Device and method for dynamically regulating and controlling spraying of dust suppression unmanned aerial vehicle for photovoltaic construction of loess

Abstract

The invention relates to the technical field of unmanned aerial vehicle dynamic dust suppression intelligent decision making based on multi-sensor data fusion, in particular to a loess photovoltaic construction dust suppression unmanned aerial vehicle spraying dynamic regulation and control device and method, and the method comprises the steps: building a dynamically updated four-dimensional concentration field through a sensor cooperation unit in combination with a turbulence diffusion model of a dust field reconstruction engine; and the strategy knowledge base is optimized to shorten the flight path planning decision-making period from the minute level of the existing offline planning to the second level response. And the anti-interference execution unit realizes accurate spray trajectory tracking under a complex wind field condition. And a real-time evaluation closed loop of the dust suppression effect is constructed by a dual-spectrum imaging and deep learning analysis technology of the efficiency feedback unit, so that a dust field model can dynamically correct boundary condition parameters. The response lag time of an existing dust suppression system is shortened, meanwhile, the spray coverage rate is increased, energy consumption is reduced on the premise that the dust suppression effect is guaranteed, and an intelligent solution is provided for photovoltaic construction flying dust treatment.

Description

Technical Field

[0001] The present invention relates to the field of intelligent decision-making technology for dynamic dust suppression by unmanned aerial vehicles (UAVs) based on multi-sensor data fusion, and in particular to a dynamic control device and method for spraying by unmanned aerial vehicles (UAVs) for dust suppression in photovoltaic construction on loess soil. Background Art

[0002] At photovoltaic construction sites on the loess plateau, the soil is fine and loose in texture. The disturbance caused by the operation of construction machinery, combined with ambient wind, causes dust particles to escape into the air, increasing the concentration of inhalable particulate matter in the air. This not only exacerbates visibility reduction in the construction area and threatens work safety, but also causes the migration and spread of fine particles, polluting the surrounding ecological environment. Given that existing ground-based spraying dust suppression systems are limited by the undulating terrain and cannot achieve comprehensive and uniform coverage, a solution is needed: drones equipped with spray platforms. Utilizing their three-dimensional maneuverability and combined with RTK high-precision positioning technology, drones can precisely navigate designated airspaces, carrying micron-sized atomizing nozzles to spray water mist or polymer dust suppressant solutions, forming a fine mist curtain to capture discrete dust particles. These atomized droplets physically contact and envelop the dust particles through a collision capture mechanism, promoting particle cohesion and increasing particle size, thereby increasing settling velocity and achieving rapid dust reduction. Simultaneously, the dust suppressant forms a film on the particle surface, suppressing secondary dust. Drone operations effectively compensate for the spatial coverage deficiencies of ground equipment, achieving dynamic dust suppression across the entire space, comprehensively reducing construction dust levels to meet environmental protection standards, and optimizing construction cleanliness and operational safety.

[0003] In the dust suppression scenario of photovoltaic construction in the Loess Plateau, the drone spray system needs to dynamically adjust the spray parameters and flight trajectory in real time according to the spatiotemporal distribution of dust concentration to improve the dust suppression efficiency. The change in dust concentration is caused by the disturbance of construction machinery activities and the unpredictability of environmental wind. The system is required to integrate the real-time dust density data, wind speed and direction vectors and terrain information collected by multi-source sensors to build a dynamic optimization model to generate control instructions. However, due to the insufficient parallel computing capability of the data processing architecture and the high complexity of the optimization algorithm, the existing technology cannot efficiently parse and respond to dynamic data input under low latency requirements, resulting in the control instructions lagging behind the evolution of the environmental state. For example, when the wind field shifts rapidly and changes the direction of dust dispersion, the monitoring data stream input decision algorithm cannot update the spray position coordinates in real time due to calculation delays, causing the drone to use the old path planning, wasting resources and increasing the risk of dust escape. This pain point is attributed to the insufficient efficiency of the real-time data processing system in the low-latency optimization decision process. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides a dynamic control device and method for spraying by drones for photovoltaic construction of loess soil to suppress dust. The present invention solves the technical problem of delayed real-time decision-making for drone dust suppression operations in scenarios with dynamic changes in the dust diffusion environment.

[0005] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows: In a first aspect, the present invention provides a dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil, comprising: A sensor collaboration unit is configured to collect dust concentration signals, wind speed vectors, and terrain point cloud data in the construction area and generate time-aligned structured data packets; a dust field reconstruction engine, connected to the sensor collaboration unit, configured to fuse discrete data in the structured data packet to establish a dust concentration distribution matrix in the spatiotemporal dimension; Optimization strategy knowledge base, which stores optimization algorithms for various environmental scenarios and is configured to match corresponding algorithms based on dust diffusion variance parameters; The prediction decision optimization unit inputs the dust concentration distribution matrix and is configured to execute a three-level optimization logic, which includes: The primary module previews the dust movement trajectory to generate the initial track; The secondary module injects meteorological variables to perform Monte Carlo simulation verification; The third-level optimization module receives the spraying status data, compares the spraying status data with the expected coverage, and dynamically adjusts the track tracking error weight and spray flow parameters; The anti-interference execution unit is connected to the prediction decision optimization unit and is configured to call the reinforcement learning strategy based on the deep Q network to generate coordinated control instructions for the UAV's flight attitude and spray flow; an efficiency feedback unit, comprising a dual-spectral imager and a convolutional neural network analysis module, wherein the dual-spectral imager is used to capture dust suppressant coverage images, the convolutional neural network analysis module is used to extract unsuppressed dust coordinates, and correction coefficients generated by the image coordinates are output to the dust field reconstruction engine; Among them, the structured data packet flows to the dust field reconstruction engine, the dust concentration distribution matrix is ​​input into the prediction decision optimization unit, the collaborative control instructions drive the UAV to perform spraying, and the three-level optimization module of the prediction decision optimization unit receives the spraying status data and the correction coefficient reconstructs the dust migration model established by the dust field reconstruction engine.

[0006] Furthermore, in the Loess Land Photovoltaic Construction Dust Suppression UAV Spray Dynamic Control Device described in the present invention, the sensor coordination unit includes a laser scattering sensor, an ultrasonic anemometer, and a rotating laser radar, wherein: The laser scattering sensor outputs 0-1000 at a frequency of 10-20 Hz Dust concentration signal; The ultrasonic anemometer generates a three-dimensional wind speed vector including east, north, and vertical components; The rotating laser radar scans the terrain to generate centimeter-level precision terrain point cloud data; The dust concentration signal, three-dimensional wind speed vector and terrain point cloud data are timestamped using a preset protocol to generate a time-aligned structured data packet.

[0007] Furthermore, in the Loess Plateau photovoltaic construction dust suppression drone spray dynamic control device described in the present invention, the dust field reconstruction engine includes a spatial interpolation module, a physical evolution module and a matrix generation module, wherein: The spatial interpolation module processes the discrete dust data in the structured data packet and constructs a plane grid concentration surface; The physical evolution module is coupled with the RANS turbulence model to calculate the dust migration rate; The matrix generation module integrates the plane grid concentration surface and the dust migration rate to output the dust concentration distribution matrix of the 15-second prediction window.

