Intelligent management method and system of rural light storage and charging station combined with point cloud data

Abstract

The present application relates to the technical field of new energy facility intelligent management, and particularly relates to a rural light storage and charging station intelligent management method and system combined with point cloud data. The method comprises the following steps: obtaining point cloud data of a temporarily deployed site; performing point cloud site modeling according to the point cloud data to obtain a point cloud site model; performing point cloud site feature extraction according to the point cloud site model to obtain point cloud site feature data; performing deployment optimization area calculation according to the point cloud site feature data to obtain deployment optimization area data; and performing intelligent placement simulation according to the deployment optimization area data to obtain intelligent placement data. The present application combines point cloud space perception and AI intelligent optimization to provide an efficient deployment management method, thereby improving the construction and operation quality of rural light storage and charging stations.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology for new energy facilities, and in particular to an intelligent management method and system for rural photovoltaic storage and charging stations that incorporates point cloud data. Background Technology

[0002] With the accelerated construction of new energy infrastructure in rural areas, integrated photovoltaic-storage-charging stations (i.e., stations integrating photovoltaic power generation, energy storage systems, and electric vehicle charging facilities) have become an important means to promote the widespread application of new energy. However, rural sites typically present problems such as complex terrain, a high proportion of soft ground, numerous dynamic obstacles, and low site space utilization. Traditional site deployment models that rely on manual experience are insufficient to meet the high requirements for deployment efficiency, equipment stability, and energy efficiency. In existing technologies, most photovoltaic-storage-charging station deployments still adopt a standardized template layout, lacking high-precision spatial modeling and intelligent deployment optimization processes for unstructured rural sites. Especially in temporary deployment scenarios (such as temporary expansion of charging stations during the busy farming season), they lack dynamic site perception and real-time deployment capabilities. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes an intelligent management method and system for rural photovoltaic storage and charging stations that incorporates point cloud data, thereby resolving at least one of the aforementioned technical issues.

[0004] This application provides a smart management method for rural photovoltaic storage and charging stations that combines point cloud data. The method includes:

[0005] S1. Obtain point cloud data of the temporary deployment site;

[0006] S2. Perform point cloud site modeling based on point cloud data to obtain point cloud site model; extract point cloud site features based on point cloud site model to obtain point cloud site feature data.

[0007] S3. Calculate the optimal deployment area based on the point cloud site feature data to obtain the optimal deployment area data;

[0008] S4. Perform intelligent deployment simulation based on the deployment optimization area data to obtain intelligent deployment data.

[0009] This invention incorporates point cloud data into the temporary deployment management process of rural photovoltaic-storage-charging stations, enabling high-precision modeling and spatial structure analysis of the deployment site. Addressing issues such as soft boundaries, irregular micro-topography, and dynamic obstacles in rural sites, it utilizes multi-dimensional feature extraction techniques to comprehensively enhance the site's status perception capabilities. During the calculation of the optimal deployment area, feature-driven risk zone elimination and spatial accessibility analysis achieve global optimization of the deployment safety, efficiency, and coordination of photovoltaic-storage-charging equipment. In the intelligent deployment simulation phase, combining dynamic simulation and AI feature field guidance effectively avoids soft depressions and high-risk shading areas, improving the operational stability of the photovoltaic-storage system and the overall efficiency of power generation and charging. This approach is suitable for the highly dynamic and heterogeneous environments of rural sites.

[0010] Optionally, S1 includes:

[0011] The control terminal laser device performs low-resample pre-scanning to obtain initial point cloud data;

[0012] The initial point cloud data is processed to perform a preliminary estimation of the land surface carrying capacity, resulting in carrying capacity data.

[0013] The carrying capacity data is processed to extract stability data for key areas, thus obtaining stability data for key areas.

[0014] Based on the stability data of key areas, the control terminal laser device performs high-precision back-scan of the initial point cloud data in key areas to obtain point cloud data.

