Adaptive path generation method and system for complex terrain photovoltaic array perception

Abstract

The application provides a complex terrain photovoltaic array perception adaptive path generation method and system, relates to the technical field of path generation, and the method comprises the following steps: performing spatiotemporal fusion processing of multi-source perception data, and constructing a three-dimensional space semantic representation model; acquiring constraint parameters to construct a dynamic passability field representing the accessibility difficulty distribution of the space; extracting the spatial arrangement topological structure features of the photovoltaic array, and acquiring the inspection attention data of each photovoltaic array; taking the spatial arrangement topological structure features as the structure guiding factor of path generation, performing global path skeleton optimization under the inspection attention data matching based on the dynamic passability field, and establishing the path optimization result, so that the technical problems that the existing technology has insufficient inspection path rationality and inaccurate traffic risk assessment when the photovoltaic array is inspected under complex terrain are solved. The technical effects of improving the path planning accuracy and inspection execution reliability are achieved through multi-source perception data fusion and dynamic passability analysis.

Description

Technical Field

[0001] This invention relates to the field of path generation technology, and more specifically to an adaptive path generation method and system for photovoltaic array sensing in complex terrain. Background Technology

[0002] In the existing photovoltaic power plant operation and maintenance field, inspection operations typically employ path planning methods based on manual experience or digital maps to design equipment inspection routes. Common methods include shortest path search based on grid maps, rule-constrained channel optimization, and path generation strategies based on simplified terrain models. These methods usually abstract the work area into a two-dimensional plane or regular grid structure, and use fixed traffic rules as the basis for path determination, such as using the shortest distance as the primary optimization objective.

[0003] However, in complex terrain conditions, photovoltaic arrays often exhibit irregular distribution characteristics, with significant slope undulations, uneven array spacing, and local shading. This results in strong spatial heterogeneity in actual access conditions, making it difficult for traditional grid-based modeling to accurately represent changes in accessibility in real-world operating environments. Consequently, discrepancies arise between path evaluation results and actual accessibility. Furthermore, existing technologies often lack constraints on the connectivity and continuity of photovoltaic array structures during path generation. This can lead to generated paths that skip across arrays or contain locally impassable sections, increasing the frequency of adjustments and operational risks during actual execution.

[0004] In summary, existing technologies for inspecting photovoltaic arrays in complex terrain rely on static rules for path generation and do not fully consider accessibility, resulting in insufficient rationality of inspection paths and inaccurate assessment of access risks. Summary of the Invention

[0005] The purpose of this application is to provide an adaptive path generation method and system for photovoltaic array perception in complex terrain, which solves the technical problems of insufficient rationality of inspection paths and inaccurate assessment of access risks when inspecting photovoltaic arrays in complex terrain, where path generation relies on static rules and does not fully consider accessibility.

[0006] In view of the above problems, this application provides an adaptive path generation method and system for photovoltaic array perception in complex terrain.

[0007] The first aspect of this application provides an adaptive path generation method for sensing photovoltaic arrays in complex terrain. This method includes: after accessing a multi-source sensing device in the target photovoltaic array area, reading multi-source sensing data and performing spatiotemporal fusion processing of the multi-source sensing data to construct a three-dimensional spatial semantic representation model containing terrain structural features and photovoltaic array topological features; obtaining constraint parameters of the target operating equipment's motion capability; performing coupled calculations of terrain slope features, array spacing features, and shading influence features at each location in the three-dimensional spatial semantic representation model to construct a dynamic navigability field representing the spatial accessibility difficulty distribution; extracting the spatial arrangement topological features of the photovoltaic array and obtaining inspection attention data for each photovoltaic array; using the spatial arrangement topological features as a structural guiding factor for path generation; performing global path skeleton optimization based on the dynamic navigability field and matching inspection attention data to establish a path optimization result.

[0008] Optionally, the array channel connection relationship characterized by the spatial arrangement topology is discretized into a set of channel nodes and adjacency relationships between nodes, and a structural connectivity graph between channel nodes is constructed based on the adjacency relationships; an initial passage cost is assigned to each channel node based on the dynamic traversability field, and constraint propagation is performed along the adjacency relationships in the structural connectivity graph to construct an initial channel cost mapping relationship; based on the initial channel cost mapping relationship, the array attention intensity corresponding to the inspection attention data is introduced to perform a first round of weight correction on the channel node set based on the influence range of adjacency relationship propagation, forming a first intermediate channel cost distribution; based on the first intermediate channel cost distribution, and using the adjacency relationships between nodes to limit the path expansion direction... The path connectivity constraint propagation is performed on the array channel connection relationship to generate an intermediate path skeleton candidate set containing the reachability propagation results. For each candidate skeleton path in the intermediate path skeleton candidate set, the local cost fluctuation consistency of the path is calculated based on the spatial gradient change characteristics of the dynamic drivability field on the path node sequence. The intermediate path skeleton candidate set is then filtered and sorted according to the local cost fluctuation consistency to obtain a second intermediate path skeleton set. In the second intermediate path skeleton set, the structural consistency correction process is performed on the node connection order of the path skeleton based on the adjacency constraint relationship of the array topology. During the correction process, the corrected cost value of the channel nodes is updated synchronously to form a converged path optimization result.

[0009] Optionally, based on the adjacency relationship between nodes, directed adjacency pairs between adjacent nodes in the candidate skeleton path are determined, and the passage cost difference of each directed adjacency pair in the dynamic drivability field is extracted to construct a cost gradient sequence corresponding to the path node sequence; directional consistency analysis is performed on the cost gradient sequence to identify the same-direction segments and opposite-direction jump segments of cost change direction between consecutive adjacency pairs, and a path local fluctuation structure description is constructed based on the length of the same-direction segment and the frequency of the opposite-direction jump; based on the path local fluctuation structure description, segmented smoothing processing is performed on the cost gradient sequence to weaken the influence of abrupt gradients caused by local abnormal terrain or occlusion, resulting in a corrected cost gradient distribution sequence; according to the continuity of gradient change in each segment of the corrected cost gradient distribution sequence, the local cost fluctuation consistency characterizing path stability is calculated.

[0010] Optionally, a path stability identifier is configured based on the consistency of local cost fluctuations; a local path replacement library is established using the path stability identifier, and the local path replacement library is embedded into the path optimization result.

[0011] Optionally, the constraint parameters are parsed into the maximum passable slope, the minimum safe passage distance, and the maximum allowable occlusion ratio. In the three-dimensional spatial semantic representation model, the terrain slope value, the effective passage distance between adjacent arrays, and the occlusion coverage ratio are extracted for each spatial location, and compared with the maximum passable slope, the minimum safe passage distance, and the maximum allowable occlusion ratio item by item to obtain the individual constraint deviation of each spatial location under the slope constraint, distance constraint, and occlusion constraint. Based on the individual constraint deviation of each spatial location, a constraint triggering state determination result is constructed, and the spatial location is divided into an untriggered constraint area, a single constraint triggered area, and a multi-constraint superimposed triggering area to form a first intermediate constraint distribution. Based on the first intermediate constraint distribution, differentiated passage capability modulation processing is performed on the single constraint triggered area and the multi-constraint superimposed triggering area respectively to establish a dynamic passability field.

[0012] Optionally, for the single constraint triggering region, directional attenuation processing is performed on the drivability of the spatial location based on the corresponding constraint deviation; for the multi-constraint superimposed triggering region, synergistically enhanced drivability suppression processing is performed based on the superposition degree of each constraint deviation to obtain a second intermediate drivability distribution; based on the joint distribution state of the first intermediate constraint distribution and the second intermediate drivability distribution, the degree of difference in constraint triggering state between adjacent locations is determined within the spatial neighborhood, and a hierarchical transfer processing of drivability is performed according to the degree of difference in constraint triggering state to form a continuously changing drivability transition distribution; the drivability transition distribution is mapped to a spatial reachability cost, and the cost is directionally consistently corrected based on the continuity of the change of the drivability transition distribution within the spatial neighborhood to construct a dynamic drivability field.

[0013] Optionally, the multi-source sensing data includes terrain elevation data, array structure spatial distribution data, and occlusion relationship data.

[0014] Optionally, the multi-source sensing device includes terrain sensing device, visual sensing device and environmental sensing device. After configuring a unified device access interface for each sensing device, timestamp identification and spatial coordinate reference benchmark correction are performed to complete the data acquisition initialization.

[0015] Optionally, during the execution process based on the path optimization results, updated environmental perception data is acquired in real time, and a dynamic accessibility field is performed for local reconstruction; the local reconstruction results are used to perform real-time perception correction of the path optimization results.

