A power transmission channel unmanned aerial vehicle terrain self-adaptive patrol route planning method and device

Abstract

The application provides a power transmission channel unmanned aerial vehicle terrain adaptive inspection flight path planning method and device, relates to the technical field of unmanned aerial vehicle autonomous inspection, and comprises the following steps: acquiring multi-source space data and inspection task data of a target power transmission channel; generating a terrain adaptive control grid based on the multi-source space data and the inspection task data; determining continuous terrain elevation based on a neural implicit terrain model and the terrain adaptive control grid, combining an electrical safety enhancement constraint and a dynamic flight control safety margin to construct a safe flight envelope; generating a control waypoint sequence by adopting a non-linear model predictive control model for simulating ground flight control under the constraint of the safe flight envelope; and optimizing the control waypoint sequence by adopting a diffusion trajectory optimization model to generate an inspection trajectory under the condition of imaging quality. The application realizes the collaborative optimization of flight safety, kinetic smoothness and high-quality inspection imaging under complex terrain.

Description

Technical Field

[0001] This application relates to the field of autonomous inspection technology for unmanned aerial vehicles (UAVs), specifically to a method and device for terrain-adaptive inspection route planning for UAVs in power transmission channels. Background Technology

[0002] Currently, in the field of UAV inspection of power transmission channels, flight path planning typically relies on digital elevation models and manually set waypoints or conventional path search algorithms. However, in scenarios where complex terrains such as mountains and hills interact with the spatial structure of power transmission lines, existing technologies still have many shortcomings. First, terrain modeling often uses fixed-resolution grids or fixed-interval waypoints, making it difficult to automatically adjust modeling accuracy according to the degree of terrain undulation and the density of inspection targets, resulting in computational redundancy in flat areas or insufficient representation of features in complex areas. Second, safety space construction usually treats obstacles as ordinary geometric shapes, failing to uniformly represent the unique constraints of power transmission equipment such as conductor sag, tower structure, and electrical safety distances, making it difficult to accurately describe the actual flyable space. Third, terrain-following flight altitude adjustment is mostly based on the current local terrain, lacking forward-looking predictive control that combines the terrain trend ahead and the UAV's dynamic constraints, easily leading to insufficient climb response or trajectory oscillations. Finally, flight path optimization often focuses on obstacle avoidance or energy consumption, resulting in insufficient coupling with inspection imaging quality, making it difficult to simultaneously meet the fine-grained acquisition requirements such as observation angle, imaging scale, and gimbal attitude during the flight path generation stage. Therefore, there is an urgent need for an adaptive flight path planning method that can balance flight safety, dynamic smoothing, and high-quality inspection imaging under complex terrain and strict electrical safety constraints. Summary of the Invention

[0003] To address the problem that existing technologies struggle to balance flight safety, dynamic smoothness, and high-quality inspection imaging under complex terrain and strict electrical safety constraints, this application proposes a terrain-adaptive inspection route planning method and device for power transmission channel UAVs, enabling adaptive generation and multi-dimensional collaborative optimization of UAV inspection routes for power transmission channels.

[0004] To achieve the above objectives, this application adopts the following technical solution: A method for planning terrain-adaptive inspection routes for unmanned aerial vehicles (UAVs) along power transmission channels includes: acquiring multi-source spatial data and inspection task data of the target power transmission channel; generating a terrain-adaptive control grid for the target power transmission channel based on the multi-source spatial data and the inspection task data; determining the continuous terrain elevation of the target power transmission channel based on a neural implicit terrain model and the terrain-adaptive control grid, and constructing a safe flight envelope using a neural symbolic distance field model in conjunction with electrical safety enhancement constraints and dynamic flight control safety margins; performing terrain-following flight control using a nonlinear model predictive control model under the constraints of the safe flight envelope to generate a control waypoint sequence; and optimizing the control waypoint sequence using a diffusion trajectory optimization model, guided by imaging quality, to generate an inspection trajectory.

[0005] Preferably, generating the terrain-adaptive control grid for the target power transmission channel includes: determining a comprehensive grid refinement index for each grid cell in the initial control grid based on terrain complexity, power corridor constraint strength, and inspection target density; determining a target grid scale for each grid cell between a preset maximum grid scale and a preset minimum grid scale based on the comprehensive grid refinement index; and performing refinement or merging operations on the initial control grid based on the target grid scale to obtain the terrain-adaptive control grid.

[0006] Preferably, the terrain complexity is obtained by weighted summation of normalized elevation undulation index, normalized slope index, and normalized curvature index according to their respective weights; the slope index is determined by the magnitude of the first gradient of the terrain elevation function output by the neural implicit terrain model, and the curvature index is determined by the norm of the second derivative matrix of the terrain elevation function; the power corridor constraint strength is obtained by weighted summation of tower proximity, conductor proximity, obstacle density, and historical hidden dangers.

[0007] Preferably, the electrical safety enhancement constraint includes an electrical safety enhancement distance, which includes: a basic electrical safety distance determined by the voltage level of the transmission line, a safety compensation distance caused by changes in conductor sag, a safety compensation distance caused by wind deflection, and a safety compensation distance caused by UAV positioning errors.

[0008] Preferably, the spatial sampling points within the safe flight envelope satisfy the following safe flight determination conditions: the safe distance estimate output by the neural symbol distance field model for the spatial sampling points, after subtracting the electrical safety enhancement distance and the dynamic flight control safety margin, is not less than a preset minimum safe distance threshold; the boundaries of the safe flight envelope include: a lower boundary for preventing the UAV from flying too low, an upper boundary for preventing the UAV from exceeding the mission's allowed altitude, a lateral boundary for limiting the UAV's lateral offset range, and a locally contracting boundary for improving the safety margin of key risk areas.

[0009] Preferably, before employing a nonlinear model predictive control model for terrain-following flight control, the method further includes: extracting candidate waypoints within the safe flight envelope and determining an initial altitude; establishing connectivity between the candidate waypoints to select an initial master inspection waypoint sequence; inserting mission-critical waypoints into the initial master inspection waypoint sequence to generate an initial terrain-following waypoint sequence; the control waypoint sequence is generated based on the initial terrain-following waypoint sequence.

[0010] Preferably, the method of using a nonlinear model predictive control model for terrain-following flight control includes: constructing a control objective function, wherein the control objective function jointly optimizes terrain-following error, safety envelope constraint error, flight stability error, and inspection imaging reserved error; and based on the control objective function, performing rolling optimization on waypoint altitude, flight speed, and attitude changes under the conditions of the safety flight envelope and UAV flight control execution constraints.

[0011] Preferably, the rolling optimization includes: selecting a control window along the direction of the initial terrain-following waypoint sequence, and dynamically adjusting the spacing of predicted points within the control window according to the terrain undulations and obstacle distribution ahead; after completing joint optimization within the control window, only the control results of the previous segment in the current window are written into the initial terrain-following waypoint sequence; in response to the UAV moving along the flight path, the control window is moved forward, and the terrain, safe flight envelope, and inspection target information are reread to perform optimization again.

[0012] Preferably, the imaging quality includes the following indicators: target imaging distance suitability, observation angle suitability, imaging scale suitability, image overlap rate suitability, and gimbal attitude reachability.