[0015] This invention estimates the surface carrying capacity in advance during the low-resampled pre-scanning stage and extracts stability information for key areas based on the carrying capacity results. This effectively avoids resource waste and unnecessary redundant data collection in traditional full-field high-precision scanning. By performing high-precision backscanning of key areas, fine-grained data enhancement is achieved for areas with high deployment potential, improving the local accuracy and spatial resolution of the overall point cloud data.

[0016] Optionally, the point cloud site modeling includes:

[0017] Based on point cloud data and a pre-set scene library model, scene type matching is performed to generate pseudo-historical site baseline data.

[0018] Pseudo-physics boundary fitting is performed on the point cloud data to obtain optimized point cloud data;

[0019] Based on the point cloud optimization data and pseudo-historical site benchmark data, the land surface morphology is fitted to generate a land surface model.

[0020] Entropy filtering is applied to the optimized point cloud data to obtain steady-state occlusion data;

[0021] Slope risk calculations were performed on the surface model to obtain slope risk data;

[0022] Spatial accessibility data is obtained by performing spatial accessibility processing based on steady-state shading data and slope risk data;

[0023] A point cloud site model is obtained by fusing spatial accessibility data and surface model.

[0024] This invention introduces a scene library model for scene type matching, enabling the rapid generation of pseudo-historical benchmarks in rural temporary deployment sites lacking historical data, thus improving fitting accuracy. Pseudo-physics boundary fitting technology effectively corrects the soft variability of point cloud data boundaries, enhancing boundary stability. A continuous surface model is generated through terrain morphology fitting, and entropy filtering eliminates the influence of short-term dynamic occlusion, ensuring the steady-state reliability of the site model. Slope risk calculation and spatial accessibility processing are performed collaboratively, refining deployment risk areas and equipment access paths. The fusion of spatial accessibility data and the surface model results in a point cloud site model with high spatial consistency, high dynamic adaptability, and high deployment availability.

[0025] Optionally, the entropy filtering process includes:

[0026] Dual-domain spatial entropy calculation is performed on the point cloud optimization data to obtain dual-domain spatial entropy data;

[0027] Based on the dual-domain spatial entropy data, preliminary marking of candidate occlusion points is performed on the point cloud optimization data to obtain candidate occlusion point marking data;

[0028] Entropy gradient field data is generated by processing the candidate occlusion point marker data.

[0029] Based on the entropy gradient field data, Markov correction modeling is performed to obtain the entropy gradient field model.

[0030] Steady-state occlusion segmentation is performed based on the entropy gradient field model to obtain steady-state occlusion data.

[0031] This invention introduces dual-domain spatial entropy calculation to finely characterize the local terrain complexity and occlusion uncertainty in rural sites with diverse spatial structures and soft obstacles or dynamic disturbances, thereby improving the accuracy of occlusion point identification. Candidate occlusion point labeling combined with entropy gradient field generation effectively distinguishes between steady-state obstacles and short-term dynamic occlusion, resolving misjudgments caused by changes in agricultural machinery, personnel, or the environment. Markov correction modeling optimizes the spatial consistency of the occlusion area using neighborhood normal continuity, further improving the integrity and coherence of the occlusion segmentation boundary. Through steady-state occlusion segmentation, an occlusion model with high robustness and dynamic adaptability is generated.

[0032] Optionally, the slope risk calculation includes:

[0033] Micro-mutation field extraction was performed on the surface model to obtain micro-mutation field data;

[0034] Slope gradient trend map data is generated from micro-mutation field data.

[0035] Surface stress propagation is calculated based on slope gradient trend map data to obtain boundary destability trend data.

[0036] Risk calculations are performed based on boundary destabilization trend data to obtain slope risk data.