[0016] The second aspect of this application provides an adaptive path generation system for sensing photovoltaic arrays in complex terrain. This system includes: a sensing data processing module, used to read multi-source sensing data after accessing multi-source sensing devices in the target photovoltaic array area, and to perform spatiotemporal fusion processing of the multi-source sensing data to construct a three-dimensional spatial semantic representation model containing terrain structural features and photovoltaic array topological features; a feature coupling calculation module, used to obtain constraint parameters of the target operating equipment's motion capability, and to perform coupled calculations of terrain slope features, array spacing features, and occlusion influence features at each location in the three-dimensional spatial semantic representation model to construct a dynamic traversability field representing the spatial accessibility difficulty distribution; a feature data acquisition module, used to extract the spatial arrangement topological features of the photovoltaic array and obtain inspection attention data for each photovoltaic array; and an optimization result establishment module, used to use the spatial arrangement topological features as structural guiding factors for path generation, and to perform global path skeleton optimization based on the dynamic traversability field and inspection attention data matching to establish path optimization results.

[0017] One or more technical solutions provided in this application have at least the following technical effects or advantages: The method provided in this application involves reading multi-source sensing data after accessing the target photovoltaic array area using multi-source sensing devices, and performing spatiotemporal fusion processing of the multi-source sensing data to construct a three-dimensional spatial semantic representation model that includes terrain structure features and photovoltaic array topology features. It also involves obtaining constraint parameters for the motion capability of the target operating equipment, performing coupled calculations of terrain slope features, array spacing features, and shading influence features at each location in the three-dimensional spatial semantic representation model, and constructing a dynamic accessibility field to characterize the spatial accessibility difficulty distribution. Furthermore, it involves extracting the spatial arrangement topology features of the photovoltaic array and obtaining inspection attention data for each photovoltaic array. Finally, it uses the spatial arrangement topology features as a structural guidance factor for path generation, performs global path skeleton optimization based on the dynamic accessibility field and the inspection attention data matching, and establishes the path optimization result. By fusing multi-source sensing data and constructing a 3D spatial semantic representation model, a unified expression of terrain structure information and array spatial distribution information in complex terrain photovoltaic array areas is achieved, improving the spatial consistency and representational completeness of heterogeneous sensing data. Through the construction and analysis of a dynamic accessibility field based on equipment motion constraints, a collaborative quantitative expression of slope constraints, spacing constraints, and occlusion constraints is achieved, transforming spatial accessibility from discrete judgment to continuous cost distribution. This improves the refinement of access risk assessment and the safety and reliability of path planning in complex environments. Furthermore, by introducing a global path skeleton optimization mechanism based on topology guidance and inspection focus information matching, the path generation process achieves joint constraint optimization of array structure constraints and task priority information. This enables the generated path to adapt to complex terrain changes and prioritize coverage of key inspection areas, thereby improving the adaptability, accuracy, and overall execution efficiency of the inspection path.

[0018] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1A flowchart illustrating the adaptive path generation method for photovoltaic array perception in complex terrain provided in this application.

[0021] Figure 2 A schematic diagram of the adaptive path generation system for photovoltaic array perception of complex terrain provided in this application.

[0022] Figure labeling: Perception data processing module 11, feature coupling calculation module 12, feature data acquisition module 13, optimization result establishment module 14. Detailed Implementation

[0023] This application provides an adaptive path generation method and system for photovoltaic array sensing in complex terrain. It addresses the technical problems of existing technologies for inspecting photovoltaic arrays in complex terrain, where path generation relies on static rules and fails to adequately consider accessibility, leading to insufficient path rationality and inaccurate access risk assessment. The method achieves the technical effect of adaptively generating inspection paths for photovoltaic arrays in complex terrain through multi-source sensing data fusion and dynamic accessibility analysis, thereby improving the accuracy of path planning and the reliability of inspection execution.

[0024] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0025] Example 1, as Figure 1 As shown, this application provides an adaptive path generation method for photovoltaic array sensing in complex terrain. The adaptive path generation method for photovoltaic array sensing in complex terrain includes: After connecting to the multi-source sensing devices in the target photovoltaic array area, the multi-source sensing data is read and spatiotemporal fusion processing of the multi-source sensing data is performed to construct a three-dimensional spatial semantic representation model that includes terrain structure features and photovoltaic array topology features.

[0026] Furthermore, the multi-source sensing data includes terrain elevation data, array structure spatial distribution data, and occlusion relationship data.

[0027] Furthermore, the multi-source sensing devices include terrain sensing devices, visual sensing devices, and environmental sensing devices. After configuring a unified device access interface for each sensing device, timestamp identification and spatial coordinate reference benchmark correction are performed to complete the data acquisition initialization.

[0028] Specifically, multi-source sensing devices are connected to the target photovoltaic array area. These devices include terrain sensing devices, visual sensing devices, and environmental sensing devices. The terrain sensing devices, such as lidar or RTK measuring instruments, are used to accurately measure the terrain elevation data of the target photovoltaic array area. This elevation data reflects the height information at different locations within the target photovoltaic array area, i.e., the absolute or relative height value of each spatial location. The visual sensing devices, such as multi-view cameras or UAV imaging systems, are used to collect structural spatial distribution data of the target photovoltaic array area, determining the specific location, arrangement, and relative positional relationships between each photovoltaic array. Environmental sensing devices, such as light sensors or panoramic cameras, are used to acquire occlusion relationship data. For example, through light projection or viewpoint analysis methods, the proportion of a location that is occluded is determined, defined as the occlusion coverage ratio, i.e., the proportion of the occluded direction to the total visible direction.

[0029] To ensure accurate fusion and unified processing of data collected by various sensing devices, a unified device access interface is configured for each sensing device, such as based on ROS or a custom data bus, to standardize and encapsulate the data formats output by different sensing devices. After interface configuration, timestamp identification and spatial coordinate reference datum correction are performed to complete the data acquisition initialization. The timestamp identification adds precise time information to each collected data point to describe the acquisition time. Timestamps align the collected data in the time dimension. Timestamp alignment refers to unifying the data collected by different sensing devices onto the same time axis through a synchronized clock source, such as GPS time or Network Time Protocol (NTP), to avoid data misalignment caused by sampling delays. Spatial coordinate reference datum correction transforms the original coordinate systems of each sensing device, such as camera coordinate systems and radar coordinate systems, to a unified global coordinate system, such as the ENU coordinate system established based on the photovoltaic field reference points. This ensures that data from different sources can correspond one-to-one in space. Spatial coordinate reference datum correction obtains the extrinsic parameter matrix, including the rotation matrix R and translation vector T, through calibration board measurement or SLAM algorithm, and then performs coordinate transformation P. global =R·P local +T is implemented, where P local P represents the coordinates of a point in the local coordinate system, i.e., a 3D point (x, y, z) in the coordinate system of a sensing device, such as a camera or LiDAR. The rotation matrix R represents the orientation transformation relationship between the coordinate systems, used to align the orientation of the local coordinate system to the global coordinate system. The translation vector T represents the positional offset between the origin and the coordinate system, used to move the rotated point to the correct position in the global coordinate system. global This indicates the point coordinates transformed to the global coordinate system. After completing the acquisition initialization, the system reads the multi-source sensing data after initialization processing, including terrain elevation data, array structure spatial distribution data, and occlusion relationship data.

[0030] Then, spatiotemporal fusion processing is performed on the multi-source sensing data. Spatiotemporal fusion processing is the process of integrating and analyzing data collected at different times and spatial locations. In the time dimension, data collected by different sensing devices within the same time period are correlated based on timestamps. For example, terrain elevation data collected by a terrain sensing device at a certain moment is matched with array structure spatial distribution data collected by a visual sensing device to ensure that data from the same point in time is used when analyzing the same area. In the spatial dimension, data collected by different sensing devices are aligned and superimposed in space according to a unified spatial coordinate reference benchmark. For example, occlusion relationship data is superimposed on terrain elevation data and array structure spatial distribution data to form a complete dataset.

[0031] Based on the spatiotemporally fused data, a three-dimensional spatial semantic representation model is constructed, incorporating both terrain structural features and photovoltaic array topological features. Specifically, the target photovoltaic array area is divided into a three-dimensional voxel grid at a preset resolution, such as 0.5m × 0.5m × 0.5m. Each voxel serves as the smallest spatial unit, and terrain structural features are calculated within each voxel. For example, the slope (i.e., the magnitude and aspect of the elevation gradient) is calculated using the elevation difference between adjacent voxels, while local undulation indicators are extracted. The fused array structure spatial distribution data is used to identify whether a voxel is occupied by photovoltaic modules, and the effective passage distance between adjacent arrays (i.e., the minimum distance between the boundaries of adjacent modules) is calculated, forming the array spatial distribution features. Furthermore, semantic labels are attached to each voxel, such as ground area, array occupied area, and passage area, and the shading coverage ratio is recorded as an environmental semantic attribute, forming a three-dimensional spatial semantic representation model.