[0013] Preferably, the optimization of the control waypoint sequence using the diffusion trajectory optimization model includes: in the iterative backpropagation step of the diffusion trajectory optimization model, introducing the gradient of the imaging quality score relative to the inspection trajectory as a guiding term to progressively correct and denoise the candidate trajectory; in each diffusion backpropagation step, after the inspection trajectory of the current diffusion backpropagation step is processed by the diffusion trajectory optimization model, the product of the imaging quality guiding intensity coefficient and the gradient of the imaging quality score relative to the inspection trajectory is superimposed to obtain the inspection trajectory of the next diffusion backpropagation step; during the correction process, if the generated waypoint exceeds the safe flight envelope or causes the gimbal angle to be unreachable, then constrained projection or resampling is performed on the candidate trajectory.

[0014] Preferably, after generating the inspection trajectory, the method further includes: allocating shooting trigger points according to the type of inspection target; for fine inspection objects, setting shooting trigger points at positions where the imaging quality score is higher than a preset threshold; for continuous inspection objects, setting continuous shooting trigger points according to image overlap rate requirements; and calculating the gimbal pitch angle and gimbal yaw angle based on the spatial relationship between the waypoint position and the inspection target position.

[0015] Furthermore, this application also provides a terrain-adaptive inspection route planning device for UAVs in power transmission channels, applicable to the terrain-adaptive inspection route planning method for UAVs in power transmission channels as described above, comprising: a data acquisition module for acquiring multi-source spatial data and inspection task data of the target power transmission channel; a grid generation module for generating a terrain-adaptive control grid for the target power transmission channel based on the multi-source spatial data and the inspection task data; a safety envelope construction module for determining the continuous terrain elevation of the target power transmission channel based on a neural implicit terrain model and the terrain-adaptive control grid, and constructing a safety flight envelope through a neural symbolic distance field model in conjunction with electrical safety enhancement constraints and dynamic flight control safety margin; a control module for performing terrain-following flight control using a nonlinear model predictive control model under the constraints of the safety flight envelope, generating a control waypoint sequence; and a trajectory optimization module for optimizing the control waypoint sequence using a diffusion trajectory optimization model guided by imaging quality, generating an inspection trajectory.

[0016] Beneficial effects: The method and apparatus for planning UAV terrain-adaptive inspection routes for power transmission channels provided in this application combine a neural implicit terrain model with a terrain adaptive control grid to convert discrete terrain data into a continuous differentiable elevation field. The grid scale is dynamically adjusted according to the terrain complexity, power corridor constraints and inspection target density, which solves the problem of balancing accuracy and efficiency in complex terrain under fixed grid modeling and provides a refined spatial benchmark for safety constraint calculation.

[0017] Furthermore, a safe flight envelope is constructed through a neural symbolic distance field model, and electrical safety enhancement constraints and dynamic flight control safety margins are integrated into the safe distance expression. This achieves unified and continuous modeling of terrain obstacles, electrical characteristics of power transmission equipment, and UAV dynamic performance, overcoming the shortcomings of traditional geometric obstacle avoidance methods that cannot accurately describe the real flyable space of power transmission channels, and significantly improving the inherent safety of flight path planning.

[0018] Furthermore, nonlinear model predictive control is used for look-ahead terrain-following flight control, and inspection imaging reserved error is introduced into the objective function. This allows the UAV to adjust its altitude and speed in advance based on the terrain trend ahead, ensuring safety envelope constraints while avoiding abrupt maneuvers when approaching the target, thus ensuring the dynamic smoothness of the flight trajectory and the stability of the imaging preparation state.

[0019] Finally, by using the diffusion trajectory optimization model and guided by the imaging quality score gradient, multi-dimensional imaging indicators such as target observation distance, angle, scale, overlap rate, and gimbal accessibility are directly injected into the trajectory generation process. This achieves deep coupling optimization between flight safety constraints and high-quality inspection imaging requirements, effectively solving the problem of poor observation results caused by the decoupling of flight path planning and image acquisition quality in traditional methods, and improving the data acquisition efficiency of intelligent inspection of power transmission lines. Attached Figure Description

[0020] Figure 1 This is a flowchart of the UAV terrain-adaptive inspection route planning method for power transmission channels according to an embodiment of this application; Figure 2 This is a schematic diagram of a three-dimensional scene and terrain adaptive control mesh of a power transmission channel according to an embodiment of this application; Figure 3 This is a schematic diagram illustrating the generation of the safe flight envelope and initial waypoint sequence according to an embodiment of this application; Figure 4 This is a comparison diagram of the flight path before and after terrain-following control and imaging optimization in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0023] Example 1 like Figure 1 As shown in the figure, this embodiment provides a method for planning UAV terrain-adaptive inspection routes for power transmission channels. This method realizes the generation of autonomous inspection routes for UAVs under complex terrain and strict electrical safety constraints through a multi-stage collaborative optimization mechanism.

[0024] S1 acquires multi-source spatial data and inspection task data for the target transmission corridor. Specifically, the multi-source spatial data includes, but is not limited to, high-precision digital terrain models, transmission line corridor boundaries, tower coordinates and 3D models, conductor spatial curves, corridor obstacle distribution, and information on restricted and prohibited flight zones. The inspection task data covers inspection start and end points, tower numbers to be inspected, image acquisition resolution requirements, and sensor parameters. After coordinate unification and outlier removal, this data forms the standardized input basis for subsequent algorithm execution, ensuring consistency between the planning results and the actual physical environment.

[0025] S2, based on multi-source spatial data and inspection task data, generates a terrain-adaptive control grid for the target power transmission corridor. In this step, instead of uniformly dividing the entire corridor with a fixed-resolution grid, the system dynamically adjusts the scale of the grid cells according to the degree of terrain undulation, the strength of power corridor constraints, and the density of inspection targets. For example, in areas with dense towers, abrupt terrain changes, or high-risk obstacles, a high-density fine grid is automatically generated to improve modeling accuracy; while in areas with gentle terrain and sparse targets, a low-density coarse grid is used to reduce computational redundancy. This adaptive mechanism effectively balances the accuracy and efficiency of route planning, providing a refined spatial benchmark for subsequent safety constraint calculations.

[0026] S3, based on a neural implicit terrain model and terrain adaptive control grid, determines the continuous terrain elevation of the target power transmission channel. Combining electrical safety enhancement constraints and dynamic flight control safety margins, a safe flight envelope is constructed using a neural symbolic distance field model. Specifically, the neural implicit terrain model transforms discrete terrain elevation points into a continuously differentiable elevation field, allowing terrain slope and curvature at any location to be obtained through gradient calculation, overcoming the non-differentiability limitation of traditional discrete digital elevation models (DEMs) at interpolation points. Building upon this, the neural symbolic distance field model not only integrates the geometric distances between terrain and obstacles but also introduces electrical safety enhancement constraints. These constraints comprehensively consider multiple factors, such as the basic safety distance determined by voltage level, conductor sag changes, wind deflection, and UAV positioning errors, rather than simply geometrically expanding obstacles. Simultaneously, the dynamic flight control safety margin conservatively expands the safety boundary based on the UAV's current braking capability and attitude response characteristics. The resulting safe flight envelope is a continuous set of flyable spaces containing multiple physical constraints, fundamentally ensuring that the planned flight path meets the specific safety requirements of the power transmission channel.

