on the designed wind field and the target wind field, optimize the UAV wind field model based on the third wind field deviation, input the target test scenario into the optimized UAV wind field model to determine the optimized predicted wind turbine data for multi-wind turbine collaborative control.

2. The multi-wind turbine collaborative control method for wind resistance testing of small and medium-sized UAVs according to claim 1, characterized in that, The method for determining the superposition effect of wind turbines includes: The effective radius of the fans is set as the fan interval. A three-dimensional fan array is arranged on the wall of the test area according to the fan interval. Monitoring points are randomly selected and projected onto the wall. The fan with the closest projection point on the wall is selected as the main fan, and the fans within the effective radius of the main fan are designated as neighboring fans. A group of working fans corresponding to the monitoring points is composed of all the main fans and neighboring fans on the wall. A set of fan operating parameters is randomly generated for the working fans using the Monte Carlo method. The fan operating parameters include the fan angle and the fan speed. The working wind turbine is operated according to the wind turbine operating parameters to generate a wind field in the test area, and the wind field data collected from the monitoring points constitutes the monitoring wind field; the wind field data includes wind speed, wind direction, wind type, turbulence intensity and turbulence type; A three-dimensional wind turbine array model is constructed based on the location and operating parameters of the working wind turbines. CFD finite element simulation is then performed to obtain wind field data for the corresponding monitoring points, forming a simulated wind field. The first wind field deviation is calculated based on the monitored and simulated wind fields. Multivariate fitting is then performed based on the monitoring point locations, the working wind turbine locations, the wind turbine operating parameters, and the first wind field deviation to determine the wind turbine superposition effect. This superposition effect includes wind speed superposition, wind direction superposition, and turbulence intensity superposition, expressed as: Where P is the location of the monitoring point, ΔV1(P) is the wind speed deviation caused by the superposition effect of wind turbine wind speed, Δθ1(P) is the wind direction angle deviation caused by the superposition effect of wind turbine wind direction, and ΔI t1 (P) represents the turbulence intensity deviation caused by the superposition effect of turbulence intensity at time t, n is the number of operating fans, α1 is the wind speed transfer coefficient, and V i Let be the wind speed of working fan i, κ1, κ2, and κ3 be the distance attenuation coefficients, and d be the wind speed of working fan i. i α is the distance from the working fan i to the monitoring point, α2 is the turbulence interference gain coefficient, and θ is the distance from the working fan i to the monitoring point. i Let θ be the angle of the working fan i. target To monitor the angle of the wind field, λ1 is the conflict coefficient. Let φ be the turbulence intensity at time t of the working fan i. i Let ξ be the angular deviation of the working fan i, and ξ be the conditional coupling coefficient, where α1, α2, κ1, κ2, κ3, λ, and η are determined by multivariate fitting.

3. The multi-wind turbine collaborative control method for wind resistance testing of small and medium-sized UAVs according to claim 1, characterized in that, The method for determining the interference effect of the unmanned aerial vehicle (UAV) includes: The system is randomly determining the status of a set of drones corresponding to the monitored wind field and running the drones to acquire the wind field sensed by the drones; the drone status includes the drone speed, direction, attitude, hovering mode, size and blade speed; The second wind field deviation is calculated based on the monitored wind field and the wind field sensed by the UAV. The UAV interference effect is determined by multivariate fitting based on the UAV status, the UAV-sensed wind field, and the second wind field deviation. The UAV interference effect includes wind speed superposition effect, wind direction superposition effect, and turbulence intensity superposition effect, expressed as: Δd P =‖P-P d ‖ Where P is the location of the monitoring point, ΔV2(P) is the wind speed deviation caused by the superposition effect of the drone's wind speed, Δθ2(P) is the wind direction angle deviation caused by the superposition effect of the drone's wind direction, and ΔI t2 (P) represents the turbulence intensity deviation caused by the superposition effect of UAV turbulence intensity at time t, β1, β2, and β3 are wind speed interference coefficients, and ω r Where Δd is the rotor speed, D is the rotor diameter, and Δd is the rotor diameter. P From monitoring point P to the neighboring wind turbine center P d The Euclidean distance, γ1 and γ2 are the distance attenuation coefficients, and θ p V is the pitch angle of the UAV, Δψ is the angle between the monitoring point and the UAV's heading, and V d For the drone's flight speed, ||PP k ‖ is the distance from monitoring point P to the k-th rotor P k Euclidean distance at Γ k Let φ be the circulation intensity of the k-th rotor. k Let k be the azimuth angle of the rotor. The roll angle φ r The rate of change, λ² is the dynamic response coefficient, and τ is the vorticity coupling coefficient. For the background flow field curl, For the curl operator, V tip η is the blade tip linear velocity, η1, η2, and η3 are turbulence gain coefficients, and ζ is the turbulence gain coefficient. v The viscosity of air motion. is the rotor speed gradient mode, β1, β2, β3, γ1, γ2, λ2, τ, η1, η2, η3, ζ v Determined through multivariate fitting.