[0032] For example, data acquisition and processing are performed within a photovoltaic array area. A terrain-sensing device (LiDAR) collects terrain elevation data at a frequency of 10Hz, a visual sensing device (camera) collects spatial distribution data of the array structure at a frequency of 30Hz, and an environmental sensing device (light sensor) collects occlusion relationship data. In the time dimension processing, GPS time is used as a unified clock source to timestamp the acquired data. A time tolerance window Δt is set to 100ms, and LiDAR data frames, camera data frames, and light sensor data frames within the same time window are grouped. Due to the low frequency of the LiDAR, a time interpolation method is used to calculate the terrain elevation data at the intermediate time point based on the time and value of two adjacent LiDAR data frames, aligning it to the time axis of the camera data. In the spatial dimension processing, the camera's extrinsic parameter matrix is ​​pre-calibrated, and the array structure spatial distribution data acquired by the camera is transformed into a global coordinate system, spatially aligned with the terrain elevation data acquired by the LiDAR and the occlusion relationship data acquired by the light sensor. Then, using a voxel grid as a spatial index structure, the entire photovoltaic array area is divided into several voxel units. Data falling into the same voxel unit are aggregated. In the fusion calculation, the average value of the terrain elevation data in each voxel unit is taken as the terrain elevation of that voxel. The position of the photovoltaic module is determined by geometric fitting method for the spatial distribution data of the array structure. The proportion of the shading direction to the total visible direction is calculated for the shading relationship data.

[0033] Based on the fused data, a three-dimensional voxel mesh was created with a resolution of 0.5m × 0.5m × 0.5m. Within each voxel, topographic features such as slope and local undulation were calculated. The system identified whether voxels were occupied by photovoltaic modules and calculated the effective passage distance. Semantic labels were added to voxels, and the shading coverage ratio was recorded. Finally, a three-dimensional spatial semantic representation model was constructed. The topographic features were represented by terrain elevation data, displaying information such as terrain undulation and slope changes within the area. The photovoltaic array topology features were represented by array spatial distribution data and shading relationship data, reflecting the connection relationships, arrangement, and shading status between arrays. Through this three-dimensional spatial semantic representation model, the actual situation of the entire target photovoltaic array area can be intuitively understood.

[0034] By constructing a three-dimensional spatial semantic representation model that includes terrain structure features and photovoltaic array topology features, a unified, accurate, and semantically informative spatial foundation is provided for adaptive path planning. This enables path generation to not only rely on geometric distance but also comprehensively consider terrain undulations, array structure, and shading effects, thereby providing reliable input for the construction of dynamic accessibility fields and path optimization, and improving the adaptability and accuracy of paths.

[0035] Obtain the constraint parameters of the target equipment's motion capability, perform coupled calculations of terrain slope features, array spacing features, and occlusion influence features at each location in the three-dimensional spatial semantic representation model, and construct a dynamic accessibility field to characterize the spatial accessibility difficulty distribution.

[0036] Furthermore, a dynamic accessibility field is constructed to characterize the spatial accessibility difficulty distribution, including: resolving the constraint parameters into the maximum passable slope, minimum safe passage distance, and maximum allowable occlusion ratio; in the three-dimensional spatial semantic representation model, extracting the terrain slope value, effective passage distance between adjacent arrays, and occlusion coverage ratio for each spatial location, and comparing them item by item with the maximum passable slope, minimum safe passage distance, and maximum allowable occlusion ratio to obtain the individual constraint deviation of each spatial location under slope constraint, distance constraint, and occlusion constraint; based on the individual constraint deviation of each spatial location, constructing a constraint triggering state determination result, dividing the spatial location into an untriggered constraint region, a single constraint triggered region, and a multi-constraint superimposed triggering region, forming a first intermediate constraint distribution; based on the first intermediate constraint distribution, performing differentiated accessibility modulation processing on the single constraint triggered region and the multi-constraint superimposed triggering region respectively, to establish a dynamic accessibility field.

[0037] Specifically, based on the equipment performance manual provided by the target equipment manufacturer or the equipment motion data obtained from actual testing, the kinematic and safe operation indicators of the target equipment are obtained, including but not limited to the maximum climbing ability of the chassis, the ground clearance capability of wheeled / tracked vehicles, the minimum turning safety passage requirement, and the safe obstacle avoidance distance. These parameters are then converted into three types of constraint parameters: maximum passable slope, minimum safe passage distance, and maximum allowable obstruction ratio. For the maximum climbing ability of the chassis, it is directly converted from the original climbing angle or slope percentage to the maximum passable slope. For the ground clearance capability of wheeled / tracked vehicles and the minimum turning safety passage requirement, the corresponding minimum turning radius or equipment outer contour envelope size is mapped to the required spatial width and further standardized to the minimum safe passage distance. For the safe obstacle avoidance distance, it is converted into the maximum allowable obstruction ratio by combining the positioning or visual failure threshold of the equipment under different obstruction levels.

[0038] Among these parameters, the maximum passable slope refers to the maximum terrain slope that the target equipment can safely navigate. Exceeding this value may result in risks such as tipping over or slipping. The minimum safe passage distance is the minimum distance required for safe operation when the target equipment passes between arrays. If the distance is less than this value, the equipment may not be able to operate normally or may be at risk of collision. The maximum permissible occlusion ratio is the maximum degree of occlusion allowed for the target equipment to operate normally. Exceeding this ratio may prevent the equipment from accurately perceiving the surrounding environment or receiving effective signals. The occlusion ratio refers to the proportion of the invisible area within a certain field of view to the total field of view, reflecting the degree of influence of the environment on perception and navigation. For example, the target equipment is an intelligent inspection robot with the following motion capability constraints: maximum passable slope 20°, minimum safe passage distance 0.8m, and maximum permissible occlusion ratio 30%.

[0039] In the three-dimensional spatial semantic representation model, multi-dimensional feature extraction is performed on each spatial location. For each voxel unit or spatial grid location, the terrain slope value is calculated based on the elevation difference of its neighborhood. The terrain slope value is obtained through the elevation gradient modulus of adjacent voxels and is used to characterize the intensity of local terrain undulation. Simultaneously, the minimum Euclidean distance between this location and the boundary of the nearest photovoltaic array is calculated based on the array spatial distribution feature data in the three-dimensional spatial semantic representation model. This distance is defined as the effective passage distance and is used to characterize the spatial margin for the passage of operating equipment. Based on the environmental semantic attributes recorded in the three-dimensional spatial semantic representation model, the occlusion coverage ratio of this location in the multi-directional field of view is obtained, which is used to characterize the degree of occlusion of observation and navigation signals by the photovoltaic array and terrain.

[0040] The extracted terrain slope value, effective passage distance between adjacent arrays, and occlusion coverage ratio for each spatial location are compared item by item with the maximum passable slope, minimum safe passage distance, and maximum allowable occlusion ratio. For example, the terrain slope value of a certain spatial location is compared with the maximum passable slope. If the terrain slope value is greater than the maximum passable slope, the difference between the two is calculated as the individual constraint deviation under the slope constraint. Similarly, the effective passage distance is compared with the minimum safe passage distance, and the occlusion coverage ratio is compared with the maximum allowable occlusion ratio, and the difference is calculated to obtain the individual constraint deviation under the spacing constraint and occlusion constraint.

[0041] Based on the deviation of individual constraints at each spatial location, a constraint triggering state determination result is constructed. If all individual constraint deviations at a certain spatial location are zero, it means that the terrain slope, array spacing, and occlusion conditions at that location all meet the equipment movement capability requirements, and this location is classified as an untriggered constraint area. If only one individual constraint deviation is not zero, it indicates that only one constraint condition is not met at that location, and this location is classified as a single constraint triggering area. If two or three individual constraint deviations are not zero, it indicates that multiple constraint conditions are not met at that location, and this location is classified as a multi-constraint superimposed triggering area, forming a first intermediate constraint distribution. Then, based on the first intermediate constraint distribution, differentiated accessibility modulation processing is performed on the single constraint triggering area and the multi-constraint superimposed triggering area respectively to establish a dynamic accessibility field, which is used to characterize the spatial accessibility difficulty distribution of the target area under equipment constraint conditions. For example, in a three-dimensional spatial semantic representation model, the terrain slope value at a certain spatial location is extracted to be 25°, the effective passage distance between adjacent arrays is 0.6m, and the occlusion coverage ratio is 40%. Comparing the terrain slope value of 25° at this location with the maximum passable slope of 20°, the deviation of the single constraint under the slope constraint is 25°-20°=5°. Comparing the effective passage distance of 0.6m with the minimum safe passage distance of 0.8m, the deviation of the single constraint under the spacing constraint is 0.8m-0.6m=0.2m. Comparing the occlusion coverage ratio of 40% with the maximum allowable occlusion ratio of 30%, the deviation of the single constraint under the occlusion constraint is 40%-30%=10%. Since the deviations of the three single constraints at this location are all non-zero, it is classified as a multi-constraint superposition triggering area.

[0042] By transforming a three-dimensional spatial semantic model into a dynamic accessibility field that simultaneously integrates equipment capability constraints and environmental spatial characteristics, it is possible to comprehensively consider the equipment's mobility as well as environmental factors such as the terrain, array distribution, and occlusion of the work area. This allows each spatial location to not only describe objective terrain and structural information but also quantify its actual accessibility difficulty for the target work equipment. This enables the planning of reasonable work paths for the target work equipment, preventing it from entering inaccessible or unsafe areas, thereby improving work efficiency and safety, and reducing the probability of equipment damage and work accidents.