[0008] Furthermore, in the Loess Land Photovoltaic Construction Dust Suppression UAV Spray Dynamic Control Device of the present invention, the optimization strategy knowledge base includes a variance analysis unit and a rule triggering module, wherein: The variance analysis unit calculates the diffusion rate variance of the dust concentration distribution matrix ; The rule triggering module is configured to execute the following strategies: when When <0.5, the preset rolling horizon optimization framework is called to generate the initial track calculation parameters; When 0.5≤ When ≤1.2, the preset chaotic particle swarm algorithm is activated to generate simulation verification parameters; when When the value is >1.2, the preset multi-objective genetic algorithm is enabled to output the optimization weights.

[0009] Furthermore, in the Loess Land Photovoltaic Construction Dust Suppression UAV Spray Dynamic Control Device described in the present invention, the secondary module includes a scene loading unit, a disturbance injection unit, and a path screening unit, wherein: The scene loading unit constructs a digital twin model of the virtual construction scene; The disturbance injection unit injects a random disturbance variable with a wind speed amplitude of ±15% into the digital twin model; The path screening unit performs 500 Monte Carlo simulations on the initial trajectory, screens out optimized trajectory instructions with a failure probability of ≤5%, and outputs the optimized trajectory instructions to the anti-interference execution unit.

[0010] Furthermore, in the Loess Plateau photovoltaic construction dust suppression drone spray dynamic control device described in the present invention, the anti-interference execution unit includes a path tracking controller, a dynamic compensation module and an instruction fusion module, wherein: The path tracking controller receives initial track calculation parameters and generates a heading angle control sequence; The dynamic compensation module constructs a Lyapunov attitude control function to output a roll angle instruction; The instruction fusion module integrates the heading angle control sequence and the roll angle instruction to generate the coordinated control instruction.

[0011] Furthermore, in the Loess Land Photovoltaic Construction Dust Suppression UAV Spray Dynamic Control Device of the present invention, the performance feedback unit includes an image acquisition module, a feature recognition module, a coordinate conversion module, and a correction generation module, wherein: The image acquisition module captures dust suppressant coverage images through a dual-spectral imager; The feature recognition module uses a convolutional neural network to extract the coordinates of unsuppressed dust; The coordinate conversion module executes a perspective transformation algorithm to map the image coordinates to the global coordinate system of the construction site; The correction generation module outputs correction coefficients to the dust field reconstruction engine based on the spatial distribution of unsuppressed dust coordinates.

[0012] Furthermore, in the loess soil photovoltaic construction dust suppression drone spray dynamic control device described in the present invention, the three-level optimization module includes a data comparison unit, a weight update unit and a coefficient feedback unit, wherein: The data comparison unit receives the drone spraying status data and compares the real-time spraying coverage coordinates with the expected coverage range to generate a coverage deviation; The weight updating unit updates the genetic algorithm weight coefficient according to the coverage deviation; The coefficient feedback unit inputs the correction coefficient into the dust field reconstruction engine.

[0013] Furthermore, in the loess soil photovoltaic construction dust suppression drone spray dynamic control device described in the present invention, the correction generation module includes an area calculation unit, a boundary adjustment unit and a coefficient output unit, wherein: The area calculation unit calculates the residual dust area ratio based on the spatial distribution of the unsuppressed dust coordinates; The boundary adjustment unit linearly adjusts the boundary condition parameters of the RANS turbulence model according to the residual dust area ratio; The coefficient output unit outputs the correction coefficient carrying the updated boundary conditions to the dust field reconstruction engine.

[0014] In a second aspect, the present invention provides a method for dynamically controlling the spraying of a dust suppression drone during photovoltaic construction in loess soil, which is applied to the aforementioned dynamic control device for the spraying of a dust suppression drone during photovoltaic construction in loess soil, comprising: Step 1: The sensor collaborative unit collects dust concentration signals, wind speed vectors, and terrain point cloud data from the construction area to generate a time-aligned structured data packet. Step 2: The dust field reconstruction engine connected to the sensor collaborative unit fuses the discrete data in the structured data packet to establish a dust concentration distribution matrix in the spatiotemporal dimension; Step 3: Call the optimization strategy knowledge base to match the corresponding algorithm based on the dust diffusion variance parameter, input the dust concentration distribution matrix to the prediction decision optimization unit to execute the three-level optimization logic. The three-level optimization logic includes: the primary step is to preview the dust motion trajectory to generate the initial track, the secondary step is to inject meteorological variables to perform Monte Carlo simulation verification, receive the spraying status data, compare the spraying status data with the expected coverage range, and dynamically adjust the optimization parameters; Step 4: The anti-interference execution unit calls the reinforcement learning strategy to generate coordinated control instructions for the UAV's flight attitude and spray flow rate; Step 5: The performance feedback unit captures the dust suppressant coverage image using a dual-spectral imager, extracts the coordinates of the unsuppressed dust using a convolutional neural network, and outputs the correction coefficients generated by the image coordinates to the dust field reconstruction engine. Among them, the structured data packet flows to the dust field reconstruction engine, the dust concentration distribution matrix is ​​input into the prediction decision optimization unit, the collaborative control instructions drive the drone to perform spraying, and the spraying status data is transmitted back to step 3. The correction coefficient reconstructs the dust migration model established by the dust field reconstruction engine.

[0015] Beneficial effects of the present invention: The present invention significantly improves the real-time performance and adaptability of UAV dust suppression operations in photovoltaic construction scenarios in loess soil by constructing a closed-loop control system of multimodal perception and intelligent decision-making. The technical solution realizes the synchronous collection and spatiotemporal alignment of dust concentration, wind speed and terrain data through the sensor collaboration unit, and establishes a dynamically updated four-dimensional concentration field in combination with the turbulent diffusion model of the dust field reconstruction engine, solving the decision-making deviation problem caused by the lag in environmental perception in existing methods. The multi-algorithm switching mechanism of the optimization strategy knowledge base and the three-level optimization logic of the prediction decision optimization unit work together to shorten the trajectory planning decision cycle from the minute level of existing offline planning to the second level response. The hierarchical control architecture of the anti-interference execution unit realizes precise spray trajectory tracking under complex wind field conditions through the fusion control of the heading angle sequence and roll angle compensation. The dual-spectral imaging and deep learning analysis technology of the performance feedback unit construct a real-time evaluation closed loop for dust suppression effect, enabling the dust field model to dynamically correct the boundary condition parameters. This technical solution shortens the response lag time of the existing dust suppression system by optimizing the entire process of perception, decision-making, execution, and feedback. At the same time, it improves the spray coverage rate, reduces energy consumption while ensuring the dust suppression effect, and provides an intelligent solution for dust control in photovoltaic construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on the drawings without paying any creative labor.

[0017] Figure 1 A flow chart of a method for dynamically controlling spraying of dust suppression drones for photovoltaic construction in loess soil provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. The technical solutions provided by each embodiment of the present invention are described in detail below in conjunction with the drawings. In order to better understand the purpose of the present invention, the present invention is further described in detail below.