[0037] This invention introduces micro-mutation field extraction during slope risk calculation, enabling accurate identification of subtle local deformation features and addressing the shortcomings of traditional slope analysis in identifying wheel ruts and soft subsidence areas. By generating slope gradient trend maps, the spatial distribution of dramatic slope changes is grasped, enhancing early warning capabilities for high-risk areas. Surface stress propagation calculation incorporates dynamic response modeling of surface structures, allowing for quantitative prediction of boundary instability trends, applicable to slippage risk identification in ditches, road shoulders, and soft slope areas. Through the collaborative modeling of these features, the generated slope risk data not only possesses higher spatial resolution but also dynamic evolution trend description capabilities, effectively improving the deployment safety and long-term operational stability of photovoltaic energy storage and charging equipment in rural sites, and preventing equipment damage or operational failure due to hidden site risks.

[0038] Optionally, the micromutation field extraction includes:

[0039] The density field of local neighborhood points is calculated on the surface model to obtain the density data of neighborhood points;

[0040] The neighborhood density change rate is calculated from the neighborhood point density data to obtain the neighborhood density change rate data;

[0041] Sparse variability fields are generated based on neighborhood density change rate data to obtain locally sparse data;

[0042] A local neighborhood topology map is generated based on the surface model to obtain local neighborhood map data;

[0043] Local self-homeomorphic defects are calculated from local neighborhood graph data to obtain local self-homeomorphic defect data;

[0044] A self-homeomorphic defect map is generated from the local self-homeomorphic defect data to obtain point cloud defect map data;

[0045] Slope risk data is obtained by performing masked multi-head self-attention calculation on point cloud defect map data based on local sparse data.

[0046] This invention, by introducing local neighborhood point density field calculation and density change rate extraction, can sensitively capture micro-scale structural anomalies such as surface subsidence and wheel ruts, overcoming the shortcomings of traditional curvature or slope features in identifying sparse and variable regions. Through local neighborhood topology map generation and self-homeomorphic defect calculation, it reveals continuity defects in the local topological structure of the site, improving the accuracy of modeling complex soft surface damage morphologies. By combining sparse and variable fields with defect map data and introducing a masked multi-head self-attention mechanism, it enhances the model's ability to identify and discriminate high-risk morphological features, improving the fine-grained spatial resolution and semantic interpretability of slope risk data. This invention effectively addresses the diverse factors of rural natural environments such as dynamic compaction, local subsidence, and multi-source disturbances, providing more forward-looking and intelligent risk warning support for the deployment of photovoltaic storage and charging stations, significantly improving equipment deployment safety and service life.

[0047] Optionally, the point cloud site feature extraction includes:

[0048] Dense height point cloud projection is performed based on the point cloud site model to obtain height map data;

[0049] Convolutional network feature extraction is performed on the height map data to obtain convolutional network feature data;

[0050] Feature map data is obtained by performing feature map saliency mapping and reorganization on the feature data of the convolutional network.

[0051] Feature entropy modulation mapping is performed on the feature data of the convolutional network to obtain feature entropy data;

[0052] Feature-guided field generation is performed based on feature map data and feature entropy data to obtain point cloud site feature data.

[0053] This invention transforms complex spatial point cloud data into high-quality continuous height maps through dense height point cloud projection, improving the input quality for deep learning feature extraction. The convolutional network feature extraction stage uses a multi-layered convolutional network to perform high-order semantic encoding of surface morphology, enhancing the model's intelligent recognition ability for irregular morphologies (such as soft depressions, wheel ruts, and localized material accumulation areas). Feature map saliency mapping and reorganization effectively highlight features related to deployment risks, improving the model's spatial interpretability. Feature entropy modulation mapping introduces an information entropy field to dynamically weight feature data, enhancing the feature representation ability of highly complex areas. Through feature-guided field generation, high-order intelligent fusion of feature maps and entropy fields is achieved, constructing point cloud site feature data with stronger site adaptability and deployment decision value, significantly improving the robustness and practicality of intelligent deployment schemes for photovoltaic storage and charging stations in rural site conditions.

[0054] Optionally, S3 includes:

[0055] Based on the point cloud site feature data, the site curvature anomaly area is removed to obtain the initial site screening data;

[0056] Topographic stability weights are calculated on the initial site screening data to obtain topographic stability data;

[0057] Based on point cloud site feature data and terrain stability data, feature avoidance zone field modeling is performed to obtain deployment optimization area data.