[0043] Furthermore, based on the first intermediate constraint distribution, differentiated accessibility modulation processing is performed on single constraint triggering regions and multi-constraint superimposed triggering regions respectively to establish a dynamic accessibility field. This includes: for the single constraint triggering region, performing directional attenuation processing on the accessibility of the spatial location according to the corresponding constraint deviation; for the multi-constraint superimposed triggering region, performing synergistically enhanced accessibility suppression processing according to the superposition degree of each constraint deviation to obtain a second intermediate accessibility distribution; based on the joint distribution state of the first intermediate constraint distribution and the second intermediate accessibility distribution, determining the degree of difference in constraint triggering state between adjacent locations within the spatial neighborhood, and performing hierarchical transfer processing of accessibility according to the degree of difference in constraint triggering state to form a continuously changing accessibility transition distribution; mapping the accessibility transition distribution to a spatial reachability value, and performing directional consistency correction on the value according to the continuity of the change of the accessibility transition distribution within the spatial neighborhood to construct a dynamic accessibility field.

[0044] Specifically, for the untriggered constraint regions in the first intermediate constraint distribution, no constraint suppression or attenuation processing is required. Only their basic passability is retained and assigned a standardized high passability weight. This weight indicates that the region does not exceed the work equipment constraint parameters under slope, spacing, and occlusion constraints, and belongs to the safe passability space under the current conditions. For single constraint-triggered regions, based on their corresponding single deviation, such as only slope deviation or only spacing deviation, a directional attenuation function is used to modulate the basic passability. Directional attenuation refers to a monotonically decreasing passability along the direction of constraint enhancement. For example, an exponential attenuation rule can be used, making the larger the deviation, the more significant the decrease in passability, thus forming a preliminary restricted passability expression, for example, C(x) = C0·e -α·Δ Where C(x) is the accessibility value at spatial location x, C0 is the basic accessibility (taken as 1 when unconstrained), Δ is the deviation from a single constraint, and α is the attenuation coefficient, used to characterize the sensitivity of the equipment to this type of constraint. It is determined through equipment characteristic analysis, experimental calibration, or scenario requirement adjustment. For example, if the working equipment has weak tolerance to a certain type of constraint, α should be larger, causing the accessibility to decrease rapidly with the deviation. Conversely, if the equipment has strong tolerance, α should be smaller, resulting in a more gradual attenuation. For example, wheeled robots use α of 0.5~1.0 for slope, while tracked vehicles use 0.1~0.3.

[0045] For areas triggered by multiple constraints, a collaborative enhancement of traffic capacity suppression is performed based on the degree of superposition of the deviations of each constraint. The degree of superposition of the deviations of each constraint is used to characterize the comprehensive severity of the spatial location being affected by slope constraints, spacing constraints, and occlusion constraints simultaneously. Collaborative enhancement means that when multiple constraints exist simultaneously, their suppression effect on traffic capacity not only comes from the independent contribution of each individual constraint, but also includes the nonlinear amplification effect generated by the mutual coupling between constraints. That is, the overall suppression intensity is greater than the linear superposition of the suppression effects of each individual constraint.

[0046] After normalizing multiple constraint deviations, a weighted fusion is performed to obtain the comprehensive constraint strength index S(x), which is expressed as: S(x) = w1·Δθ + w2·Δd + w3·Δo, where w1, w2, and w3 are the weight coefficients corresponding to slope constraint, spacing constraint, and occlusion constraint, respectively, used to characterize the sensitivity of different constraint types to the passage capacity of the target operating equipment. Each weight coefficient is preset or calibrated according to the structural form, drive method, and environmental adaptability of the target operating equipment. Δθ, Δd, and Δo... These are the slope deviation, spacing deviation, and occlusion deviation after normalization, respectively, used to characterize the degree to which each constraint index exceeds the equipment's passable threshold. Normalization refers to converting constraint deviations with different physical dimensions and numerical ranges into a standardized representation of the same dimensionless or unified numerical range, making the slope deviation, spacing deviation, and occlusion deviation comparable and additive within the same mathematical space. The normalized slope deviation is calculated as: slope deviation / maximum passable slope. Other constraint deviations are converted in a similar way.

[0047] Based on this, the basic suppression term is calculated using the comprehensive constraint strength index S(x), and a constraint coupling enhancement term is introduced to characterize the nonlinear effect generated by the superposition of multiple constraints, thus performing capacity suppression processing: C(x) = C0·e -β·S(x) ·(1+γ·Π(Δ)), where β is the basic suppression coefficient, γ is the synergistic enhancement coefficient, and Π(Δ) is the product of the deviations of each constraint. By enhancing the coupling effect between constraints, the traffic capacity of the multi-factor superposition area decreases more than the linear superposition result of a single constraint, thereby avoiding underestimating the risks of complex environments and obtaining the second intermediate traffic capacity distribution.

[0048] Based on the joint distribution of the first intermediate constraint distribution and the second intermediate accessibility distribution, the degree of difference in constraint triggering states between adjacent locations is determined within a spatial neighborhood. The spatial neighborhood can be set according to actual needs, for example, a region of 6 neighborhoods centered on each spatial location. A set of spatial neighborhood relationships is established based on a three-dimensional voxel mesh, where each spatial unit contains a constraint triggering state identifier Sc(x) and a corresponding accessibility value C(x). Sc(x) includes untriggered constraint regions, single-constraint triggered regions, and multi-constraint superimposed triggered regions. On this basis, for any spatial unit i and its neighboring units j within its neighborhood, the degree of difference in constraint triggering states Ds(i,j) is defined. The difference is calculated by numerically encoding the state identifier. For example, a value of 1 is assigned from an untriggered constraint region to a single-triggered constraint region, a value of 1 is assigned from a single-triggered constraint region to a multi-triggered constraint region, and a value of 2 is assigned from an untriggered constraint region to a multi-triggered constraint region, thereby quantifying the intensity of constraint level changes between adjacent units.

[0049] Simultaneously, the capacity difference Dc(i,j) = |C(i) - C(j)| is used to reflect the magnitude of continuous cost changes, where C(i) is the capacity of spatial unit i, and C(j) is the capacity of neighboring unit j within the neighborhood. These two are weighted and fused to form a comprehensive difference degree D(i,j) = λ1·Ds(i,j) + λ2·Dc(i,j), where λ1 and λ2 are used to balance the impact of abrupt changes in constraint structure and the impact of continuous changes in capacity, respectively. They are configured based on the principle of consistent dimensions after normalization. For example, when the impact of topological changes on path constraints is emphasized, the weight of λ1 is increased, such as 0.6~0.8; when the smoothness of the cost field is emphasized, the weight of λ2 is increased, such as 0.6~0.8, and λ1 + λ2 = 1 is satisfied to ensure the scale stability of the comprehensive difference degree D(i,j).

[0050] A hierarchical transmission rule is constructed based on the comprehensive difference degree. When D(i,j) is small, the consistency of the constraint state of adjacent regions is considered to be high, and a higher propagation coefficient η is used for capacity diffusion update. When D(i,j) is large, the propagation weight is reduced to suppress cross-propagation, thus obtaining the hierarchical transmission update method C'(j)=C(j)+η·(C(i)-C(j))·exp(-D(i,j)). Among them, the distinction of the comprehensive difference degree D(i,j) is based on statistical distribution or empirical threshold. For example, the mean μ and standard deviation σ are calculated in the global adjacency difference degree set. D(i,j)≤μ is regarded as a small difference region, indicating that the adjacent units have a high consistency in constraint state and capacity. μ<D(i,j)≤μ+σ is defined as a medium difference region. D(i,j)>μ+σ is defined as a large difference region, indicating that there is a significant constraint mutation or capacity gradient change. Based on this, the propagation coefficient η is determined using an adaptive decay mechanism that varies with the degree of difference. η is set as η0·exp(-k·D(i,j)), where η0 is the basic propagation coefficient (e.g., 0.5), and k is the decay control coefficient used to adjust the propagation sensitivity. When D(i,j) is small, exp(-k·D(i,j)) approaches 1, allowing the traffic capacity to diffuse sufficiently within the neighborhood, achieving a smooth transition. When D(i,j) is large, the exponential term decays rapidly, reducing the propagation intensity and thus suppressing cross-regional abrupt propagation, ensuring the clarity of the cost field boundary. By iteratively propagating and converging all adjacency relationships in the global space, the originally discrete constraint triggering regions and traffic capacity distributions are spatially transformed into a smooth transition zone, ultimately yielding a traffic capacity transition distribution field C'(x) with continuous gradient variation characteristics. For example, assuming λ1=0.6 and λ2=0.4, if the constraint state between adjacent cells i and j changes from a single constraint triggering region to a multi-constraint superposition triggering region, then Ds(i,j)=1, the throughput C(i)=0.7, C(j)=0.4, then Dc(i,j)=0.3, and the comprehensive difference degree D(i,j)=0.6×1+0.4×0.3=0.72. Assuming η0=0.8 and k=1, then η=0.8·exp(-0.72)≈0.39, and the throughput is updated to C'(j)=0.4+0.39×(0.7-0.4)=0.517, thus enabling the neighborhood to achieve a smooth transition while maintaining the difference constraints.