[0019] In a first aspect, the present invention provides a dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil, comprising: A sensor collaboration unit is configured to collect dust concentration signals, wind speed vectors, and terrain point cloud data in the construction area and generate time-aligned structured data packets; a dust field reconstruction engine, connected to the sensor collaboration unit, configured to fuse discrete data in the structured data packet to establish a dust concentration distribution matrix in the spatiotemporal dimension; Optimization strategy knowledge base, which stores optimization algorithms for various environmental scenarios and is configured to match corresponding algorithms based on dust diffusion variance parameters; The prediction decision optimization unit inputs the dust concentration distribution matrix and is configured to execute a three-level optimization logic, which includes: The primary module previews the dust movement trajectory to generate the initial track; The secondary module injects meteorological variables to perform Monte Carlo simulation verification; The third-level optimization module receives the spraying status data, compares the spraying status data with the expected coverage, and dynamically adjusts the track tracking error weight and spray flow parameters; The anti-interference execution unit is connected to the prediction decision optimization unit and is configured to call the reinforcement learning strategy based on the deep Q network to generate coordinated control instructions for the UAV's flight attitude and spray flow; an efficiency feedback unit, comprising a dual-spectral imager and a convolutional neural network analysis module, wherein the dual-spectral imager is used to capture dust suppressant coverage images, the convolutional neural network analysis module is used to extract unsuppressed dust coordinates, and correction coefficients generated by the image coordinates are output to the dust field reconstruction engine; Among them, the structured data packet flows to the dust field reconstruction engine, the dust concentration distribution matrix is ​​input into the prediction decision optimization unit, the collaborative control instructions drive the UAV to perform spraying, and the three-level optimization module of the prediction decision optimization unit receives the spraying status data and the correction coefficient reconstructs the dust migration model established by the dust field reconstruction engine.

[0020] The present invention provides a dynamic control device for spraying dust suppression drones during photovoltaic construction in loess soil. Its technical solution achieves dynamic dust suppression control through multi-level data processing and decision optimization. The sensor collaboration unit integrates laser scattering, ultrasonic wind measurement, and lidar technologies to simultaneously collect dust concentration, wind speed vectors, and terrain point cloud data within the construction area. This unit uses a precision clock protocol to time-align heterogeneous data from multiple sources, generating structured data packets that provide standardized input for subsequent analysis.

[0021] After receiving structured data packets, the dust field reconstruction engine reconstructs discrete dust monitoring data into a continuous planar grid concentration surface using a spatial interpolation algorithm. The physical evolution module calculates dust migration rates based on the RANS turbulence model and simulates the particle diffusion process using wind speed field data. The matrix generation module integrates spatial concentration distribution and temporal evolution characteristics to output a spatiotemporal dust concentration matrix with predictive capabilities.

[0022] The optimization strategy knowledge base stores intelligent algorithms for various environmental scenarios, and the variance analysis unit calculates the dust diffusion fluctuation characteristics. The rule triggering module dynamically selects an optimization algorithm based on the diffusion variance threshold, including rolling horizon optimization, chaotic particle swarm optimization, and multi-objective genetic algorithm, to achieve adaptive matching between the algorithm and the scenario.

[0023] The predictive decision-making optimization unit utilizes a three-level progressive optimization architecture. The primary module predicts dust motion trajectories based on a fluid dynamics model and generates an initial flight path. The secondary module injects random meteorological disturbances into the digital twin environment and verifies the trajectory robustness through Monte Carlo simulation. The tertiary optimization module receives real-time spraying status feedback and dynamically adjusts trajectory tracking weights and spray flow parameters.

[0024] The anti-interference execution unit deploys a deep Q-network reinforcement learning strategy to convert optimization commands into UAV flight control parameters. The path tracking controller generates a heading angle sequence, the dynamic compensation module outputs attitude adjustment commands, and the command fusion module ultimately generates coordinated control signals for flight attitude and spray flow.

[0025] The performance feedback unit uses bispectral imaging technology to capture images of dust suppressant coverage, and the convolutional neural network analysis module extracts the coordinates of unsuppressed dust areas. The coordinate conversion module maps the image coordinates to a global coordinate system, and the correction generation module uses this information to update the boundary conditions of the fluid dynamics model, forming a closed-loop optimization system.

[0026] Specifically, in the Loess Plateau photovoltaic construction dust suppression drone spray dynamic control device described in the present invention, the sensor coordination unit includes a laser scattering sensor, an ultrasonic anemometer and a rotating laser radar, wherein: The laser scattering sensor outputs 0-1000 at a frequency of 10-20 Hz Dust concentration signal; The ultrasonic anemometer generates a three-dimensional wind speed vector including east, north, and vertical components; The rotating laser radar scans the terrain to generate centimeter-level precision terrain point cloud data; The dust concentration signal, three-dimensional wind speed vector and terrain point cloud data are timestamped using a preset protocol to generate a time-aligned structured data packet.

[0027] In the dynamic spray control device for photovoltaic construction dust suppression using a drone on loess soil, the sensor collaboration unit utilizes multimodal sensing technology to collaboratively collect and integrate construction environment parameters. The laser scattering sensor, based on the Mie scattering principle, detects the light scattering intensity of suspended particles in the air. A calibration curve is used to convert the light intensity signal into a dust concentration value. Its dynamic sampling frequency adapts to the rapid fluctuations of dust generated during loess soil construction.

[0028] The ultrasonic anemometer uses the time difference measurement principle. It captures the time delay differences of sound waves propagating in three-dimensional space through an array of ultrasonic transducers, and calculates the wind speed vector including east, north and vertical components, providing fluid dynamics input parameters for dust migration modeling.

[0029] The rotating lidar emits a linear laser beam and receives the reflected signal from the ground. Combining this with inertial measurement unit and global positioning system data, it constructs a centimeter-level accurate 3D point cloud model of the construction area. This model not only characterizes the terrain's undulations but also identifies the spatial distribution of obstacles such as photovoltaic arrays and construction machinery, providing a geographic reference for drone trajectory planning.

[0030] The timing synchronization of multi-source heterogeneous data is achieved through a dual mechanism of hardware triggering and software protocol. The laser scattering sensor outputs a pulse signal as the main clock source, and the sampling periods of the ultrasonic anemometer and lidar are aligned with it, ultimately generating a structured data packet with strictly matched timestamps.

[0031] The spatial discreteness of the dust concentration signal is compensated for using a Kriging interpolation algorithm. Combined with the convection-diffusion parameters of the three-dimensional wind speed vector, a spatiotemporal representation of the dust concentration field is established. Terrain point cloud data, after voxel gridding, is then overlaid with the interpolated concentration data to generate a dust distribution matrix with elevation attributes, providing a unified input format for subsequent optimization decisions.

[0032] This technical solution solves the problem of insufficient spatial coverage of fixed monitoring equipment through multi-sensor data fusion. Its dynamic sampling characteristics significantly improve the timeliness of dust field reconstruction, which is in line with the application characteristics of intelligent decision-making systems in the field of environmental monitoring.

[0033] Specifically, the dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil described in the present invention, the dust field reconstruction engine includes a spatial interpolation module, a physical evolution module and a matrix generation module, wherein: The spatial interpolation module processes the discrete dust data in the structured data packet and constructs a plane grid concentration surface; The physical evolution module is coupled with the RANS turbulence model to calculate the dust migration rate; The matrix generation module integrates the plane grid concentration surface and the dust migration rate to output the dust concentration distribution matrix of the 15-second prediction window.