[0027] S4, under the constraints of the safe flight envelope, performs the generation of the initial waypoint sequence and terrain-following flight control. Specifically, candidate waypoints are first extracted within the safe flight envelope and connectivity is established to generate an initial terrain-following waypoint sequence that satisfies basic navigability and safety constraints, serving as feasible initial values ​​for subsequent optimization. Then, a nonlinear model predictive control model is used for terrain-following flight control to generate a control waypoint sequence. This step focuses on solving the real-time problems of flight dynamics smoothing and terrain following. Unlike static altitude adjustment that relies solely on current location terrain information, the nonlinear model predictive control model uses a look-ahead window to predict the terrain trend ahead and changes in the safe envelope boundary, and jointly optimizes terrain-following error, safe envelope constraint error, flight stability error, and inspection imaging reserved error in the objective function. Through a rolling optimization mechanism, the UAV can adjust its altitude and speed in advance before entering steep slopes or narrow passages, avoiding abrupt maneuvers when approaching targets and ensuring the dynamic feasibility and stability of the flight trajectory.

[0028] S5, guided by imaging quality, employs a diffusion trajectory optimization model to optimize the control waypoint sequence and generate an inspection trajectory. The diffusion trajectory optimization model incorporates the imaging quality score as a conditional guide into the inverse denoising process of trajectory generation. This means that, under the premise of satisfying the aforementioned safe flight envelope and dynamic constraints, the algorithm will actively search for and converge to trajectory shapes that yield better observation results. For example, it automatically adjusts waypoint positions to place the insulator string at the optimal shooting angle, or optimizes the continuity of conductor tracking while ensuring a safe distance. The final output inspection trajectory is not only a safe flight path but also a set of operational plans that directly guide the UAV to acquire high-quality inspection data.

[0029] After completing the above optimizations, the method also includes converting the generated inspection trajectory into an inspection route task package executable by the flight control system, and performing a final safety and reachability verification. Through the technical approach of continuous safety field construction and multi-stage collaborative optimization, this embodiment effectively solves problems in existing technologies such as insufficient terrain modeling accuracy, single safety constraints, lagging terrain-following control, and decoupling of flight routes and imaging quality, significantly improving the safety and intelligence level of UAV inspections of power transmission channels.

[0030] Example 2 In some embodiments, generating a terrain-adaptive control grid for the target power transmission channel includes: determining a comprehensive grid refinement index for each grid cell in the initial control grid based on terrain complexity, power corridor constraint strength, and inspection target density; determining a target grid scale for each grid cell between a preset maximum grid scale and a preset minimum grid scale based on the comprehensive grid refinement index; and performing refinement or merging operations on the initial control grid based on the target grid scale to obtain a terrain-adaptive control grid. This embodiment solves the problem of balancing accuracy and efficiency in complex power transmission channel scenarios using a multi-dimensional feature fusion adaptive mechanism. Specifically, a larger grid scale is used in areas with flat terrain and no important inspection targets to reduce computational redundancy, while the grid is automatically densified in areas with drastic terrain changes, dense towers, or high-risk obstacles, thereby providing a refined spatial reference for subsequent safe flight envelope construction.

[0031] To achieve the aforementioned adaptive adjustment, a continuously differentiable terrain representation is first required. Unlike traditional discrete digital elevation models (DEMs), this embodiment employs a neural implicit terrain model to fit the terrain elevation. This is because subsequent calculations of terrain slope and curvature require the elevation field to possess continuous differentiability, while discrete raster models are often non-differentiable or have discontinuous derivatives at interpolation points, failing to meet the requirements for refined complexity assessment. Specifically, the neural implicit terrain elevation function is represented as follows: ,in, This represents the predicted terrain elevation output by the neural implicit terrain model. The parameter is The neural implicit terrain elevation function, Let represent the parameters of the neural network model, where x represents the horizontal coordinate in the local coordinate system of the transmission channel, and y represents the vertical coordinate in the local coordinate system of the transmission channel. The training objective of this model is to minimize the deviation between the predicted elevation and the actual elevation, and its loss function is: ,in, The terrain elevation fitting loss is represented by M, where M represents the number of terrain sampling points used in training, and j represents the index of the terrain sampling point. This represents the lateral coordinate of the j-th terrain sampling point. This represents the longitudinal coordinate of the j-th terrain sampling point. This represents the true terrain elevation of the j-th terrain sampling point. This represents the predicted terrain elevation output by the neural implicit terrain model at the j-th terrain sampling point. Using this model, the terrain elevation and its derivative at any location can be accurately queried, laying the foundation for subsequent index calculations.

[0032] Furthermore, the terrain complexity is obtained by weighted summation of the normalized elevation undulation index, normalized slope index, and normalized curvature index according to their respective weights. The specific calculation formula is as follows: ,in Indicates the first The terrain complexity of each grid cell, , , These represent the normalized elevation undulation index, slope index, and curvature index, respectively. , , These are the corresponding weighting coefficients. Among them, the elevation fluctuation index... Used to characterize the relative elevation difference within the grid. and The first Maximum and minimum terrain elevations within each grid cell This is the characteristic scale of the grid cell. The slope index is determined by the magnitude of the first-order gradient of the neural implicit terrain elevation function, as shown in the formula: ,in Indicates the first Slope index of each grid cell This indicates the number of sampling points used to calculate the slope within the grid cell. Indicates the sampling point number. and Indicates the first The coordinates of each sampling point This represents the first gradient of the elevation function. This represents the magnitude of the gradient. The curvature index is determined by the norm of the second derivative matrix of the terrain elevation function, as shown in the formula: ,in Indicates the first Curvature index of each grid cell The matrix representing the second derivative of the elevation function. This represents the matrix norm. By introducing the first-order gradient and the second-order derivative, it is possible to accurately capture the steepness and morphological changes of the terrain, which is difficult to achieve with discrete DEMs.

[0033] Meanwhile, the constraint strength of the power corridor is obtained by weighted summation of tower proximity, conductor proximity, obstacle density, and historical hazard weights. The specific formula is as follows: ,in Indicates the first The power corridor constraint strength of each grid cell and These represent the normalized tower proximity and conductor proximity, respectively. This represents the normalized obstacle density. This represents the normalized weight of historical risks. to This represents the corresponding weighting coefficient. This indicator reflects the degree of spatial correlation between grid cells and power transmission equipment and risk sources. The higher the correlation, the more stringent the requirements for inspection accuracy and safety constraints in this area, and the higher the priority of grid refinement should be.

[0034] More often, in power transmission line inspection tasks, objects such as poles, conductors, insulators, fittings, tree obstructions, foreign objects, and potential external damage points have different inspection priorities. This step calculates the inspection target density of the grid cells based on the inspection target type and task priority, so that subsequent grid division adaptively concentrates on high-value inspection areas.

[0035] The density of inspection targets can be expressed as: ,in, This represents the inspection target density of the i-th grid cell. This represents the number of inspected targets within the i-th grid cell, where m represents the index of the inspected target within the i-th grid cell. This represents the target type weight of the m-th inspection target. This represents the task priority weight of the m-th inspection target. A higher density of inspection targets indicates that there are more objects within the grid cell that need to be observed or inspected in detail, and the corresponding area should receive a higher refinement priority in subsequent adaptive grid partitioning.