4. The multi-wind turbine collaborative control method for wind resistance testing of small and medium-sized UAVs according to claim 1, characterized in that, The method for constructing a drone wind field model includes: A comprehensive wind field set is composed of monitored wind field, UAV-sensed wind field, simulated wind field, UAV status and corresponding wind turbine data. The comprehensive wind field set is divided into a training set and a test set in a 6:4 ratio using the random forest algorithm. The training set is used to train the UAV wind field model, and the test set is used to evaluate the performance of the UAV wind field model. The UAV wind field model includes an input layer, a first adjustment module, a second prediction adjustment module, and an output layer. The first adjustment module embeds a constraint on UAV interference effect. It uses the MDCN multimodal interference compensation network to calculate the UAV interference effect based on the UAV status and the target wind field. Based on the UAV interference effect, it restores the target wind field to the initial guessed monitoring wind field without considering the UAV interference effect. The MDCN multimodal interference compensation network includes a bottom layer and a fusion layer. The bottom layer processes the UAV status features through three parallel physical constraint convolution channels. The fusion layer uses a vortex attention mechanism to obtain vortex weights to adjust the UAV status features. Based on the adjusted UAV status features and the corresponding monitoring position, it calculates the UAV interference effect. Based on the UAV interference effect and the target wind field, it calculates the initial guessed monitoring wind field. The second prediction and adjustment module embeds a constraint on the wind turbine superposition effect. It iteratively eliminates the wind turbine superposition effect and predicts the wind turbine data for the target wind field. The specific steps are as follows: receiving the initial guess of the monitored wind field output by the first adjustment module, inputting the initial guess of the monitored wind field and the corresponding wind turbine location into the wind field graph neural network to obtain the initial guess of the wind turbine data, calculating the wind turbine superposition effect based on the wind turbine data and the monitored wind field, calculating the initial guess of the simulated wind field based on the wind turbine superposition effect and the monitored wind field, updating the initial guess of the simulated wind field as the initial guess of the monitored wind field input for the second iteration, calculating the simulated wind field deviation between the first and second iterations, inputting the simulated wind field deviation into the gradient inverse propagator to obtain the wind turbine data correction amount, and directly using the wind turbine data corrected in the previous iteration to update the initial guess of the simulated wind field and calculate the simulated wind field deviation in subsequent iterations until the simulated wind field deviation is less than a set threshold, at which point the wind turbine data for the target wind field is output; the gradient inverse propagator is a regularized Jacobi pseudo-inverse solver.

5. The multi-wind turbine collaborative control method for wind resistance testing of small and medium-sized UAVs according to claim 1, characterized in that, The method for determining representative measuring points and working fans includes: Calculate the curvature change on the three-dimensional target path of the UAV, and take the point where the curvature change is greater than the curvature change threshold as the first measurement point. Calculate the straight-line distance between each first measurement point as the measurement point distance. When the measurement point distance is greater than the measurement point distance threshold, take the point that is closest to the point where the target path distance between the two measurement points is divided equally by the line connecting the two measurement points as the second measurement point. The first measurement point and the second measurement point constitute the representative measurement points of the target test scene. Representative test points of the target test scenario are projected onto each wall surface. The fan closest to each wall surface projection point is selected as the main fan, and the fans within the effective radius of the main fan are selected as the neighboring fans. The working fans of the target test scenario are obtained by taking the union of the main fans and neighboring fans of all representative test points at each wall surface projection point.

6. The multi-wind turbine collaborative control method for wind resistance testing of small and medium-sized UAVs according to claim 1, characterized in that, The method for calculating the third wind field deviation includes: The target wind field, UAV status, representative measuring points, and corresponding working wind turbines of the representative measuring points are input into the UAV wind field model to obtain the initial predicted wind turbine data corresponding to each representative measuring point. The position of the UAV representative measuring point is used as the hovering position on the UAV target path and the hovering mode is selected. Based on the initial predicted wind turbine data, UAV status, and target path, the UAV is run to obtain the design wind field of each representative measuring point. The deviation between the design wind field of each representative measuring point and the target wind field is calculated to obtain the third wind field deviation.

7. The multi-wind turbine collaborative control method for wind resistance testing of small and medium-sized UAVs according to claim 1, characterized in that, The method for optimizing the UAV wind field model based on the third wind field deviation includes: The multi-index objective function is determined based on the third wind field deviation, and its expression is: in For the multi-index objective function χ k R represents the weight of the wind field index k. m For a representative set of measuring points, ΔW k The third wind field deviation at representative measurement point i includes wind speed deviation, wind direction deviation, turbulence intensity deviation, spatial gradient deviation, and time stability, τ. k This is the wind field deviation threshold; The hyperparameter set of the UAV wind field model is defined as a particle swarm optimization algorithm. The MPSO-improved particle swarm optimization algorithm is used to optimize the hyperparameters of the UAV wind field model. The specific steps are as follows: Initialize population positions and particle velocities, determine the optimal individual particle and the global optimal particle, and calculate the initial values ​​of the multi-index objective function. The expression for updating particle velocity and position is: in For the velocity update of particle i in iteration t+1, For the position update of particle i in iteration t+1, c1, c2, and c3 are learning factors, and r1, r2, and r3 are random factors in (0,1). P i Let G be the individual historical best position of particle i, and G be the global historical best position. The gradient of the objective function in iteration t Norm, w t For dynamic inertia weights, w min For the minimum inertia weight, w max For maximum inertia weight, For attenuation intensity, The rate of change intensity is T, and the maximum number of iterations is T. Let be the rate of change of the objective function after t iterations. Let Θ be the initial value of the objective function, θ be the current parameter vector of the UAV wind field model, and δ be the difference step size. The target value is the positive disturbance. The target value is the negative disturbance. Update the individual optimal particle and the global optimal particle, calculate the multi-index objective function value, and repeat the iteration until the maximum number of iterations is reached or the objective function change rate is less than 0.05% for three consecutive iterations. Output the optimal UAV wind field model hyperparameters and update the UAV wind field model. Input the target test scenario into the optimized UAV wind field model to determine the optimized prediction wind turbine data for multi-wind turbine collaborative control.

Images