[0034] In the dynamic control device for drone spraying during photovoltaic construction in loess soil, the dust field reconstruction engine utilizes multiple modules to achieve dynamic modeling and prediction of the dust field. The spatial interpolation module uses a Kriging interpolation algorithm to process discrete dust monitoring data, converting sparsely distributed sensor collection points within the construction area into a continuously distributed planar grid concentration surface. The grid resolution matches the drone spray coverage accuracy, addressing the issue of insufficient spatial coverage of existing discrete monitoring data. This module also incorporates terrain elevation data to modify interpolation weights, enabling the planar grid concentration surface to reflect the impact of loess terrain fluctuations on dust deposition.

[0035] The physical evolution module constructs the dust migration dynamics equation based on the RANS turbulence model. It couples the convection term of the three-dimensional wind velocity vector with the turbulent diffusion term to calculate the migration rate of dust particles in the fluid field. The model boundary conditions are dynamically adjusted based on the dust suppressant coverage status provided by the effectiveness feedback unit. By introducing a boundary layer parameter correction factor, the turbulence model adapts to changes in surface roughness in the photovoltaic construction area. This module uses the finite volume method to discretize the governing equations, with a time step synchronized with the sensor sampling period to ensure timely calculations.

[0036] The matrix generation module spatially and temporally aligns the gridded concentration surface output by the spatial interpolation module with the migration rate field calculated by the physical evolution module to generate a four-dimensional concentration distribution matrix with predictive capabilities. The matrix's temporal dimension includes historical observations and future prediction windows, while the spatial dimension utilizes a latitude-longitude-altitude hierarchical structure, with each grid layer storing the predicted concentration value for the corresponding spatial location. The matrix data format is compatible with the input requirements of the optimization strategy knowledge base and supports a sliding time window update mechanism, retaining the latest observation data and eliminating outdated data during each cycle to maintain the spatiotemporal continuity of the matrix.

[0037] The dust field reconstruction engine realizes data fusion and prediction functions through a three-level processing pipeline. The grid concentration surface output by the spatial interpolation module provides the initial field for the physical evolution module. The migration rate calculated by the physical evolution module is fed back to the spatial interpolation module to optimize the interpolation weight of the next cycle, forming a closed-loop optimization system. When the matrix generation module integrates the intermediate results, a double verification mechanism is used to verify the consistency of the data, including grid alignment verification and physical dimension verification, to avoid invalid data from flowing into the decision-making link. This technical solution solves the problem of disconnection between model prediction and measured data in existing dust monitoring through multi-module collaboration, which is in line with the application characteristics of intelligent decision-making systems in the field of environmental engineering.

[0038] Specifically, the dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil according to the present invention, the optimization strategy knowledge base includes a variance analysis unit and a rule triggering module, wherein: The variance analysis unit calculates the diffusion rate variance of the dust concentration distribution matrix ; The rule triggering module is configured to execute the following strategies: when When <0.5, the preset rolling horizon optimization framework is called to generate the initial track calculation parameters; When 0.5≤ When ≤1.2, the preset chaotic particle swarm algorithm is activated to generate simulation verification parameters; when When the value is >1.2, the preset multi-objective genetic algorithm is enabled to output the optimization weights.

[0039] In the dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil, the optimization strategy knowledge base achieves adaptive decision-making for dust diffusion scenarios through a multimodal algorithm matching mechanism. The variance analysis unit uses a sliding window method to calculate the spatiotemporal variation characteristics of the dust concentration distribution matrix, quantifies the degree of fluctuation of the diffusion rate through discrete difference operations, and outputs the standardized variance parameter. As the basis for scene classification, the unit integrates the Kalman filter algorithm to eliminate the interference of sensor noise on variance calculation, making the diffusion feature representation more consistent with the laws of fluid mechanics.

[0040] The rule triggering module builds a three-level decision tree to implement dynamic algorithm scheduling. When the dust concentration is below the threshold, the Receding Horizon Optimization framework employs the principles of model predictive control to decompose trajectory planning into an optimal control problem within continuous time windows. Within each window, the optimal solution to the UAV's equation of motion is found based on the current dust concentration gradient. This framework incorporates Lyapunov stability constraints to ensure smoothness and convergence of the trajectory in low-disturbance environments.

[0041] When the chaotic particle swarm algorithm is triggered in a medium-variance interval, the system first initializes the particle swarm's position and velocity vectors, where the position parameter corresponds to the combined solution of spray flow rate and flight speed. The algorithm uses a tent chaotic map to generate the initial population, enhancing global search capabilities. The fitness function comprehensively considers track coverage and energy efficiency, and dynamic inertia weighting is used during the iteration process to balance exploration and development capabilities. During the simulation verification phase, random wind speed perturbations are injected through the digital twin engine to evaluate the robustness of the particle swarm solution.

[0042] The multi-objective genetic algorithm used in high-variance scenarios utilizes the NSGA-II framework to construct a Pareto frontier encompassing dust suppression efficiency, range duration, and battery loss. Chromosome encoding utilizes a mixed real and binary representation, a simulated binary crossover strategy is introduced as the crossover operator, and polynomial mutation is employed during the mutation phase to maintain population diversity. A weight optimization module dynamically adjusts the weighting coefficients of the objective function through a fuzzy inference system to adapt to the priorities of different construction phases.

[0043] Each algorithm module interacts with data through a shared memory pool. The output parameters of the rolling time-domain optimization serve as the initial solution space boundary for the chaotic particle swarm. The non-dominated solution set of the genetic algorithm is fed back into the knowledge base to update the threshold logic of the rule base. This closed-loop optimization architecture enables the system to adapt to the dynamic characteristics of dust dispersion during photovoltaic construction in loess soil, and aligns with the technical characteristics of intelligent decision-making systems in the field of environmental control.

[0044] Specifically, the dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil according to the present invention comprises a secondary module including a scene loading unit, a disturbance injection unit, and a path screening unit, wherein: The scene loading unit constructs a digital twin model of the virtual construction scene; The disturbance injection unit injects a random disturbance variable with a wind speed amplitude of ±15% into the digital twin model; The path screening unit performs 500 Monte Carlo simulations on the initial trajectory, screens out optimized trajectory instructions with a failure probability of ≤5%, and outputs the optimized trajectory instructions to the anti-interference execution unit.

[0045] In the dynamic spray control device for photovoltaic construction dust suppression using a drone on loess soil, described in this invention, a secondary module achieves dynamic trajectory optimization and robustness verification through digital twinning and probabilistic simulation technologies. The scene loading unit constructs a three-dimensional virtual scene model based on terrain point cloud data collected by lidar and photovoltaic module layout information, including elevation features, obstacle distribution, and the dynamic positions of construction machinery. This model maintains dynamic consistency with the actual construction progress through a timestamp synchronization mechanism and integrates structural parameters from the BIM model, providing a high-fidelity environment foundation for subsequent simulations.

[0046] The disturbance injection unit uses stochastic process theory to generate a Weibull-distributed wind speed perturbation sequence and superimposes the fluctuating component of the three-dimensional wind speed vector on the digital twin model. The perturbation variable is linked to the surface roughness of the loess landform via a correction coefficient for the fluid dynamics boundary layer parameter, ensuring that the injected random perturbation conforms to the actual wind field characteristics. This unit also simultaneously adjusts the turbulence intensity parameters of the dust transport model to ensure physical consistency between the fluid simulation and the trajectory verification.