[0036] After obtaining the above sub-indicators, the comprehensive grid refinement index... Normalized terrain complexity Normalized power corridor constraint strength and normalized inspection target density The decision is made jointly, and the formula is as follows: ,in , , These are the weighting coefficients for the three factors. Based on this index, the target grid scale... At the preset maximum grid size and preset minimum grid size The dynamic determination between them is as follows: , The larger, The smaller the value, the more adaptive the grid density can be.

[0037] Furthermore, to avoid discontinuities or drastic jumps in the generated waypoint sequence due to excessive differences in adjacent grid scales, this embodiment also performs neighborhood smoothing on the target grid scale. The smoothing formula is as follows: ,in This indicates the target mesh size after smoothing. This is the grid-scale smoothing coefficient, with a value ranging from 0 to 1. Indicates the first The set of adjacent grid cells of a grid cell. Indicates the number of adjacent grid cells. Indicates the sequence number of adjacent grid cells. Indicates the first The target grid scale is determined by adjacent grid cells. This weighted averaging method preserves the detail requirements of local areas while ensuring the smoothness of the overall grid transition, effectively improving the feasibility of subsequent route planning and flight safety.

[0038] After obtaining the smoothed target grid scale, adaptive refinement or merging operations are performed on the initial control grid. Specifically, the system determines each grid cell based on the relationship between the smoothed target grid scale and a preset threshold as follows: when the smoothed target grid scale is smaller than the grid refinement threshold, the grid cell is refined, dividing it into smaller control grids; when the smoothed target grid scale is larger than the grid merging threshold, the grid cell is merged, combining it with adjacent grid cells into a larger control grid; when the smoothed target grid scale is between the two thresholds, the grid cell remains unchanged. This adjustment rule can be expressed as: ; in, Represents the i-th grid cell. This indicates that a refinement operation is performed on the i-th grid cell. This indicates that a merge operation is performed on the i-th grid cell. This represents the target mesh scale after smoothing the i-th mesh cell. This indicates the threshold for the mesh refinement scale. This represents the grid merging scale threshold. The above rules ensure that the final generated control grid maintains a higher resolution in key inspection areas and a lower resolution in ordinary areas, thus balancing computational efficiency and flight path planning accuracy.

[0039] After mesh refinement and merging, the resulting terrain-adaptive control mesh set is output. The final control mesh can be represented as: ; in, This represents the final set of terrain-adaptive control meshes generated. This represents the i-th terrain adaptive control grid cell. This represents the total number of terrain-adaptive control grid cells generated. This terrain-adaptive control grid serves as input to the subsequent safe flight envelope construction step, used to construct the terrain, conductors, towers, obstacles, and electrical safety distance constraints within the transmission channel at different grid scales, and further generate the safe flight envelope.

[0040] Example 3 In some embodiments, the electrical safety enhancement constraint includes an electrical safety enhancement distance, which comprises: a basic electrical safety distance determined by the transmission line voltage level, a safety compensation distance caused by conductor sag variation, a safety compensation distance caused by wind deflection, and a safety compensation distance caused by UAV positioning errors. This design is one of the core features that distinguishes this application from general UAV obstacle avoidance planning. In conventional path planning, obstacles are usually treated as static geometry, requiring only a fixed expansion radius. However, conductors within transmission channels are live bodies, and their safety boundaries depend not only on geometric dimensions but also on the coupling effect of electrical characteristics and environmental dynamics. Specifically, the formula for calculating the electrical safety enhancement distance is expressed as: ; in, Indicates the enhanced electrical safety distance. This indicates the basic electrical safety distance determined by the voltage level of the transmission line. This value is determined according to relevant national or industry electrical safety work regulations. For example, different minimum safe working distances correspond to 110kV, 220kV, or 500kV lines. This indicates the safety compensation distance caused by changes in conductor sag. Taking into account the differences in conductor sag under different temperatures, loads, and spans, this compensation ensures that the drone will not intrude into the safety clearance even under maximum sag conditions. This indicates the safety compensation distance caused by wind deflection, used to offset the risk of electrical breakdown caused by the lateral swaying or drifting of the conductor due to strong winds. This indicates the safety compensation distance caused by the drone's positioning error; Indicates the safety distance weighting for voltage levels. Indicates the conductor sag compensation weight. Indicates the windage compensation weight. This represents the positioning error compensation weight. Through the above four-dimensional compensation mechanism, the safety boundary is no longer a simple geometric envelope, but a dynamic safety domain that integrates electrical physical laws and system error models, fundamentally avoiding discharge accidents caused by neglecting electrical characteristics or blind spots in inspections caused by excessive conservatism.

[0041] To effectively integrate the aforementioned electrical safety enhancement constraints into a continuous spatial representation, this embodiment employs a neural symbolic distance field model to construct the fundamental field function of the safe flight envelope. Specifically, the neural symbolic distance field model... The feature is trained to fit the base field and embed secure enhancement features, and its expression is: ,in, This represents the estimated safe distance from the neural symbol distance field output. The parameter is The neural symbol distance field model, This represents the parameters of the neural symbol distance field model. This represents the coordinates of the sampling point in the three-dimensional space of the power transmission channel. This represents the security enhancement feature vector associated with the sampling point. It is worth emphasizing that the feature vector... It's not a simple category label, but a dense vector containing multi-dimensional semantic information such as object type, voltage level, sag compensation parameters, obstacle attributes, and safety control level. It is precisely through this... The injection of [something] enables the same neural network to distinguish the essential differences in safety distance requirements between 220kV live conductors and ordinary trees, thereby accurately expressing the differentiated safety constraints of heterogeneous risk objects in a unified continuous field.

[0042] Furthermore, the spatial sampling points within the safe flight envelope satisfy the following safe flight criterion: the safe distance estimate output by the neural symbolic distance field model for the spatial sampling points, after subtracting the electrical safety enhancement distance and the dynamic flight control safety margin, is not less than a preset minimum safe distance threshold. This criterion can be expressed by the inequality: ; in, For the original safe distance estimate of the aforementioned neural symbol distance field output, Indicates the enhanced electrical safety distance. Indicates the dynamic flight control safety margin. This represents the preset minimum safe distance threshold. The dynamic flight control safety margin introduced here... This is to compensate for the inability of a static safety field to reflect the real-time motion capabilities of a drone. The calculation formula is as follows: ; in, This indicates the emergency braking distance compensation determined by the current flight speed and braking performance; and These represent altitude adjustment capability compensation calculated based on the maximum climb rate and the maximum descent rate, respectively, to prevent the drone from hitting the ground or crashing into obstacles due to sudden terrain changes that prevent it from being unable to climb or descend in time. This indicates attitude adjustment capability compensation, reserving sufficient space for gimbal pointing adjustment or fuselage attitude stabilization; to These are the corresponding weighting coefficients. This inequality essentially defines a feasible safety region: a point is considered safe to fly only if the original safety distance minus all electrical and dynamic risk reserves still has a remainder. This approach achieves deep coupling between environmental constraints and platform capabilities, ensuring that the planned flight path is not only theoretically safe but also sufficiently robust in actual execution.