[0051] The capacity transition distribution is then mapped to a spatial reachability cost, which quantifies the accessibility difficulty of each spatial location for the target equipment. A higher cost indicates greater accessibility difficulty. A standardized mapping function Ccost(x) = 1 / (C'(x) + ε) is used, where ε is a small constant (e.g., 10⁻⁶) to prevent the denominator from being zero. This transforms the capacity transition distribution into a spatial cost field that can be used for path search.

[0052] To ensure the physical consistency of the cost field in space, a directional consistency correction mechanism based on neighborhood gradient continuity is introduced, which calculates the cost gradient vector within the spatial neighborhood. The format is: And analyze its direction of change and its traffic capacity gradient. The reverse consistency relationship, The gradient represents the capacity field. If a sudden change in cost value is found in a local region but the capacity change is gradual, it is considered that there is a non-physical jump or propagation error in that region. A neighborhood averaging correction term is introduced to smooth the cost value, for example, using Ccost'(x) = Ccost(x) + μ·(mean). n (Ccost)-Ccost(x)) is iteratively updated, where mean n (Ccost) represents the average cost value in the neighborhood of the current spatial cell, and μ is a correction coefficient used to control the correction strength. When the gradient changes of traffic capacity and cost value are gradual and there are few non-physical jumps, a smaller μ value, such as 0.1~0.3, can be used to avoid excessive smoothing. When the gradient of traffic capacity changes drastically and the cost value is prone to sudden changes, a larger μ value, such as 0.5~0.8, can be used to enhance the correction strength. This ensures that the cost value increases monotonically in space along the direction of decreasing traffic capacity and decreases monotonically along the direction of increasing traffic capacity, so that the direction of the cost field changes in line with the physical constraints, forming a dynamic trafficability field with continuity, directional consistency and the characteristics that can be used for path optimization calculation.

[0053] By comprehensively considering the impact of single constraints and the superposition of multiple constraints on spatial accessibility, as well as the differences in constraint triggering states between adjacent locations, the continuous change and accurate quantification of accessibility in space are achieved. Accessibility is mapped to spatial reachability cost and corrected, so that the dynamic accessibility field can more intuitively and accurately reflect the accessibility difficulty of equipment in different locations, providing a reliable basis for equipment path planning and operation scheduling, thereby improving the efficiency and safety of equipment operation.

[0054] Extract the spatial arrangement topology features of the photovoltaic arrays and obtain the inspection data of each photovoltaic array.

[0055] Specifically, the spatial arrangement topology features of the photovoltaic array are extracted based on a three-dimensional spatial semantic representation model. These features refer to structured information describing the arrangement of the photovoltaic array in space and their interconnections. They include not only the position and geometric boundaries of individual photovoltaic array units, but also the adjacency relationships, channel connectivity relationships, and overall row and column organization structure between arrays. Through connected component analysis or clustering algorithms, such as density-based DBSCAN or region-growing segmentation methods, continuous array voxels are grouped to identify independent photovoltaic array units. Each array unit is assigned a unique identifier (ID). Based on the geometric center position, main direction, and boundary range of each array unit, the row and column arrangement structure of the array is constructed, that is, the arrangement direction and relative sequence relationship of the array in the global coordinate system are determined.

[0056] By analyzing the spatial adjacency relationships between array units, a topological connection structure is constructed, forming the spatial arrangement topological structure characteristics of the photovoltaic array: Using array units as nodes, the existence of passable channels between adjacent arrays is used as the criterion for edge determination. When the minimum distance between two array units is greater than the minimum passable distance threshold and the intermediate area is not occupied by obstacles, a connection is considered to exist between the two array units. Thus, undirected edges are established between the corresponding nodes, forming the array topological graph structure. The array topological graph structure is represented as a graph model, where nodes represent array units or channel units, and edges represent spatially reachable connections, serving as structural guidance constraints in subsequent path generation.

[0057] Simultaneously, inspection and monitoring data for each photovoltaic array is acquired. This data reflects the importance or priority of different photovoltaic arrays during operation and maintenance. By connecting to the power plant monitoring SCADA system or equipment EMS management system, the operating status parameters of each array are obtained and quantified. For example, the number of fault occurrences, output power deviation rate, or temperature anomaly level are processed using a minimum-maximum normalization algorithm to generate a monitoring intensity index, representing the inspection priority of that array. Furthermore, time factors, such as the time interval between the most recent inspection, can be incorporated to dynamically adjust the monitoring intensity, thus forming a time-varying distribution of inspection and monitoring data.

[0058] By extracting the spatial layout topology features of the photovoltaic array and introducing inspection and maintenance data that reflects the needs of operation and maintenance, subsequent path planning can not only follow the actual layout structure of the photovoltaic array, but also prioritize the coverage of high-risk or high-concern areas, thereby achieving a unity of structural rationality and task orientation, and improving the reliability, adaptability and efficiency of path planning.

[0059] Using the spatial arrangement topology features as structural guiding factors for path generation, global path skeleton optimization is performed based on dynamic drivability field and inspection focus data matching to establish path optimization results.

[0060] Furthermore, global path skeleton optimization based on inspection attention data matching under the dynamic dépendability field includes: discretizing the array channel connection relationship characterized by the spatial arrangement topology into a set of channel nodes and adjacency relationships between nodes, and constructing a structural connectivity graph between channel nodes based on the adjacency relationships; assigning an initial dépendency value to each channel node based on the dynamic dépendability field, and propagating constraints along the adjacency relationships in the structural connectivity graph to construct an initial channel cost mapping relationship; based on the initial channel cost mapping relationship, introducing the array attention intensity corresponding to the inspection attention data to perform a first round of weight correction on the channel node set based on the influence range of adjacency relationship propagation, forming a first intermediate channel cost distribution; based on the first intermediate channel cost distribution, The path expansion direction is defined by the adjacency relationship between nodes, and path connectivity constraints are propagated on the array channel connection relationship to generate an intermediate path skeleton candidate set containing the reachability propagation results. For each candidate skeleton path in the intermediate path skeleton candidate set, the local cost fluctuation consistency of the path is calculated based on the spatial gradient change characteristics of the dynamic drivability field on the path node sequence. The intermediate path skeleton candidate set is then filtered and sorted according to the local cost fluctuation consistency to obtain a second intermediate path skeleton set. In the second intermediate path skeleton set, the structural consistency correction process is performed on the node connection order of the path skeleton based on the adjacency constraint relationship of the array topology, and the corrected cost value of the channel nodes is updated synchronously during the correction process to form a converged path optimization result.

[0061] Specifically, after extracting the spatial arrangement topology features and constructing the dynamic accessibility field, the spatial arrangement topology features are used as structural guidance factors for path generation. Under the constraints of the dynamic accessibility field, global path skeleton optimization is performed by matching inspection focus data. The structural guidance factor refers to using the inherent channel arrangement and adjacency relationship of the photovoltaic array to constrain the path search direction and connection method, thereby avoiding path planning from deviating from the actual accessible channels.

[0062] First, the array channel connection relationships characterized by the spatial arrangement topology are discretized, that is, the channel region in the continuous space is abstracted into a set of discrete channel nodes. Each channel node is defined as a key location point or voxel cluster center on the channel centerline. Based on spatial adjacency relationships, such as whether adjacent channels are connected, a set of adjacency relationships between nodes is established, thereby constructing a structural connectivity graph between channel nodes. The structural connectivity graph is presented in graph theory, where nodes are channel nodes and edges represent passable connections.

[0063] A dynamic accessibility field is introduced to assign an initial access cost to each channel node. The node-level cost is obtained by sampling the cost of the corresponding spatial voxel at each channel node location or by averaging the sampling within the node's neighborhood. Constraint propagation is then performed along the adjacency relationship of the structural connectivity graph. For example, through a graph traversal algorithm, the initial access cost of a node is passed to neighboring nodes according to the adjacency relationship, thus constructing an initial channel cost mapping relationship. This ensures that the cost of each node not only reflects the local access difficulty but also includes the neighborhood propagation influence.

[0064] Based on the initial channel cost mapping, inspection attention data is introduced to map the array attention intensity of each array unit to the corresponding channel node. Influence range propagation is then performed based on adjacency relationships, using array attention intensity as a weighting factor to correct the initial channel cost of adjacent nodes, forming a first intermediate channel cost distribution. This reduces the cost of nodes in high-attention array regions, making them more likely to be selected into the path. Based on this first intermediate channel cost distribution, and utilizing the adjacency relationships between nodes to limit the path expansion direction (allowing path expansion only along channel connection directions), path connectivity constraint propagation is performed using a graph-based path expansion algorithm, such as Dijkstra's algorithm. The search is performed in the structural connectivity graph to obtain a set of candidate path sequences that satisfy connectivity constraints, which serve as the intermediate path skeleton candidate set. These candidate paths are potential inspection paths that initially meet the connectivity requirements.