[0058] This invention effectively eliminates areas with significant local morphological abrupt changes or surface defects by removing curvature anomalies based on point cloud site feature data, reducing deployment risk points and improving the overall accuracy of deployment area selection. The terrain stability weight calculation incorporates multi-dimensional surface features (such as slope smoothness, sparsity variability, and topological defects) to dynamically assess the impact of micro-topographic changes on equipment stability, providing a fine-grained risk assessment basis for deployment strategies. Feature avoidance zone field modeling automatically generates avoidance zones for high-risk feature areas (such as soft depressions and areas prone to dynamic disturbances) by fusing site feature data and terrain stability data, guiding the planning of optimal deployment areas and ensuring the deployment safety and long-term operational stability of equipment in complex and dynamic rural environments. This invention significantly improves the intelligent deployment capabilities and spatial decision-making reliability of photovoltaic, energy storage, and charging stations in unstructured rural sites.

[0059] Optionally, S4 includes:

[0060] Deployment attitude simulation is performed based on the deployment preference area data to obtain deployment attitude data;

[0061] Lighting and shadow evolution simulation was performed on the deployment attitude data to obtain lighting simulation data;

[0062] The deployment posture data is optimized based on the lighting simulation data to obtain optimized deployment posture data.

[0063] By performing device collaborative deployment conflict detection on the deployment posture optimization data, intelligent deployment data is obtained.

[0064] This invention introduces deployment posture simulation based on the preferred deployment area, enabling dynamic simulation of equipment posture adaptability in complex rural sites and early detection of deployment stability issues caused by uneven ground or localized soft depressions. Illumination and shadow evolution simulation, combined with time-series data, accurately assesses illumination availability and dynamic changes in shading during the deployment cycle, providing a basis for optimizing equipment energy efficiency. Intelligent optimization of deployment posture based on illumination simulation results achieves a synergistic improvement in posture adjustment and maximizing illumination benefits. Equipment collaborative deployment conflict detection, through spatial layout simulation, proactively avoids spatial conflicts and access obstacles during multi-device deployment, significantly improving the intelligent deployment efficiency and safety of photovoltaic energy storage and charging stations in highly dynamic and obstacle-prone rural sites, ensuring long-term stable operation of equipment and optimal solar energy utilization.

[0065] Optionally, this application also provides a smart management system for rural photovoltaic, energy storage, and charging stations that incorporates point cloud data, for executing the smart management method for rural photovoltaic, energy storage, and charging stations that incorporates point cloud data as described above. The smart management system for rural photovoltaic, energy storage, and charging stations that incorporates point cloud data includes:

[0066] The temporary site point cloud data acquisition module is used to acquire point cloud data of temporary deployment sites;

[0067] The point cloud site modeling and feature extraction module is used to perform point cloud site modeling based on point cloud data to obtain a point cloud site model; and to extract point cloud site features based on the point cloud site model to obtain point cloud site feature data.

[0068] The deployment optimization region calculation module is used to calculate the deployment optimization region based on the point cloud site feature data to obtain the deployment optimization region data.

[0069] The intelligent deployment simulation module is used to simulate intelligent deployment based on the data of the preferred deployment area, and obtain intelligent deployment data.

[0070] The purpose of this invention is to introduce point cloud data into the intelligent management of rural photovoltaic-storage-charging stations, enabling high-precision reconstruction of the real spatial morphology of temporary deployment sites and solving the problem of insufficient identification capabilities for soft terrain and irregular obstacle areas in traditional deployment processes. Through point cloud site modeling, combined with pseudo-physics field fitting and topology optimization methods, a site model with good continuity and accurate boundaries is constructed, laying a high-quality foundation for feature extraction. During point cloud site feature extraction, convolutional neural feature fields and entropy field mapping are integrated to enhance the model's intelligent perception capability of irregular shapes (such as wheel ruts, soft depressions, and temporary material piles). In the deployment optimization area calculation stage, curvature anomaly removal, terrain stability assessment, and avoidance zone modeling are used to achieve fine screening of the spatial safety and equipment stability of the deployment area. In intelligent deployment simulation, through attitude simulation, illumination optimization, and cooperative conflict detection, an intelligent deployment scheme that balances energy efficiency, safety, and cooperative accessibility is generated, comprehensively improving the intelligent deployment efficiency and long-term operational reliability of photovoltaic-storage-charging stations in complex and dynamic rural environments. Attached Figure Description