[0043] Based on the aforementioned safe flight criteria, the boundaries of the safe flight envelope include: a lower boundary to prevent the UAV from flying too low, an upper boundary to prevent the UAV from exceeding the mission's permissible altitude, a lateral boundary to limit the UAV's lateral deviation range, and a locally contracting boundary to improve the safety margin of key risk areas. Combined with... Figure 3 The diagram illustrating the safe flight envelope and initial waypoint sequence generation provides a more intuitive understanding of the physical formation mechanism of these four boundaries. For example... Figure 3 As shown in the partial cross-section in the lower left corner, the lower boundary is not simply parallel to the ground, but rather an undulating surface formed by the combined effects of terrain elevation, tree height, and minimum ground clearance, effectively avoiding collision risks from ridges, valleys, and vegetation. The upper boundary is determined by a combination of airspace height restrictions, the effective imaging height of sensors, and the maximum permissible altitude for the mission, preventing the drone from flying too high, resulting in insufficient image resolution or intrusion into no-fly zones. The lateral boundary extends along the transmission corridor and is jointly defined by the corridor boundary, tower avoidance zones, and electrical safety distances to conductors, ensuring that the drone always operates within the legal inspection channel. The locally contracted boundary is reflected in high-risk areas, due to... or As the numerical value in this region increases, the spatial range satisfying the decision inequality automatically shrinks inward, thus adaptively providing higher safety redundancy for high-risk targets without changing global parameters. It should be understood that these four boundaries are not independent geometric planes, but rather continuous isosurfaces naturally formed in three-dimensional space by the same safety decision inequality. Together, they enclose a real flyable space that satisfies both electrical safety and flight dynamics, providing a solid safety benchmark for subsequent waypoint generation and trajectory optimization.

[0044] Example 4 In some embodiments, a nonlinear model predictive control model is used for terrain-following flight control, including: constructing a control objective function; jointly optimizing terrain-following error, safety envelope constraint error, flight stability error, and inspection imaging reserved error based on the control objective function; and performing rolling optimization on waypoint altitude, flight speed, and attitude changes under the conditions of safety flight envelope and UAV flight control execution constraints. This embodiment solves the problem that traditional terrain-following flight methods only focus on current terrain matching while ignoring future mission requirements and dynamic smoothness by introducing a multi-dimensional joint optimization objective and a look-ahead rolling mechanism. Specifically, this step uses the initial terrain-following waypoint sequence generated in Embodiment 1 as the base reference trajectory, and uses a nonlinear model predictive control (NMPC) algorithm to predict and optimize the future flight state in a finite time domain, so that the generated control waypoint sequence not only meets the immediate safe passage requirements but also has preparatory characteristics for inspection missions.

[0045] To achieve the aforementioned forward-looking control, a predictive state model for the UAV is first required. This model describes the state transition law of the UAV at discrete time steps, expressed as follows: ,in Indicates the first The predicted state of the drone in each prediction step. Indicates the first The current state of the drone in the predicted step. Indicates the first The control quantity applied in each prediction step This represents the function for predicting the motion state of the UAV. In its specific implementation, the state vector... Typically includes information such as three-dimensional position, flight altitude, flight speed, heading angle, and attitude angle; control variables This includes commands such as altitude adjustment rate, acceleration, and turning rate. Through this model, the system can predict the flight status at several future moments within the control window, providing a quantitative basis for subsequent cost assessment and constraint checks.

[0046] Furthermore, the control objective function The design of this embodiment is the core innovation, and its complete expression is: ; in, This represents the value of the control objective function predicted by the nonlinear model constrained by the inspection imaging. This indicates the total number of prediction steps within the look-ahead control window. This indicates the prediction step number. Indicates the first The terrain following error for each prediction step is used to measure the deviation between the actual predicted altitude of the UAV and the expected terrain-following altitude, ensuring that the flight path conforms to the terrain undulations. Indicates the first The safe flight envelope constraint error for each prediction step increases sharply when the predicted state approaches or exceeds the safe flight envelope boundary constructed in Example 3, thus playing a soft constraint penalty role. Indicates the first The flight stability error of each prediction step is usually composed of the altitude difference, rate of change of velocity, or rate of change of attitude between adjacent prediction steps, and is designed to suppress track oscillations and violent maneuvers. Indicates the first The error allowance for each prediction step's inspection imaging is a key factor that distinguishes it from general UAV path planning. , , , These are the weighting coefficients for the four errors mentioned above, which can be dynamically adjusted according to the task stage.

[0047] It is particularly important to note that the inspection imaging has a tolerance for error. The physical significance of this is to reserve potential energy in advance for subsequent refined inspection actions. In power transmission channel inspections, drones often need to adjust their gimbal pointing, flight lateral offset, and relative speed before reaching the towers or insulator strings to meet optimal shooting conditions. If relying solely on traditional terrain following and obstacle avoidance, the drone may only begin to drastically adjust its attitude or position near the target point, which can easily lead to blurred images, missed shots, and even trigger safety protection due to instantaneous overload. By introducing this term into the NMPC objective function, the system can predict whether there are high-value inspection targets ahead within the look-ahead window and begin fine-tuning the waypoint altitude and speed hundreds of meters in advance. This ensures that the drone is in a stable and suitable observation state when entering the shooting area, effectively achieving deep coupling between flight control and mission payload requirements.

[0048] When solving the above objective function, the following constraints must be strictly satisfied: .in, This means that all predicted state points must be within the safe flight envelope set constructed in Example 3, which is the bottom line for ensuring electrical safety and physical obstacle avoidance; This means that the control variables must be within the UAV's permissible action space, including dynamic constraints such as maximum rate of climb, maximum rate of descent, maximum turn rate, and acceleration / deceleration limits. By simultaneously incorporating environmental safety constraints and platform capability constraints into the optimization problem, it is ensured that the output control waypoint sequence is theoretically optimal and engineering-feasible.

[0049] Furthermore, rolling optimization includes: selecting a control window along the initial waypoint sequence and dynamically adjusting the spacing of predicted points within the control window based on terrain undulations and obstacle distribution; after completing joint optimization within the control window, only the control results from the previous segment of the current window are written into the initial waypoint sequence; in response to the UAV moving along the flight path, the control window is moved forward, and terrain, safe flight envelope, and inspection target information are reread to perform optimization again. This rolling temporal control mechanism gives the system robustness in the face of uncertainty. Specifically, the length of the control window is not fixed but adaptively expands and contracts according to environmental complexity. For example, when a steep ridge, dense tree cover, or tower structure is detected ahead, the system automatically reduces the spacing between predicted points and increases the sampling density within the window to improve the fitting accuracy and safety margin of the local trajectory; conversely, above long straight passages or open water, the spacing between predicted points is increased to reduce computational load. After each optimization iteration, the system does not accept the solution for the entire window entirely. Instead, it only adopts the control results from the previous stage and writes them into the waypoint sequence. Then, the window slides forward with the actual position of the UAV and uses the latest sensor data or map information for the next round of optimization. This method effectively overcomes the shortcomings of global planning in dealing with sudden local obstacles or model mismatches, while also avoiding the cumulative errors caused by long line-of-sight planning, ensuring that the UAV can both see far and move steadily in complex power transmission channel environments.

[0050] Example 5 In some embodiments, guided by imaging quality, a diffusion trajectory optimization model is used to optimize the control waypoint sequence to generate an inspection trajectory. This step follows the look-ahead terrain-following control waypoint sequence output in Embodiment 4, aiming to solve the deep coupling problem between safe flight and high-quality data acquisition. Unlike traditional methods that treat the shooting action as an additional step after the flight path generation, this embodiment uses generative artificial intelligence technology to directly inject multi-dimensional imaging indicators into the reverse denoising process of trajectory generation, so that the final generated inspection trajectory actively converges to the optimal observation state while satisfying the safety envelope and dynamic constraints.