[0047] The path screening unit decomposes the initial trajectory into a sequence of discrete waypoints, each of which associates spray flow with flight attitude parameters. During Monte Carlo simulation, the system resamples the wind speed perturbation sequence at each iteration to calculate the interaction between the drone's dynamic model and the dust concentration field. The failure determination logic comprehensively evaluates trajectory deviation, spray coverage blind spots, and battery power thresholds, ultimately selecting an optimized trajectory instruction set whose robustness meets pre-set reliability criteria.

[0048] Specifically, the anti-interference execution unit of the dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil of the present invention includes a path tracking controller, a dynamic compensation module, and an instruction fusion module, wherein: The path tracking controller receives initial track calculation parameters and generates a heading angle control sequence; The dynamic compensation module constructs a Lyapunov attitude control function to output a roll angle instruction; The instruction fusion module integrates the heading angle control sequence and the roll angle instruction to generate the coordinated control instruction.

[0049] In the dynamic control device for spraying by a drone used in photovoltaic construction dust suppression in loess soil, described in the present invention, an anti-interference execution unit achieves flight stability and spray accuracy in complex environments through a hierarchical control architecture. Based on the drone's kinematic model, the path tracking controller decomposes the initial trajectory calculation parameters into a continuous sequence of waypoints and uses a feedforward-feedback composite control strategy to generate heading angle control instructions. The controller integrates the distributed characteristic parameters of the photovoltaic array and dynamically adjusts the heading angle change rate through a curvature adaptive algorithm to avoid spray coverage blind spots caused by the gap between photovoltaic panels. At the same time, a trajectory smoothness constraint is introduced to ensure that the generated heading angle sequence conforms to the drone's dynamic characteristics.

[0050] The dynamic compensation module addresses the unique turbulent wind field characteristics of loess landforms by constructing a nonlinear attitude control function based on Lyapunov stability theory. This function calculates the spatial deviation vector between the drone's center of mass and the desired trajectory in real time, and, combined with angular velocity data collected by the inertial measurement unit, outputs roll angle compensation commands. The module employs sliding-mode variable structure control to suppress crosswind disturbances and adjusts control gains using the boundary layer thickness parameter. This reduces high-frequency buffeting while maintaining stability, enabling the drone to maintain a preset flight attitude in complex airflow conditions.

[0051] The command fusion module uses a weighted fusion algorithm to integrate the heading angle control sequence and the roll angle command to generate the final coordinated control command. During the fusion process, the module dynamically adjusts the weight coefficients of track tracking accuracy and attitude stability based on the gradient change characteristics of the dust concentration distribution matrix. When a high-concentration dust area is detected, the priority of the spray flow parameters is automatically increased, and the UAV flight speed and nozzle opening parameters are simultaneously optimized to achieve a balance between dust suppression effect and energy efficiency. The module has a built-in command conflict detection mechanism. When there is a physical constraint conflict between the heading angle command and the roll angle command, the relaxation factor adjustment strategy is activated to redistribute the control amount within the allowable error range.

[0052] Specifically, the Loess Plateau photovoltaic construction dust suppression drone spray dynamic control device of the present invention includes an image acquisition module, a feature recognition module, a coordinate conversion module, and a correction generation module, wherein: The image acquisition module captures dust suppressant coverage images through a dual-spectral imager; The feature recognition module uses a convolutional neural network to extract the coordinates of unsuppressed dust; The coordinate conversion module executes a perspective transformation algorithm to map the image coordinates to the global coordinate system of the construction site; The correction generation module outputs correction coefficients to the dust field reconstruction engine based on the spatial distribution of unsuppressed dust coordinates.

[0053] In the dynamic spray control device for photovoltaic construction dust suppression using a drone on the Loess Plateau, the performance feedback unit utilizes multimodal sensing and spatial mapping technology to achieve closed-loop evaluation of dust suppression effectiveness. The image acquisition module, equipped with a dual-spectral imager, simultaneously captures images of dust suppressant coverage in both the visible and near-infrared bands. Multispectral fusion processing enhances the contrast between dust and the area covered by the dust suppressant. The module also incorporates an active fill-light system to adapt to changing lighting conditions in the photovoltaic construction area, ensuring color consistency across images captured at different times.

[0054] The feature recognition module utilizes a deep convolutional neural network architecture, loading a pre-trained dust feature extraction model through transfer learning. An attention mechanism layer is embedded within the network structure, focusing on the scattering characteristics of dust particles in the image and the optical interference characteristics of the dust suppressant film, outputting a pixel-level coordinate mask of unsuppressed dust areas. This module utilizes an online incremental learning strategy to continuously optimize the model's recognition accuracy for mineral dust unique to the Loess Plateau.

[0055] The coordinate conversion module establishes a mapping relationship between the image coordinate system and the global construction site coordinate system. The perspective transformation algorithm incorporates the drone's pose parameters as an extrinsic matrix and combines them with terrain point cloud data collected by the LiDAR to construct a 3D projection model. The module uses a feature point matching algorithm to eliminate image distortion caused by drone flight jitter, achieving sub-meter accuracy mapping of dust coordinates from the 2D image space to the 3D construction scene.

[0056] The correction generation module calculates a dust suppression correction coefficient based on the spatial distribution characteristics of unsuppressed dust coordinates. This module constructs a vector field model of dust diffusion trends and identifies primary dust escape paths and secondary diffusion areas by analyzing the concentration and directional consistency of coordinate points. The output correction coefficients, including coverage compensation parameters and spray trajectory adjustment weights, are fed back to the dust field reconstruction engine to drive a dynamic optimization closed loop.

[0057] Specifically, the dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil according to the present invention comprises a three-level optimization module including a data comparison unit, a weight update unit, and a coefficient feedback unit, wherein: The data comparison unit receives the drone spraying status data and compares the real-time spraying coverage coordinates with the expected coverage range to generate a coverage deviation; The weight updating unit updates the genetic algorithm weight coefficient according to the coverage deviation; The coefficient feedback unit inputs the correction coefficient into the dust field reconstruction engine.

[0058] In the dynamic spray control device for photovoltaic construction using a loess soil dust suppression drone, described in this invention, a three-level optimization module continuously optimizes the dust suppression strategy through a dynamic feedback mechanism. A data comparison unit uses a convolutional neural network to extract the boundary coordinates of the effective dust suppressant coverage area from real-time spray coverage images captured by the drone's onboard dual-spectral imager. This data comparison unit then spatially overlays the coordinates with the expected coverage polygons generated by the digital twin model to calculate a coverage deviation matrix. This matrix includes parameters such as spatial position deviation, area loss rate, and concentration gradient difference, providing a quantitative basis for weight adjustment.

[0059] Specifically, the dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil according to the present invention comprises an area calculation unit, a boundary adjustment unit, and a coefficient output unit, wherein: The area calculation unit calculates the residual dust area ratio based on the spatial distribution of the unsuppressed dust coordinates; The boundary adjustment unit linearly adjusts the boundary condition parameters of the RANS turbulence model according to the residual dust area ratio; The coefficient output unit outputs the correction coefficient carrying the updated boundary conditions to the dust field reconstruction engine.