[0065] For each candidate skeleton path in the intermediate path skeleton candidate set, the consistency of local cost fluctuations is calculated based on the spatial gradient change characteristics of the dynamic accessibility field on the path node sequence. The spatial gradient change characteristics reflect the changes of the dynamic accessibility field at different locations. The spatial gradient change characteristics are measured by calculating the difference in access cost values ​​between adjacent nodes on the path. Local cost fluctuation consistency is used to evaluate the smoothness of cost changes on the path. Higher consistency indicates a more stable path and greater stability when inspection equipment passes through. The intermediate path skeleton candidate set is then filtered and sorted based on the local cost fluctuation consistency, removing paths with large fluctuations and instability to obtain the second intermediate path skeleton set.

[0066] In the second intermediate path skeleton set, combined with the adjacency constraint relationship of the array topology, the node connection order of the path skeleton is subjected to structural consistency correction processing. By checking whether the path node sequence conforms to the actual row and column arrangement logic of the array, such as avoiding cross-row jumps or reverse connections, path segments that do not conform to the structural constraints are reordered or replaced. During the correction process, the cost value of the channel nodes is locally updated and rebalanced according to the new connection relationship, so that the path achieves consistency between structural rationality and cost optimization, and finally forms a converged path optimization result. The path optimization result is an inspection path that not only conforms to the actual physical connection structure of the photovoltaic array, but also avoids obstacles and prioritizes passing through high-interest areas.

[0067] By comprehensively considering the spatial arrangement topology characteristics, dynamic accessibility field, and inspection focus data, the path generation process simultaneously satisfies structural feasibility, access safety, and task orientation, thereby avoiding invalid path searches, improving path planning efficiency, and ensuring the feasibility and inspection coverage effectiveness of the generated path in actual complex photovoltaic array scenarios, thus improving inspection efficiency and accuracy and reducing inspection costs and risks.

[0068] Furthermore, calculating the consistency of local cost fluctuations along the path includes: determining directed adjacency pairs between adjacent nodes in the candidate skeleton path based on the adjacency relationships between nodes, extracting the difference in travel costs for each directed adjacency pair in the dynamic drivability field, and constructing a cost gradient sequence corresponding to the path node sequence; performing directional consistency analysis on the cost gradient sequence to identify the same-direction segments and opposite-direction jump segments in the cost change direction between consecutive adjacency pairs, and constructing a description of the local fluctuation structure of the path based on the length of the same-direction segments and the frequency of opposite-direction jumps; performing segmented smoothing processing on the cost gradient sequence based on the description of the local fluctuation structure of the path to weaken the influence of abrupt gradients caused by local abnormal terrain or occlusion, and obtaining a corrected cost gradient distribution sequence; and calculating the consistency of local cost fluctuations characterizing path stability based on the continuity of gradient changes in each segment of the corrected cost gradient distribution sequence.

[0069] Specifically, after obtaining the intermediate path skeleton candidate set, the stability of each candidate path in the intermediate path skeleton candidate set is quantitatively evaluated by calculating the consistency of local cost fluctuations. First, the directed adjacency pairs between adjacent nodes in the candidate skeleton path are determined based on the adjacency relationship between nodes. The adjacency relationship between nodes comes from the structural connectivity graph, which clarifies the connection order and direction between each node on the path. For example, in a candidate skeleton path composed of four nodes A, B, C, and D connected in sequence, the directed adjacency pairs are (A,B), (B,C), and (C,D). The directionality is used to reflect the travel direction of the inspection equipment on the path.

[0070] The spatial reachability cost of each node is extracted from the dynamic dépendency field, and the cost difference between each directed adjacency pair is calculated. Specifically, for each directed adjacency pair, the dépendency cost of the previous node is subtracted from the dépendency cost of the subsequent node to obtain the dépendency cost difference, forming the cost gradient sequence corresponding to the path node sequence. The cost gradient describes the trend of difficulty change along the travel direction; a positive value indicates increasing dépendency, and a negative value indicates decreasing dépendency. Then, directional consistency analysis is performed on the cost gradient sequence. Directional consistency analysis involves determining the sign of the gradient sequence, dividing consecutive directed adjacency pairs with the same direction of dépendency cost change into co-directional segments, while the portion between adjacent directed adjacency pairs with opposite directions of dépendency cost change is defined as a reverse jump segment. For example, consider the cost gradient sequence [3,2,-1,-2,4]. The first two 3s and 2s are in the same direction and are both positive, indicating an increase in travel cost, forming a segment in the same direction. -1 and -2 are also in the same direction and are both negative, indicating a decrease in travel cost, forming another segment in the same direction. The transition from 2 to -1 is a reverse transition, forming a reverse transition segment. By statistically analyzing the length of the segments in the same direction and the frequency of reverse transitions, a description of the local fluctuation structure of the path is constructed. The length of the segments in the same direction refers to the number of directed adjacency pairs contained in the segments, and the frequency of reverse transitions refers to the number of times reverse transitions occur.

[0071] Based on the description of local fluctuations in the path, a piecewise smoothing process is performed on the cost gradient sequence to weaken atypical gradient fluctuations caused by local abnormal terrain, such as abrupt slope changes or occlusion errors. The piecewise smoothing process employs methods such as moving averages to smooth the cost gradient values ​​within each segment, reducing the impact of abrupt gradient changes. For example, for the same-direction segment [3,2], after moving average processing, it may become [2.5,2.5], thus obtaining a corrected cost gradient distribution sequence, making the change in passage cost smoother and more reasonable. Based on the continuity of gradient changes in each segment of the corrected cost gradient distribution sequence, the consistency of local cost fluctuations, which characterizes path stability, is calculated. The continuity of gradient changes is measured by calculating the standard deviation of the gradient values ​​within each segment. The smaller the standard deviation, the more continuous the gradient change, the more stable the path, and the more suitable it is for actual equipment execution.

[0072] By calculating the consistency of local cost fluctuations, paths with stable traffic costs and minimal impact from local anomalies can be selected. This avoids instability caused by excessive fluctuations in path traffic costs during inspection, thereby improving the safety, accuracy, and overall efficiency of inspection paths and reducing operational risks.

[0073] Furthermore, after calculating the consistency of local cost fluctuations that characterize path stability, the method further includes: configuring a path stability identifier based on the consistency of local cost fluctuations; establishing a local path replacement library using the path stability identifier; and embedding the local path replacement library into the path optimization result.

[0074] Specifically, after completing the calculation of path local cost fluctuation consistency, stability marking is performed on the path according to the calculation result of path local cost fluctuation consistency. Said path stability marking refers to the result of classifying and marking the stability degree of passage cost variation in different sections of the path. Specifically, based on the calculated local cost fluctuation consistency index, the path node sequence is segmented and marked in combination with preset threshold intervals. For example, two thresholds T1 and T2 are set, where T1<T2. When the local cost fluctuation consistency is less than T1, the path section is marked as highly stable and assigned a stability marking value of 3; when the local cost fluctuation consistency is between T1 and T2, it is marked as moderately stable and assigned a stability marking value of 2; when the local cost fluctuation consistency is greater than T2, it is marked as unstable and assigned a stability marking value of 1. The preset thresholds T1 and T2 are obtained through statistical analysis of historical data and the requirements of inspection equipment for path stability in actual application scenarios, and the data sources include the passage cost fluctuation conditions of a large number of actual inspection paths and the operation performance of inspection equipment on paths with different fluctuation degrees. For example, statistical analysis is performed on the collected passage cost fluctuation values of 500 historical path sections, and the distribution characteristics of the fluctuation values are analyzed based on a frequency distribution histogram, wherein the horizontal axis represents the local cost fluctuation consistency index and the vertical axis represents the occurrence frequency. The corresponding relationship between the fluctuation values and path stability is determined by observing the actual operation performance of the inspection robot in different fluctuation intervals, including indexes such as traveling stability, number of path deviations and speed fluctuation degree. When the fluctuation value is less than 0.4, the overall operation performance of the inspection robot is stable, the path deviation rate is low and the control error is small, so the threshold T1 is set to 0.4, and the path section with local cost fluctuation consistency less than T1 is defined as a highly stable path section; when the fluctuation value is between 0.4 and 0.78, although the inspection robot has slight speed adjustment and local path correction, it can still stably perform tasks as a whole, so the threshold T2 is set to 0.78, and this interval is defined as a moderately stable path section; when the local cost fluctuation consistency is greater than 0.78, the operation performance of the inspection robot is poor, and the path is identified as an unstable path section.

[0075] After completing path stability marking, a local path replacement library is constructed for path sub-intervals marked with stability marking value 1 or stability marking value 2. Said local path replacement library refers to a set that stores multiple alternative path fragments in units of path sub-segments. Its main data sources include: candidate path sub-segments that are not selected as the optimal path but satisfy connectivity constraints during the current path search process, actually feasible path fragments recorded in historical inspection processes, and alternative paths obtained through re-searching based on local subgraphs. For example, constrained searching is performed within the problematic section with local start and end nodes as boundaries.