[0051] Specifically, image quality includes the following indicators: target imaging distance suitability, observation angle suitability, imaging scale suitability, image overlap rate suitability, and gimbal attitude reachability. To quantify these indicators to support gradient optimization, this embodiment constructs an image quality scoring function. The calculation formula is as follows: ; in, This indicates the overall image quality score; This indicates the suitability of the target imaging distance, used to assess whether the distance between the UAV and the inspection target is within the optimal focus and resolution range of the sensor; This indicates the suitability of the observation angle, used to measure whether the angle between the line-of-sight vector and the normal vector of the target surface meets the requirements for defect identification; This indicates the suitability of the imaging scale, reflecting whether the proportion of the target in the image pixels meets the detection threshold of the algorithm. This indicates the suitability of the image overlap rate, ensuring that the continuously acquired image sequence meets the requirements of 3D reconstruction or panoramic stitching. This indicates the gimbal's attitude reachability and penalizes waypoints that, although geometrically ideal, exceed the gimbal's mechanical limits or cause the gimbal to lock up. to These are the weighting coefficients for the corresponding indicators, which can be dynamically configured according to the type of inspection task. Through this scoring function, the quality of the shooting is transformed into a continuously differentiable numerical expression, providing a mathematical basis for the gradient guidance of the subsequent diffusion model.

[0052] Furthermore, a diffusion trajectory optimization model is used to optimize the control waypoint sequence, including: in the iterative backpropagation step of the diffusion trajectory optimization model, the gradient of the imaging quality score relative to the inspection trajectory is introduced as a guiding term to progressively correct and denoise the candidate trajectories; in each diffusion backpropagation step, the inspection trajectory of the current diffusion backpropagation step is processed by the diffusion trajectory optimization model, and then the product of the imaging quality guiding intensity coefficient and the gradient of the imaging quality score relative to the inspection trajectory is superimposed to obtain the inspection trajectory for the next diffusion backpropagation step. The mathematical expression of this process is: ; in, Indicates the first The inspection trajectory obtained from the diffusion reverse step Indicates the first The inspection trajectory in the reverse diffusion step The parameter is Pre-trained diffusion trajectory denoising model, Indicates the current diffusion step number. This represents the trajectory optimization input, which includes the inspection target type, priority, and environmental context. The key lies in the additional items. : It is the gradient of the image quality score relative to the trajectory parameters, indicating the direction in which adjusting the waypoint position, velocity, or attitude in the current state will result in the fastest improvement in image quality; This is the image quality guiding intensity coefficient, used to balance the relationship between adhering to the original data distribution and pursuing high image quality. When When the value is large, the generated trajectory will more aggressively approach the optimal shooting point; when When the size is smaller, the trajectory tends to be more conservative and smooth. This mechanism allows the diffusion model to no longer merely mimic the operation of a human pilot, but to explore the optimal solution in the potential space that transcends human experience and balances safety with ultimate image quality.

[0053] During the above correction process, if the generated waypoints exceed the safe flight envelope or cause the gimbal angle to become unreachable, constrained projection or resampling is performed on the candidate trajectory. Since the generation of the diffusion model is random, relying solely on gradient guidance may produce intermediate results that violate hard constraints. Therefore, a constraint verification step is added after each reverse step: for paths exceeding the safe flight envelope described in Example 3... The waypoints are projected back into the feasible region along the normal direction of the safety boundary; for attitude commands that cause the gimbal to exceed limits, they are truncated to the nearest feasible angle or local resampling is triggered to find alternative paths. This mechanism ensures that no matter how the generation process explores, the final output trajectory always strictly meets electrical safety and platform physical constraints, achieving a unity of free generation and rigid constraints.

[0054] After obtaining the optimized inspection trajectory, specific execution parameters need to be generated. Shooting trigger points are assigned based on the type of inspection target. For fine-grained inspection objects, shooting trigger points are set at locations where the image quality score exceeds a preset threshold; for continuous inspection objects, continuous shooting trigger points are set according to image overlap requirements. Based on the spatial relationship between waypoint locations and inspection target locations, the gimbal pitch and yaw angles are calculated. This differentiated triggering strategy adapts to the acquisition needs of heterogeneous targets within the transmission channel. For example, for fine-grained inspection objects such as tower insulators and fittings, the system only... The shutter is triggered at waypoints where peak speeds and stability are achieved, avoiding invalid and redundant photos; while for continuously inspected objects such as guide wires and tree obstacles in passageways, the shutter speed is determined based on... The trigger interval is dynamically calculated to ensure full coverage without any omissions. The gimbal attitude is calculated based on the real-time solution of the relative position vector between the UAV and the target, and compensated for by the body attitude to ensure that the camera optical axis is accurately pointed to the center of the target.

[0055] Combination Figure 4The comparison diagram of flight paths before and after terrain-following control and imaging optimization, shown below, provides a visual understanding of the technical effects of this embodiment. The solid line in the diagram represents the imaging-optimized flight path, which, compared to the long dashed line representing only terrain-following control, exhibits significant local adaptive adjustment characteristics near the inspection target marked with a pentagram: the flight path does not simply skim across the target area, but pre-adjusts altitude and lateral offset before entering the target region, maintains a stable, hovering flight attitude within the optimal observation window, and precisely sets the shooting trigger point indicated by the camera icon at the corresponding location. This demonstrates that, guided by the imaging quality gradient, the diffusion trajectory optimization model successfully transforms abstract observation requirements into concrete spatial maneuvering strategies. While ensuring that the target image does not touch the safety envelope dashed line boundary throughout the entire flight, it significantly improves the clarity, integrity, and usability of the target image, resolving the dilemma in traditional planning methods where safe flight results in unclear images or safety margins are sacrificed for shooting.

[0056] Example 6 like Figure 1 As shown, this embodiment provides a terrain-adaptive inspection route planning device for power transmission channel UAVs, applicable to the terrain-adaptive inspection route planning method for power transmission channel UAVs described in the foregoing embodiments. This device achieves fully automated processing from environmental perception to trajectory generation by encapsulating complex route planning algorithms into functionally decoupled processing modules. Specifically, the device includes a data acquisition module, a mesh generation module, a safety envelope construction module, a control module, and a trajectory optimization module. These modules are cascaded through standardized data interfaces, forming a highly efficient data processing pipeline.

[0057] The data acquisition module is used to acquire multi-source spatial data and inspection task data of the target power transmission channel. Specifically, this module is responsible not only for extracting basic information from multi-source heterogeneous data such as lidar point clouds, oblique photogrammetry models, vector maps, and flight control logs, but also for data cleaning and coordinate system unification preprocessing tasks. For example, this module can spatially register terrain data of different precision with line equipment data, remove outliers, and fill in missing areas, thereby constructing a standardized 3D inspection scene model of the power transmission channel, providing a consistent input benchmark for subsequent algorithms. It should be understood that although this module is depicted as a single frame in the figure, in actual implementation, it can also be composed of multiple parallel data preprocessing sub-units to improve throughput efficiency under large data volumes.