[0060] In the dynamic control device for spraying dust suppression drones used in photovoltaic construction on loess soil, the correction generation module achieves dynamic dust field correction through spatial analysis coupled with fluid modeling. The area calculation unit, based on the global coordinate set of unsuppressed dust output by the coordinate conversion module, employs a Delaunay triangulation algorithm to construct an irregular polygonal mesh of the dust-covered area. This unit calculates the ratio of the overlapping area of ​​the polygonal mesh to the preset dust suppression target area to generate a residual dust area ratio parameter. A Gaussian filter algorithm is introduced to eliminate discrete noise interference during coordinate acquisition, improving the stability of the area calculation.

[0061] The boundary conditioning unit employs a layered conditioning strategy to process the boundary condition parameters of the RANS turbulence model. The primary conditioning layer dynamically adjusts the baseline values ​​of turbulence intensity and turbulent viscosity ratio based on the gradient change rate of the residual dust area fraction. The secondary conditioning layer, incorporating real-time drone altitude and wind speed sensor data, corrects the boundary layer thickness parameters using an eddy scale compensation algorithm. The unit incorporates a built-in database of wind field characteristics from the Loess Plateau. When a zonal distribution of dust accumulation is detected, the unit automatically activates the crosswind compensation submodel, superimposing a transverse momentum transport term on the mainstream turbulence model.

[0062] The coefficient output unit generates correction coefficients that carry updated boundary conditions through multi-channel data fusion. The dust diffusion coefficient is processed using a double buffering mechanism, with the correction coefficients for the current frame weightedly fused with the correction results from previous frames to avoid parameter jumps caused by image recognition delays. The output correction coefficient set includes three parameters: turbulent diffusion intensity, particle settling rate, and spatial distribution weight. These are encapsulated in JSON format and transmitted to the dust field reconstruction engine. This unit also generates a boundary condition change log, recording the parameter adjustment path for subsequent optimization analysis.

[0063] In a second aspect, the present invention provides a method for dynamically controlling the spraying of a dust suppression drone during photovoltaic construction in loess soil, which is applied to the aforementioned dynamic control device for the spraying of a dust suppression drone during photovoltaic construction in loess soil, comprising: Step 1: The sensor collaborative unit collects dust concentration signals, wind speed vectors, and terrain point cloud data from the construction area to generate a time-aligned structured data packet. Step 2: The dust field reconstruction engine connected to the sensor collaborative unit fuses the discrete data in the structured data packet to establish a dust concentration distribution matrix in the spatiotemporal dimension; Step 3: Call the optimization strategy knowledge base to match the corresponding algorithm based on the dust diffusion variance parameter, input the dust concentration distribution matrix to the prediction decision optimization unit to execute the three-level optimization logic. The three-level optimization logic includes: the primary step is to preview the dust motion trajectory to generate the initial track, the secondary step is to inject meteorological variables to perform Monte Carlo simulation verification, receive the spraying status data, compare the spraying status data with the expected coverage range, and dynamically adjust the optimization parameters; Step 4: The anti-interference execution unit calls the reinforcement learning strategy to generate coordinated control instructions for the UAV's flight attitude and spray flow rate; Step 5: The performance feedback unit captures the dust suppressant coverage image using a dual-spectral imager, extracts the coordinates of the unsuppressed dust using a convolutional neural network, and outputs the correction coefficients generated by the image coordinates to the dust field reconstruction engine. Among them, the structured data packet flows to the dust field reconstruction engine, the dust concentration distribution matrix is ​​input into the prediction decision optimization unit, the collaborative control instructions drive the drone to perform spraying, and the spraying status data is transmitted back to step 3. The correction coefficient reconstructs the dust migration model established by the dust field reconstruction engine.

[0064] The proposed method for dynamically controlling spraying by a dust suppression drone during photovoltaic construction in loess soil achieves dynamic perception and closed-loop control of dust fields through multi-module collaboration. The sensor collaboration unit integrates heterogeneous data from multiple sources, including lidar, anemometers, and dust sensors. Using a timestamp synchronization mechanism, it aligns the acquisition sequence and generates a structured data packet containing spatial coordinates, dust concentration gradients, and three-dimensional wind speed vectors. This data packet uses a spatial interpolation algorithm to eliminate sensor blind spots, providing a high-precision input reference for dust field reconstruction.

[0065] The dust field reconstruction engine uses a kriging interpolation algorithm to process data from discrete monitoring points within structured data packages, constructing a spatiotemporal dust concentration distribution matrix. Each grid node in the matrix associates a wind speed vector with a terrain roughness parameter. Using a turbulent diffusion model, the dust migration rate is calculated, resulting in a dynamically updateable four-dimensional concentration field model (three-dimensional space + time). This model uses boundary layer parameter correction coefficients to adapt to the unique wind field characteristics of loess landforms, improving the accuracy of dust dispersion predictions.

[0066] The optimization strategy knowledge base stores multiple dust dispersion prediction algorithms and automatically matches the optimal algorithm combination based on the real-time calculated dust dispersion variance parameters. The three-level optimization logic executed by the prediction and decision optimization unit includes the following: the initial step uses a Lagrangian particle tracking model to predict dust motion trajectories and generate an initial track covering high-concentration areas; the secondary step injects a Weibull-distributed wind speed disturbance variable into the digital twin model and verifies the track robustness through 500 Monte Carlo simulations; the tertiary step receives spraying status data from the performance feedback unit and uses an adaptive genetic algorithm to dynamically adjust the track planning weight coefficients, forming an interference-resistant optimized track instruction set.

[0067] The anti-interference execution unit utilizes a hierarchical control architecture to achieve coordinated control of flight and spraying. The path tracking controller decomposes the optimized trajectory into a heading angle control sequence. The dynamic compensation module generates roll angle commands based on the Lyapunov function to combat crosswind disturbances. The command fusion module uses a weighted fusion algorithm to output coordinated control commands that balance trajectory accuracy and spray coverage. This command synchronously adjusts the drone's flight speed and nozzle opening to ensure that the spray trajectory spatially matches the dust concentration gradient.

[0068] The performance feedback unit uses a dual-spectral imager to capture dust suppressant coverage images in the visible and near-infrared bands. A deep convolutional neural network is then used to extract pixel-level coordinate masks of unsuppressed dust areas. The coordinate conversion module, combined with the drone's pose parameters, performs a perspective transformation, mapping the image coordinates to the global construction site coordinate system. The correction generation module analyzes the spatial aggregation characteristics of residual dust and outputs feedback data, including turbulence intensity correction coefficients and spray flow compensation parameters. This data drives the dust field reconstruction engine to update the model boundary conditions, forming a closed-loop optimization chain.

[0069] This technical solution addresses the sluggish response and limited adaptability of existing dust suppression systems through a closed-loop architecture encompassing perception, decision-making, execution, and feedback. The sensor collaboration unit and dust field reconstruction engine form the environmental perception layer, while the optimization strategy knowledge base and predictive decision optimization unit form the intelligent decision-making layer. The anti-interference execution unit enables precise control, and the performance feedback unit performs effect evaluation and model calibration. Structured parameters are transferred between modules via standardized data interfaces, enabling dynamic dust suppression control in loess plate photovoltaic construction scenarios, in line with the technical characteristics of intelligent control systems in the field of environmental engineering.