[0076] During the construction of the local path replacement library, the start and end nodes corresponding to unstable or moderately stable segments are used as index keys to retrieve or generate several alternative paths in the structural connectivity graph. For each alternative path, evaluation indicators such as local cost fluctuation consistency, average cost value, and path length are recalculated, forming a candidate replacement set with multi-dimensional attribute labels. This local path replacement library is embedded into the path optimization result. After obtaining the final path optimization result, all path segments in the result are traversed to check if there is a path segment that matches the original unstable or moderately stable path segment in the local path replacement library. If so, the unstable or moderately stable path segment is replaced with the corresponding alternative path segment from the local path replacement library, thereby obtaining the optimized path optimization result. This avoids global path replanning, ensures that the final planned path has high stability and accuracy, and reduces potential problems encountered by inspection equipment during inspection due to path instability.

[0077] By configuring path stability indicators, establishing a local path replacement library, and embedding it into the path optimization results, the path planning scheme was further optimized. This enabled the effective identification and handling of unstable areas in the path. By replacing unstable path segments, the stability of the entire generated path was improved, reducing the risk of accidents caused by path problems during the inspection process, while ensuring the efficient execution of the inspection task.

[0078] Furthermore, after establishing the path optimization results, the process also includes: acquiring updated environmental perception data in real time during the execution based on the path optimization results, and performing local reconstruction of the dynamic navigability field; and using the local reconstruction results to perform real-time perception correction of the path optimization results.

[0079] Specifically, during the execution process based on path optimization results, updated environmental perception data is acquired in real time through multi-source sensing devices deployed on the target operating equipment. The updated environmental perception data includes local terrain point cloud data acquired by lidar, array structure and obstacle image data collected by visual sensors, and near-range occlusion and dynamic obstacle information acquired by environmental perception devices. The above environmental perception data undergoes timestamp alignment and spatial coordinate transformation processing after acquisition.

[0080] The newly acquired environmental perception data is compared with the existing 3D spatial semantic representation model for difference detection. Voxel-level comparison or local subgraph matching is used to identify areas of environmental change, such as the detection of new obstructions, the occupation of existing passages, or local terrain changes. The detected areas of change are defined as local reconstruction areas, and key parameters such as terrain slope characteristics, array spacing characteristics, and obstruction coverage ratio are recalculated within these areas. Then, based on the updated constraint deviations, the accessibility calculation process is re-executed, including constraint trigger determination, accessibility modulation, and neighborhood propagation, thereby completing the local reconstruction of the dynamic accessibility field.

[0081] Then, utilizing the locally reconstructed dynamic traversability field, real-time perceptual correction is performed on the original path optimization results. Specifically, the current execution path is mapped to the updated dynamic traversability field, the spatial reachability value of the node sequence on the execution path is resampled, and path segments with significantly increased costs or decreased traversability are detected, such as exceeding preset thresholds or deteriorating stability indicators. For affected path segments, a fast path re-search is performed using the aforementioned established local path replacement library or in the local topology subgraph, such as local... Alternatively, the Dijkstra algorithm can be used to generate alternative path fragments and embed them into the original path, thereby achieving local path correction. If the range of change is large, it can trigger a larger-scale path replanning, enabling the optimized path to continuously adapt to environmental changes during execution and maintain drivability and safety.

[0082] By introducing a real-time environmental perception and dynamic update mechanism based on the path planning results, the path is transformed from a static planning result into a perceptible and adaptive dynamic execution plan. This effectively addresses uncertainties in the photovoltaic array environment, such as changes in shading, equipment movement, or terrain disturbances, thereby improving the safety and reliability of path execution and ensuring that the inspection equipment can safely and smoothly complete its inspection tasks.

[0083] Example 2, based on the same inventive concept as the adaptive path generation method for photovoltaic array perception of complex terrain in the foregoing examples, such as... Figure 2 As shown, this application provides an adaptive path generation system for photovoltaic array perception in complex terrain, wherein the adaptive path generation system for photovoltaic array perception in complex terrain includes: The sensing data processing module 11 is used to read multi-source sensing data after the multi-source sensing devices in the target photovoltaic array area are connected, and to perform spatiotemporal fusion processing of the multi-source sensing data to construct a three-dimensional spatial semantic representation model that includes terrain structure features and photovoltaic array topology features. The feature coupling calculation module 12 is used to obtain the constraint parameters of the target operating equipment's motion capability, and to perform coupling calculations of terrain slope features, array spacing features, and shading influence features at each location in the three-dimensional spatial semantic representation model to construct a dynamic traversability field that represents the spatial accessibility difficulty distribution. The feature data acquisition module 13 is used to extract the spatial arrangement topology features of the photovoltaic array and obtain the inspection attention data of each photovoltaic array. The optimization result establishment module 14 is used to use the spatial arrangement topology features as the structural guidance factor for path generation, and to perform global path skeleton optimization under the matching of inspection attention data based on the dynamic traversability field to establish the path optimization result.

[0084] Furthermore, the optimization result establishment module 14 is also used to: discretize the array channel connection relationship characterized by the spatial arrangement topology into a set of channel nodes and adjacency relationships between nodes, and construct a structural connectivity graph between channel nodes based on the adjacency relationships; assign an initial accessibility cost to each channel node based on the dynamic accessibility field, and propagate constraints along the adjacency relationships in the structural connectivity graph to construct an initial channel cost mapping relationship; based on the initial channel cost mapping relationship, introduce the array attention intensity corresponding to the inspection attention data to perform a first round of weight correction on the channel node set based on the influence range of adjacency relationship propagation, forming a first intermediate channel cost distribution; based on the first intermediate channel cost distribution, and utilizing the adjacency relationships between nodes... The system defines the path expansion direction and propagates path connectivity constraints on the array channel connections to generate an intermediate path skeleton candidate set containing reachability propagation results. For each candidate skeleton path in the intermediate path skeleton candidate set, the local cost fluctuation consistency of the path is calculated based on the spatial gradient change characteristics of the dynamic drivability field on the path node sequence. The intermediate path skeleton candidate set is then filtered and sorted according to the local cost fluctuation consistency to obtain a second intermediate path skeleton set. In the second intermediate path skeleton set, based on the adjacency constraint relationship of the array topology, the node connection order of the path skeleton is subjected to structural consistency correction processing, and the corrected cost value of the channel nodes is updated synchronously during the correction process to form a converged path optimization result.

[0085] Furthermore, the optimization result establishment module 14 is also used to: determine the directed adjacency pairs between adjacent nodes in the candidate skeleton path based on the adjacency relationship between nodes, and extract the passage cost difference of each directed adjacency pair in the dynamic drivability field to construct the cost gradient sequence corresponding to the path node sequence; perform directional consistency analysis on the cost gradient sequence to identify the same-direction segments and opposite-direction jump segments of the cost change direction between consecutive adjacency pairs, and construct a path local fluctuation structure description based on the length of the same-direction segment and the frequency of the opposite-direction jump; perform segmented smoothing processing on the cost gradient sequence based on the path local fluctuation structure description to weaken the influence of abrupt gradients caused by local abnormal terrain or occlusion, and obtain a corrected cost gradient distribution sequence; calculate the local cost fluctuation consistency characterizing the path stability according to the continuity of gradient changes in each segment of the corrected cost gradient distribution sequence.

[0086] Furthermore, the optimization result establishment module 14 is also used to: configure a path stability identifier based on the consistency of local cost fluctuations; establish a local path replacement library using the path stability identifier; and embed the local path replacement library into the path optimization result.

[0087] Furthermore, the feature coupling calculation module 12 is also used to: parse the constraint parameters into the maximum passable slope, the minimum safe passage distance, and the maximum allowable occlusion ratio; in the three-dimensional spatial semantic representation model, extract the terrain slope value, the effective passage distance between adjacent arrays, and the occlusion coverage ratio for each spatial location, and compare them item by item with the maximum passable slope, the minimum safe passage distance, and the maximum allowable occlusion ratio to obtain the individual constraint deviation of each spatial location under the slope constraint, the distance constraint, and the occlusion constraint; based on the individual constraint deviation of each spatial location, construct the constraint triggering state judgment result, divide the spatial location into an untriggered constraint area, a single constraint triggering area, and a multi-constraint superimposed triggering area, forming a first intermediate constraint distribution; based on the first intermediate constraint distribution, perform differentiated passage capability modulation processing on the single constraint triggering area and the multi-constraint superimposed triggering area respectively to establish a dynamic passability field.

[0088] Furthermore, the feature coupling calculation module 12 is also used to: perform directional attenuation processing on the drivability of the spatial location according to the corresponding constraint deviation amount for the single constraint triggering region; perform synergistic enhanced drivability suppression processing on the multi-constraint superimposed triggering region according to the superposition degree of each constraint deviation amount to obtain a second intermediate drivability distribution; determine the degree of difference in constraint triggering state between adjacent locations within the spatial neighborhood based on the joint distribution state of the first intermediate constraint distribution and the second intermediate drivability distribution, and perform hierarchical transmission processing of drivability according to the degree of difference in constraint triggering state to form a continuously changing drivability transition distribution; map the drivability transition distribution to a spatial reachability cost, and perform directional consistency correction on the cost based on the continuity of the change of the drivability transition distribution within the spatial neighborhood to construct a dynamic drivability field.

[0089] Furthermore, the multi-source sensing data includes terrain elevation data, array structure spatial distribution data, and occlusion relationship data.