[0058] The mesh generation module generates terrain-adaptive control meshes for target power transmission corridors based on multi-source spatial data and inspection task data. This module integrates terrain feature analysis and dynamic mesh adjustment algorithms, automatically calculating the comprehensive mesh refinement index based on terrain complexity, power corridor constraint strength, and inspection target density. During execution, the module receives the scene model output from the data acquisition module, fits a continuous elevation field using a neural implicit terrain model, and determines the target size of each mesh cell within a preset maximum and minimum scale range, ultimately outputting a well-spaced adaptive control mesh. This approach allows the device to effectively reduce the computational load on non-critical areas while ensuring modeling accuracy in key areas.

[0059] The safety envelope construction module is used to determine the continuous terrain elevation of the target power transmission channel based on a neural implicit terrain model and terrain adaptive control grid. It then combines electrical safety enhancement constraints and dynamic flight control safety margins to construct a safe flight envelope using a neural symbolic distance field model. This is the core unit for achieving intrinsic safety in this device. This module utilizes the adaptive grid provided by the grid generation module as a sampling reference, quantifying electrical safety factors such as voltage level, conductor sag, wind deflection, and positioning error into enhanced distances. These enhanced distances are integrated with the UAV's dynamic response capabilities into the inference process of the neural symbolic distance field. Its output safe flight envelope is a continuous set of three-dimensional feasible regions, rather than a discrete list of obstacles, thus providing mathematically continuous and differentiable safety constraint boundaries for subsequent path search and trajectory optimization.

[0060] The control module, under the constraints of the safe flight envelope, employs a nonlinear model predictive control model for terrain-following flight control, generating a sequence of control waypoints. This module receives the safety domain information output by the safety envelope construction module and performs look-ahead rolling optimization within it. Specifically, the control module integrates a UAV dynamics prediction model and a multi-objective cost function solver, enabling joint optimization of multiple metrics within a limited prediction time domain, including terrain following, safety boundary preservation, flight stability, and inspection imaging reservation. By dynamically adjusting the sampling density within the prediction window and applying only the results of previous optimizations, this module can respond in real-time to abrupt changes in the terrain and environment ahead, outputting a sequence of control waypoints that meets both safe passage requirements and exhibits good dynamic smoothness.

[0061] The trajectory optimization module, guided by imaging quality, optimizes the control waypoint sequence using a diffusion trajectory optimization model to generate inspection trajectories. As the device's backend fine-tuning unit, this module converts the imaging quality score into a gradient guidance signal, which is injected into the inverse denoising iteration of the diffusion model. This means the device no longer simply generates a flyable path, but actively explores and converges to trajectory shapes that yield the best observation results. Simultaneously, this module includes a constrained projection mechanism to ensure that, in the pursuit of high imaging quality, the generated trajectory always remains strictly within the safe flight envelope, achieving a rigid balance between safety and quality.

[0062] It should be noted that the above modules can be integrated and deployed in the ground station's server or high-performance workstation for offline generation of high-precision inspection route task packages; or they can be deployed in a lightweight manner in the UAV's onboard computer or edge computing device to support online real-time replanning and dynamic obstacle avoidance. Regardless of the deployment method, the logical relationships and data flow order between the modules remain consistent and are all within the scope of protection of this application. Through the above modular architecture, the device provided in this embodiment transforms the abstract planning method into an engineeringable product entity, significantly improving the intelligence level and safety of UAV inspection operations on power transmission channels.

[0063] Example 7 To further verify the effectiveness of the UAV terrain-adaptive inspection route planning method for power transmission channels provided in this application in real-world complex environments, this embodiment constructs a power transmission channel simulation scenario containing typical mountainous features for comprehensive testing. This simulation scenario covers a continuous power transmission section from tower T1 to tower T9, with a total length of approximately 3.5 kilometers. The area incorporates various terrain features such as steep ridges, deep V-shaped valleys, open water areas, and dense vegetation. The scenario includes not only tower models of nine different tower types and precise conductor sag curves, but also artificially set up multiple tree-obstruction areas and historical external damage hazard points to comprehensively evaluate the algorithm's performance in three dimensions: terrain adaptability, electrical safety assurance, and inspection imaging quality.

[0064] First, regarding the terrain modeling and mesh generation results, combined with Figure 2 The diagram illustrates the 3D scene of the power transmission channel and the terrain-adaptive control mesh. Figure 2The top-down projection of the lower grid clearly shows that the terrain-adaptive control grid generated based on the neural implicit terrain model exhibits a significant non-uniform distribution. Specifically, in the valley area between towers T3 and T4 and the steep slope area near tower T7, due to the high terrain complexity index and power corridor constraint strength, the system automatically refines the grid scale to near the preset minimum value, forming a high-density dark grid area; while in the gentle woodland and above the water area between T1 and T2, the grid remains relatively sparse. The physical essence of this adaptive mechanism lies in dynamically balancing computational resources and environmental representation accuracy through a comprehensive grid refinement index. Compared with the traditional fixed-resolution raster method, this strategy significantly reduces the overall number of grid cells while ensuring that the modeling error of towers and dangerous terrain areas is less than 0.1 meters, thereby significantly reducing the computational load of subsequent safety field construction and trajectory optimization, verifying the efficiency and accuracy of this application in terrain perception.

[0065] Secondly, regarding the effectiveness of constructing a safe flight space, combined with Figure 3 The diagram showing the safe flight envelope and initial waypoint sequence generation is analyzed. Figure 3 The area enclosed by the dashed line in the main view is the safe flight envelope generated by the neural symbolic distance field model. As can be seen from the partial cross-section in the lower left corner, this envelope is not a simple geometric expansion, but a continuous feasible region incorporating multiple constraints. Below the conductor, the upper boundary of the envelope strictly adheres to the electrical safety enhancement constraints. Its vertical distance from the conductor not only includes the basic electrical safety distance but also incorporates sag and wind deflection compensation calculated based on current weather conditions, ensuring that the UAV will not intrude into the discharge risk zone even under extreme conditions. Above tree obstacles, the lower boundary of the envelope undulates with the terrain and canopy height, and additionally retains a dynamic flight control safety margin to prevent collisions caused by positioning drift or braking delay. Particularly noteworthy is the significant local contraction of the safe envelope in the complex crossing area near tower T5, which is a direct reflection of the increased electrical safety enhancement distance in this region. Initial waypoints W1 to W15 are all confined within this envelope, demonstrating that the safety baseline constructed in this application can fundamentally avoid both electrical and physical risks, solving the problem of mismatch between safety boundaries and actual risks in traditional methods.

[0066] Finally, regarding the overall performance of route planning, combined with Figure 4 A detailed evaluation was conducted using the comparison charts of the flight path before and after terrain-following control and imaging optimization. Figure 4The altitude change curves in the lower left corner illustrate the differences between the three flight path morphologies: the broken line connected by hollow circles represents the initial flight path that only meets the requirements for passability, with abrupt altitude changes and no consideration of dynamic constraints; the long dashed line represents the terrain-following control flight path optimized by the nonlinear model predictive control model, showing that its altitude curve has become smooth and continuous, and exhibits a clear forward-looking climb and descent trend when approaching ridges and valleys, verifying the role of the inspection imaging reserved error term in suppressing violent maneuvers; the solid line represents the final inspection trajectory processed by the diffusion trajectory optimization model. This trajectory, while maintaining smoothness, shows fine local adjustments near the inspection targets marked with pentagrams, such as reducing the flight speed and adjusting the lateral offset in advance before the insulator string observation point. The magnified view in the lower right corner shows that the shooting trigger points indicated by the camera icon are precisely distributed at the positions with the highest image quality scores, rather than being mechanically arranged at equal intervals.