[0070] The specific implementation of the dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil provided by the present invention is as follows: In photovoltaic construction scenarios, drones collect environmental data using a multi-sensor fusion system. LiDAR scans the construction area terrain to generate 3D point cloud data. A dust sensor array detects spatial dust concentration distribution at a fixed sampling frequency, and an ultrasonic anemometer measures wind speed vectors in real time. Sensor data is aligned using a timestamp synchronization module to form a structured data packet containing spatial coordinates, concentration gradients, and wind speed vectors. This data is then transmitted to the dust field reconstruction engine for spatiotemporal fusion processing.

[0071] The dust field reconstruction engine uses the Kriging spatial interpolation algorithm to process discrete monitoring data and construct a four-dimensional dust concentration field model. The model divides the construction area into a three-dimensional grid. Each grid node associates a wind speed vector with a terrain roughness parameter, and calculates dust migration rates using the turbulent diffusion equation. The engine incorporates a database of wind field characteristics from the Loess Plateau and dynamically modifies boundary layer parameters based on historical data, improving the model's prediction accuracy in complex terrains.

[0072] The optimization strategy knowledge base stores multiple dust dispersion prediction algorithms, including Lagrangian particle tracking models and Euler fluid dynamics models. The system automatically selects the optimal algorithm combination based on the real-time calculated dust dispersion variance parameters. The prediction and decision optimization unit implements a three-level optimization logic: The primary optimization uses a particle tracking model to preview dust motion trajectories and generate an initial trajectory covering high-concentration areas; the secondary optimization injects a Weibull-distributed wind speed perturbation variable into the digital twin model and verifies the trajectory robustness through Monte Carlo simulation; the tertiary optimization receives spray status data transmitted by the performance feedback unit and uses an adaptive genetic algorithm to dynamically adjust the trajectory planning weight coefficients.

[0073] The anti-interference execution unit utilizes a hierarchical control architecture to coordinate flight attitude and spray flow. The path tracking controller decomposes the optimized trajectory into a heading angle control sequence. The dynamic compensation module generates roll angle commands based on the Lyapunov function to combat crosswind disturbances. The command fusion module outputs coordinated control commands using a weighted fusion algorithm. These commands synchronously adjust the drone's flight speed and nozzle opening parameters to spatially match the spray trajectory to the dust concentration gradient. The spray system utilizes a dual-mode design: a high-pressure, fine-stream mode for targeted cleaning of strong dust sources, and a low-pressure, atomizing mode for large-scale dust suppression.

[0074] The performance feedback unit uses a dual-spectral imager to capture images of dust suppressant coverage in the visible and near-infrared bands. A deep convolutional neural network extracts pixel-level coordinate masks of unsuppressed dust areas. The coordinate conversion module, combined with the drone's pose parameters, performs a perspective transformation, mapping the image coordinates to the global construction site coordinate system. The correction generation module analyzes the spatial aggregation characteristics of residual dust and outputs feedback data, including turbulence intensity correction coefficients and spray flow compensation parameters, to drive the dust field reconstruction engine to update the model boundary conditions. The feedback cycle is dynamically adjusted based on construction intensity, with a high-frequency feedback mode employed in dust-active areas and an intermittent detection mode in stable areas.

[0075] The device features a modular design, with the sensor unit, control unit, and actuator connected via standard interfaces. The drone's platform utilizes a carbon fiber composite frame and is equipped with a redundant power supply system and dual-link communication modules. The control software is deployed on edge computing nodes, enabling dual-mode operation for real-time data processing and offline simulation. The system supports manual intervention, allowing operators to adjust and optimize weight parameters through a ground station to adapt to specific operating conditions.

[0076] The present invention solves the problem of decision lag under dynamic changes in the dust diffusion environment by constructing a closed-loop control system of multimodal perception-dynamic modeling-intelligent decision-making. The technical solution first integrates a lidar, a multispectral dust sensor, and an ultrasonic anemometer through a sensor collaboration unit to collect three-dimensional point cloud data, dust concentration gradients, and wind speed vectors of the construction area in real time to form a time-space aligned environmental state matrix. The dust field reconstruction engine uses the Kriging spatial interpolation algorithm to process discrete monitoring data, combines the turbulent diffusion model to establish a four-dimensional concentration field, and adapts to the unique wind field disturbance characteristics of the loess landform through the boundary layer parameter dynamic correction module to achieve minute-level updates of dust migration trends.

[0077] The optimization strategy knowledge base stores a hybrid algorithm combining Lagrangian particle tracking and Eulerian fluid dynamics, automatically switching prediction models based on real-time dust dispersion variance parameters. The prediction and decision optimization unit implements a three-level optimization logic: primary optimization generates an initial trajectory covering high-concentration areas; secondary optimization injects a Weibull-distributed wind speed disturbance variable into the digital twin model, verifying trajectory robustness through 500 Monte Carlo simulations; and tertiary optimization receives spraying status data transmitted by the performance feedback unit and uses an adaptive genetic algorithm to dynamically adjust trajectory planning weights, forming an interference-resistant optimization instruction set. This process, through the real-time coupling of the dust concentration distribution matrix and the trajectory cost function, reduces the decision cycle for existing offline planning from hours to seconds.

[0078] The anti-interference execution unit utilizes a hierarchical control architecture to achieve coordinated control of flight and spraying. The path tracking controller decomposes the optimized trajectory into a heading angle control sequence. The dynamic compensation module generates roll angle commands based on the Lyapunov function to combat crosswind disturbances. The command fusion module outputs coordinated control parameters using a weighted fusion algorithm, synchronously adjusting the drone's flight speed and nozzle opening to achieve a spatial match between the spray trajectory and the dust concentration gradient. The spray system utilizes a dual-mode design: a high-pressure, fine water stream mode targets targeted dust sources, while a low-pressure atomization mode achieves wide-area dust suppression coverage. Pulse-width modulation technology achieves millisecond-level flow rate response.

[0079] The performance feedback unit uses a dual-spectral imager to capture images of dust suppressant coverage in the visible and near-infrared bands. A deep convolutional neural network extracts pixel-level coordinate masks of unsuppressed dust areas. The coordinate conversion module combines the drone's pose parameters to perform a perspective transformation, mapping the image coordinates to the global construction site coordinate system. The correction generation module analyzes the spatial aggregation characteristics of residual dust and outputs feedback data, including turbulence intensity correction coefficients and spray flow compensation parameters, to drive the dust field reconstruction engine to update the model boundary conditions. This closed-loop feedback mechanism reduces the lag time of existing manual inspections from two hours to less than three minutes, achieving real-time alignment between numerical simulations of the dust diffusion field and measured data.