[0090] Furthermore, the multi-source sensing devices include terrain sensing devices, visual sensing devices, and environmental sensing devices. After configuring a unified device access interface for each sensing device, timestamp identification and spatial coordinate reference benchmark correction are performed to complete the data acquisition initialization.

[0091] Furthermore, the optimization result establishment module 14 is also used to: acquire updated environmental perception data in real time during the execution of path optimization results, and perform local reconstruction of the dynamic drivability field; and use the local reconstruction results to perform real-time perception correction of the path optimization results.

[0092] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The adaptive path generation method and specific examples for complex terrain photovoltaic array perception in the foregoing Embodiment 1 are also applicable to the adaptive path generation system for complex terrain photovoltaic array perception in this embodiment. Through the foregoing detailed description of the adaptive path generation method for complex terrain photovoltaic array perception, those skilled in the art can clearly understand the adaptive path generation system for complex terrain photovoltaic array perception in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0093] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0094] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. An adaptive path generation method for photovoltaic array sensing in complex terrain, characterized in that, The method includes: After connecting to the multi-source sensing devices in the target photovoltaic array area, the multi-source sensing data is read and spatiotemporal fusion processing of the multi-source sensing data is performed to construct a three-dimensional spatial semantic representation model that includes terrain structure features and photovoltaic array topology features. Obtain the constraint parameters of the target operating equipment's motion capability, perform coupled calculations of terrain slope features, array spacing features, and occlusion influence features at each location in the three-dimensional spatial semantic representation model, and construct a dynamic drivability field to characterize the spatial accessibility difficulty distribution; Extract the spatial arrangement topology features of the photovoltaic array and obtain the inspection focus data for each photovoltaic array; Using the spatial arrangement topology features as structural guiding factors for path generation, global path skeleton optimization is performed based on dynamic drivability field and inspection focus data matching to establish path optimization results.

2. The adaptive path generation method for photovoltaic array sensing in complex terrain as described in claim 1, characterized in that, Global path skeleton optimization based on dynamic accessibility field inspection focus data matching includes: The array channel connection relationship characterized by the spatial arrangement topology is discretized into a set of channel nodes and the adjacency relationship between nodes, and a structural connectivity graph between channel nodes is constructed based on the adjacency relationship. Based on the dynamic accessibility field, each channel node is assigned an initial access cost value, and constraints are propagated along the adjacency relationships in the structural connectivity graph to construct the initial channel cost mapping relationship. Based on the initial channel cost mapping relationship, the array attention intensity corresponding to the inspection attention data is introduced to perform a first round of weight correction on the channel node set based on the propagation influence range of the adjacency relationship, forming a first intermediate channel cost distribution; Based on the cost distribution of the first intermediate channel and using the adjacency relationship between nodes to limit the path expansion direction, the path connectivity constraint propagation is performed on the array channel connection relationship to generate a candidate set of intermediate path skeletons containing the reachability propagation results. For each candidate skeleton path in the intermediate path skeleton candidate set, the consistency of local cost fluctuations of the path is calculated based on the spatial gradient change characteristics of the dynamic drivability field on the path node sequence. The intermediate path skeleton candidate set is then filtered and sorted according to the consistency of local cost fluctuations to obtain the second intermediate path skeleton set. In the second intermediate path skeleton set, based on the adjacency constraint relationship of the array topology, the node connection order of the path skeleton is subjected to structural consistency correction processing, and the correction cost of the channel node is updated synchronously during the correction process to form the converged path optimization result.

3. The adaptive path generation method for photovoltaic array sensing in complex terrain as described in claim 2, characterized in that, Consistency of local cost fluctuations along the computation path, including: Based on the adjacency relationship between the nodes, the directed adjacency pairs between adjacent nodes in the candidate skeleton path are determined, and the passage cost difference of each directed adjacency pair in the dynamic drivability field is extracted to construct the cost gradient sequence corresponding to the path node sequence. Perform directional consistency analysis on the cost gradient sequence to identify the same-direction segments and opposite-direction jump segments of cost change direction between consecutive adjacent pairs, and construct a path local fluctuation structure description based on the length of the same-direction segment and the frequency of the opposite-direction jump. Based on the description of the local fluctuation structure of the path, the cost gradient sequence is processed by segmented smoothing to weaken the influence of abrupt gradients caused by local abnormal terrain or occlusion, and a corrected cost gradient distribution sequence is obtained. Based on the continuity of gradient changes in each segment of the corrected cost gradient distribution sequence, the consistency of local cost fluctuations, which characterizes path stability, is calculated.

4. The adaptive path generation method for photovoltaic array sensing in complex terrain as described in claim 3, characterized in that, After calculating the consistency of local cost fluctuations that characterize path stability, the following steps are also included: Configure a path stability identifier based on the local cost fluctuation consistency; A local path replacement library is established using path stability identifiers, and the local path replacement library is embedded into the path optimization results.

5. The adaptive path generation method for photovoltaic array sensing in complex terrain as described in claim 1, characterized in that, Construct a dynamic drivability field to characterize the spatial accessibility difficulty distribution, including: The constraint parameters are resolved into the maximum passable slope, the minimum safe passage distance, and the maximum allowable occlusion ratio; In the three-dimensional spatial semantic representation model, the terrain slope value, the effective passage distance between adjacent arrays and the occlusion coverage ratio are extracted for each spatial location, and compared with the maximum passable slope, the minimum safe passage distance and the maximum allowable occlusion ratio respectively to obtain the individual constraint deviation of each spatial location under the slope constraint, spacing constraint and occlusion constraint. Based on the deviation of individual constraints at each spatial location, a constraint triggering state determination result is constructed, and the spatial location is divided into non-triggered constraint area, single constraint triggered area and multi-constraint superimposed triggering area, forming the first intermediate constraint distribution. Based on the first intermediate state constraint distribution, differentiated accessibility modulation processing is performed on single constraint triggering regions and multi-constraint superposition triggering regions respectively to establish a dynamic accessibility field.

6. The adaptive path generation method for photovoltaic array sensing in complex terrain as described in claim 5, characterized in that, Based on the first intermediate state constraint distribution, differentiated accessibility modulation processing is performed in single-constraint triggering regions and multi-constraint superimposed triggering regions respectively to establish a dynamic accessibility field, including: For the single constraint triggering area, directional attenuation processing is performed on the accessibility of the spatial location based on the corresponding constraint deviation. For the multi-constraint superposition triggering region, a collaborative enhanced capacity suppression process is performed based on the superposition degree of each constraint deviation to obtain a second intermediate capacity distribution; Based on the joint distribution state of the first intermediate constraint distribution and the second intermediate capacity distribution, the degree of difference in constraint triggering state between adjacent positions is determined within the spatial neighborhood, and the hierarchical transfer processing of capacity is performed according to the degree of difference in constraint triggering state to form a continuously changing capacity transition distribution. The transitional distribution of accessibility is mapped to a spatial accessibility value, and the value is corrected for directional consistency based on the continuity of the transitional distribution of accessibility in the spatial neighborhood, thereby constructing a dynamic accessibility field.

7. The adaptive path generation method for photovoltaic array sensing in complex terrain as described in claim 1, characterized in that, Multi-source sensing data includes terrain elevation data, array structure spatial distribution data, and occlusion relationship data.

8. The adaptive path generation method for photovoltaic array sensing in complex terrain as described in claim 1, characterized in that, The multi-source sensing devices include terrain sensing devices, visual sensing devices, and environmental sensing devices. After configuring a unified device access interface for each sensing device, timestamp identification and spatial coordinate reference benchmark correction are performed to complete the data acquisition initialization.

9. The adaptive path generation method for photovoltaic array sensing in complex terrain as described in claim 1, characterized in that, After establishing the path optimization results, it also includes: During the execution process based on path optimization results, updated environmental perception data is acquired in real time, and dynamic accessibility field is performed for local reconstruction. Real-time perception correction of path optimization results is performed using the results of local reconstruction.

10. An adaptive path generation system for photovoltaic array sensing in complex terrain, characterized in that, The step of implementing the adaptive path generation method for sensing complex terrain photovoltaic arrays according to any one of claims 1 to 9, wherein the adaptive path generation system for sensing complex terrain photovoltaic arrays comprises: The sensing data processing module is used to read multi-source sensing data after the multi-source sensing devices in the target photovoltaic array area are connected, and to perform spatiotemporal fusion processing of the multi-source sensing data to construct a three-dimensional spatial semantic representation model that includes terrain structure features and photovoltaic array topology features. The feature coupling calculation module is used to obtain the constraint parameters of the target operation equipment's motion capability, perform coupled calculations of terrain slope features, array spacing features, and occlusion influence features at each location in the three-dimensional spatial semantic representation model, and construct a dynamic drivability field to characterize the spatial accessibility difficulty distribution. The feature data acquisition module is used to extract the spatial arrangement topology features of the photovoltaic array and obtain the inspection attention data of each photovoltaic array. The optimization result establishment module is used to take the spatial arrangement topology features as the structural guidance factor for path generation, perform global path skeleton optimization based on the dynamic accessibility field and the matching of inspection attention data, and establish path optimization results.

Images