[0067] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for terrain-adaptive inspection route planning for unmanned aerial vehicles (UAVs) in power transmission channels, characterized in that, The method includes: Acquire multi-source spatial data and inspection task data of the target power transmission channel; Based on the multi-source spatial data and the inspection task data, a terrain-adaptive control grid for the target power transmission channel is generated. The continuous terrain elevation of the target power transmission channel is determined based on the neural implicit terrain model and the terrain adaptive control grid. Combined with electrical safety enhancement constraints and dynamic flight control safety margin, a safe flight envelope is constructed through the neural symbolic distance field model. Under the constraints of the safe flight envelope, a nonlinear model predictive control model is used for terrain-following flight control to generate a sequence of control waypoints; Guided by imaging quality, the control waypoint sequence is optimized using a diffusion trajectory optimization model to generate an inspection trajectory; The generation of the terrain-adaptive control grid for the target power transmission channel includes: Based on terrain complexity, power corridor constraint strength, and inspection target density, the comprehensive grid refinement index of each grid cell in the initial control grid is determined. Based on the comprehensive grid refinement index, the target grid scale of each grid cell is determined between the preset maximum grid scale and the preset minimum grid scale; Based on the target grid scale, the initial control grid is refined or merged to obtain the terrain adaptive control grid. The terrain complexity is obtained by weighted summation of normalized elevation undulation index, normalized slope index, and normalized curvature index according to their respective weights; the slope index is determined by the magnitude of the first gradient of the terrain elevation function output by the neural implicit terrain model, and the curvature index is determined by the norm of the second derivative matrix of the terrain elevation function. The constraint strength of the power corridor is obtained by weighted summation of tower proximity, conductor proximity, obstacle density, and historical hazard weights.

2. The method for terrain-adaptive inspection route planning of power transmission channels by unmanned aerial vehicles according to claim 1, characterized in that, The electrical safety enhancement constraint includes an electrical safety enhancement distance, which includes: The basic electrical safety distance is determined by the voltage level of the transmission line, the safety compensation distance is caused by the change of conductor sag, the safety compensation distance is caused by the influence of wind deflection, and the safety compensation distance is caused by the positioning error of the UAV.

3. The method for terrain-adaptive inspection route planning of power transmission channels by UAVs according to claim 2, characterized in that, The spatial sampling points within the safe flight envelope meet the following safe flight determination conditions: the safe distance estimate output by the neural symbol distance field model for the spatial sampling points, after subtracting the electrical safety enhancement distance and the dynamic flight control safety margin, is not less than the preset minimum safe distance threshold. The boundaries of the safe flight envelope include: a lower boundary to prevent the UAV from flying too low, an upper boundary to prevent the UAV from exceeding the mission-permitted altitude, a lateral boundary to limit the lateral deviation range of the UAV, and a locally contracting boundary to improve the safety margin of key risk areas.

4. The method for terrain-adaptive inspection route planning of UAVs for power transmission channels according to claim 1, characterized in that, Before employing a nonlinear model predictive control model for terrain-following flight control, the method further includes: Extract candidate waypoints and determine the initial altitude within the safe flight envelope; Establish connectivity between the candidate waypoints to select an initial main inspection waypoint sequence; Insert mission-critical waypoints into the initial main inspection waypoint sequence to generate an initial terrain-following waypoint sequence; The control waypoint sequence is generated based on the initial geodetic waypoint sequence.

5. The method for terrain-adaptive inspection route planning of UAVs for power transmission channels according to claim 4, characterized in that, The method of using a nonlinear model predictive control model for terrain-following flight control includes: A control objective function is constructed, which jointly optimizes the terrain following error, safety envelope constraint error, flight stability error, and inspection imaging reserved error. Based on the control objective function, under the conditions of the safe flight envelope and UAV flight control execution constraints, rolling optimization is performed on waypoint altitude, flight speed, and attitude changes.

6. The method for terrain-adaptive inspection route planning of power transmission channels by unmanned aerial vehicles according to claim 5, characterized in that, The rolling optimization includes: A control window is selected along the direction of the initial waypoint sequence, and the spacing between the predicted points within the control window is dynamically adjusted according to the terrain undulations and obstacle distribution ahead. After completing the joint optimization within the control window, only the control results from the previous stage in the current window are written into the initial geodetic waypoint sequence; In response to the UAV moving along the flight path, the control window is moved forward, and terrain, safe flight envelope, and inspection target information are reread to perform optimization again.

7. The method for terrain-adaptive inspection route planning of power transmission channels by unmanned aerial vehicles according to claim 1, characterized in that, The imaging quality includes the following indicators: The suitability of target imaging distance, observation angle, imaging scale, image overlap rate, and gimbal attitude reachability.

8. The method for planning a terrain-adaptive inspection route for power transmission channels using unmanned aerial vehicles (UAVs) according to claim 7, characterized in that, The optimization of the control waypoint sequence using the diffusion trajectory optimization model includes: In the iterative backpropagation step of the diffusion trajectory optimization model, the gradient of the imaging quality score relative to the inspection trajectory is introduced as a guiding term to progressively correct and denoise the candidate trajectory. In each diffusion backpropagation step, the inspection trajectory of the current diffusion backpropagation step is processed by the diffusion trajectory optimization model and then superimposed with the product of the imaging quality guiding intensity coefficient and the gradient of the imaging quality score relative to the inspection trajectory to obtain the inspection trajectory of the next diffusion backpropagation step. During the correction process, if the generated waypoint exceeds the safe flight envelope or causes the gimbal angle to be unreachable, then the candidate trajectory is subjected to constrained projection or resampling.

9. The method for terrain-adaptive inspection route planning of UAVs for power transmission channels according to claim 8, characterized in that, After generating the inspection trajectory, the method further includes: Shooting trigger points are assigned according to the type of inspection target. For fine inspection objects, shooting trigger points are set at positions where the imaging quality score is higher than a preset threshold. For continuous inspection objects, continuous shooting trigger points are set according to the image overlap rate requirements. Based on the spatial relationship between waypoint location and inspection target location, calculate gimbal pitch angle and gimbal yaw angle.

10. A terrain-adaptive inspection route planning device for power transmission channel UAVs, applicable to the terrain-adaptive inspection route planning method for power transmission channel UAVs as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire multi-source spatial data and inspection task data of the target power transmission channel; The grid generation module is used to generate a terrain adaptive control grid for the target power transmission channel based on the multi-source spatial data and the inspection task data. The safety envelope construction module is used to determine the continuous terrain elevation of the target power transmission channel based on the neural implicit terrain model and the terrain adaptive control grid, and to construct the safety flight envelope through the neural symbolic distance field model by combining electrical safety enhancement constraints and dynamic flight control safety margin. The control module is used to perform terrain-following flight control using a nonlinear model predictive control model under the constraints of the safe flight envelope, and to generate a sequence of control waypoints. The trajectory optimization module is used to optimize the control waypoint sequence using a diffusion trajectory optimization model, guided by imaging quality, to generate an inspection trajectory.

Images