Claims

1. A dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil, characterized in that: include: A sensor collaboration unit is configured to collect dust concentration signals, wind speed vectors, and terrain point cloud data in the construction area and generate time-aligned structured data packets; a dust field reconstruction engine, connected to the sensor collaboration unit, configured to fuse discrete data in the structured data packet to establish a dust concentration distribution matrix in the spatiotemporal dimension; Optimization strategy knowledge base, which stores optimization algorithms for various environmental scenarios and is configured to match corresponding algorithms based on dust diffusion variance parameters; The prediction decision optimization unit inputs the dust concentration distribution matrix and is configured to execute a three-level optimization logic, which includes: The primary module previews the dust movement trajectory to generate the initial track; The secondary module injects meteorological variables to perform Monte Carlo simulation verification; The third-level optimization module receives the spraying status data, compares the spraying status data with the expected coverage, and dynamically adjusts the track tracking error weight and spray flow parameters; The anti-interference execution unit is connected to the prediction decision optimization unit and is configured to call the reinforcement learning strategy based on the deep Q network to generate coordinated control instructions for the UAV's flight attitude and spray flow; an efficiency feedback unit, comprising a dual-spectral imager and a convolutional neural network analysis module, wherein the dual-spectral imager is used to capture dust suppressant coverage images, the convolutional neural network analysis module is used to extract unsuppressed dust coordinates, and correction coefficients generated by the image coordinates are output to the dust field reconstruction engine; Among them, the structured data packet flows to the dust field reconstruction engine, the dust concentration distribution matrix is ​​input into the prediction decision optimization unit, the collaborative control instructions drive the UAV to perform spraying, and the three-level optimization module of the prediction decision optimization unit receives the spraying status data and the correction coefficient reconstructs the dust migration model established by the dust field reconstruction engine.

2. The dynamic control device for spraying dust suppression drones for photovoltaic construction in loess land according to claim 1 is characterized in that: The sensor cooperative unit includes a laser scattering sensor, an ultrasonic anemometer and a rotating laser radar, wherein: The laser scattering sensor outputs 0-1000 at a frequency of 10-20 Hz Dust concentration signal; The ultrasonic anemometer generates a three-dimensional wind speed vector including east, north, and vertical components; The rotating laser radar scans the terrain to generate centimeter-level precision terrain point cloud data; The dust concentration signal, three-dimensional wind speed vector and terrain point cloud data are timestamped using a preset protocol to generate a time-aligned structured data packet.

3. The dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil according to claim 1 is characterized in that: The dust field reconstruction engine includes a spatial interpolation module, a physical evolution module, and a matrix generation module, wherein: The spatial interpolation module processes the discrete dust data in the structured data packet and constructs a plane grid concentration surface; The physical evolution module is coupled with the RANS turbulence model to calculate the dust migration rate; The matrix generation module integrates the plane grid concentration surface and the dust migration rate to output the dust concentration distribution matrix of the 15-second prediction window.

4. The dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil according to claim 3 is characterized in that: The optimization strategy knowledge base includes a variance analysis unit and a rule triggering module, wherein: The variance analysis unit calculates the diffusion rate variance of the dust concentration distribution matrix ; The rule triggering module is configured to execute the following strategies: when When <0.5, the preset rolling horizon optimization framework is called to generate the initial track calculation parameters; When 0.5≤ When ≤1.2, the preset chaotic particle swarm algorithm is activated to generate simulation verification parameters; when When the value is >1.2, the preset multi-objective genetic algorithm is enabled to output the optimization weights.

5. The dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil according to claim 1 is characterized in that: The secondary module includes a scene loading unit, a disturbance injection unit and a path screening unit, wherein: The scene loading unit constructs a digital twin model of the virtual construction scene; The disturbance injection unit injects a random disturbance variable with a wind speed amplitude of ±15% into the digital twin model; The path screening unit performs 500 Monte Carlo simulations on the initial trajectory, screens out optimized trajectory instructions with a failure probability of ≤5%, and outputs the optimized trajectory instructions to the anti-interference execution unit.

6. The dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil according to claim 1 is characterized in that: The anti-interference execution unit includes a path tracking controller, a dynamic compensation module and an instruction fusion module, wherein: The path tracking controller receives initial track calculation parameters and generates a heading angle control sequence; The dynamic compensation module constructs a Lyapunov attitude control function to output a roll angle instruction; The instruction fusion module integrates the heading angle control sequence and the roll angle instruction to generate the coordinated control instruction.

7. The dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil according to claim 1 is characterized in that: The performance feedback unit includes an image acquisition module, a feature recognition module, a coordinate conversion module, and a correction generation module, wherein: The image acquisition module captures dust suppressant coverage images through a dual-spectral imager; The feature recognition module uses a convolutional neural network to extract the coordinates of unsuppressed dust; The coordinate conversion module executes a perspective transformation algorithm to map the image coordinates to the global coordinate system of the construction site; The correction generation module outputs correction coefficients to the dust field reconstruction engine based on the spatial distribution of unsuppressed dust coordinates.

8. The dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil according to claim 7 is characterized in that: The three-level optimization module includes a data comparison unit, a weight update unit, and a coefficient feedback unit, wherein: The data comparison unit receives the drone spraying status data and compares the real-time spraying coverage coordinates with the expected coverage range to generate a coverage deviation; The weight updating unit updates the genetic algorithm weight coefficient according to the coverage deviation; The coefficient feedback unit inputs the correction coefficient into the dust field reconstruction engine.

9. The dynamic control device for spraying dust suppression drones for photovoltaic construction in loess soil according to claim 8, characterized in that: The correction generation module includes an area calculation unit, a boundary adjustment unit and a coefficient output unit, wherein: The area calculation unit calculates the residual dust area ratio based on the spatial distribution of the unsuppressed dust coordinates; The boundary adjustment unit linearly adjusts the boundary condition parameters of the RANS turbulence model according to the residual dust area ratio; The coefficient output unit outputs the correction coefficient carrying the updated boundary conditions to the dust field reconstruction engine.

10. A method for dynamically controlling the spraying of a dust suppression drone during photovoltaic construction in loess soil, applied to the dynamic control device for spraying of a dust suppression drone during photovoltaic construction in loess soil as claimed in any one of claims 1 to 9, characterized in that: include: Step 1: The sensor collaborative unit collects dust concentration signals, wind speed vectors, and terrain point cloud data from the construction area to generate a time-aligned structured data packet. Step 2: The dust field reconstruction engine connected to the sensor collaborative unit fuses the discrete data in the structured data packet to establish a dust concentration distribution matrix in the spatiotemporal dimension; Step 3: Call the optimization strategy knowledge base to match the corresponding algorithm based on the dust diffusion variance parameter, input the dust concentration distribution matrix to the prediction decision optimization unit to execute the three-level optimization logic. The three-level optimization logic includes: the primary step is to preview the dust motion trajectory to generate the initial track, the secondary step is to inject meteorological variables to perform Monte Carlo simulation verification, receive the spraying status data, compare the spraying status data with the expected coverage range, and dynamically adjust the optimization parameters; Step 4: The anti-interference execution unit calls the reinforcement learning strategy to generate coordinated control instructions for the UAV's flight attitude and spray flow rate; Step 5: The performance feedback unit captures the dust suppressant coverage image using a dual-spectral imager, extracts the coordinates of the unsuppressed dust using a convolutional neural network, and outputs the correction coefficients generated by the image coordinates to the dust field reconstruction engine. Among them, the structured data packet flows to the dust field reconstruction engine, the dust concentration distribution matrix is ​​input into the prediction decision optimization unit, the collaborative control instructions drive the drone to perform spraying, and the spraying status data is transmitted back to step 3. The correction coefficient reconstructs the dust migration model established by the dust field reconstruction engine.

Images