[0071] Other features, objects, and advantages of this application will become more apparent from the following detailed description of the non-limiting embodiments, taken with reference to the accompanying drawings:

[0072] Figure 1 A flowchart illustrating the steps of an embodiment of an intelligent management method for rural photovoltaic storage and charging stations that incorporates point cloud data is shown.

[0073] Figure 2 A flowchart illustrating the steps of a temporary site cloud data acquisition method according to an embodiment is shown;

[0074] Figure 3 A flowchart illustrating the steps of a point cloud site modeling method according to an embodiment is shown.

[0075] Figure 4 A flowchart illustrating the steps of a method for calculating a preferred deployment area according to one embodiment is shown.

[0076] Figure 5 A flowchart illustrating the steps of an embodiment of an intelligent deployment simulation method is shown.

[0077] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0078] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0079] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. Functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0080] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0081] Please see Figures 1 to 5 This application provides a smart management method for rural photovoltaic storage and charging stations that combines point cloud data, the method comprising:

[0082] S1. Obtain point cloud data of the temporary deployment site;

[0083] In one embodiment, a mobile terminal laser scanning device, such as a vehicle-mounted LiDAR, a drone-mounted LiDAR, or a handheld LiDAR device, is used to perform a rapid low-resamplement pre-scan of the deployment site to collect initial point cloud data. For the initial point cloud data, the local point cloud density distribution is evaluated. If low-density void areas are detected, a supplementary scanning strategy is automatically triggered to fill in the missing point cloud information. The original point cloud data is preprocessed, with isolated noise points removed based on a point cloud density threshold. Surface points are identified and separated through plane fitting, and the surface point set is optimized using normal vector filtering technology. After the above processing, point cloud data with uniform spatial distribution, sufficient noise point removal, and clear surface structure is obtained.

[0084] S2. Perform point cloud site modeling based on point cloud data to obtain point cloud site model; extract point cloud site features based on point cloud site model to obtain point cloud site feature data.

[0085] In one embodiment, the point cloud site modeling process includes introducing a preset scene library model to match the currently deployed site type. Scene types may include soft farmland, hardened ground, composite sites, etc., and generating corresponding pseudo-historical benchmark surface data based on the matching results. Pseudo-physics boundary fitting is then performed on the currently collected point cloud data, according to a preset... Parameters are used to construct a Delaunay triangulation network for point cloud data, and then parameters that meet the requirements are selected. Boundary cells with radius constraints are pruned by trimming high-curvature or long boundary chains to form smooth and spatially connected effective site boundaries, eliminating isolated points or soft boundary noise, and generating optimized site point cloud data. The optimized point cloud is then surface-fitted using a local neighborhood approach to generate a spatially continuous surface model (Z-axis represents elevation, X and Y axes represent horizontal position). For dynamic occlusion areas in the site, an entropy filtering method combining spatial and temporal entropy is used to remove short-term dynamic obstacle point clouds, resulting in a steady-state occlusion area layer. Combining the surface model, slope risk calculation results, and spatial accessibility analysis results, a point cloud site model containing multi-level spatial features is constructed. During point cloud site feature extraction, the surface model is projected to generate a dense two-dimensional height map, which is used as input to extract multi-level spatial feature maps using a convolutional neural network. Through feature entropy modulation, the extracted feature maps are weighted according to their information entropy magnitude to enhance the feature representation of high-risk areas and areas with complex terrain. By performing global average pooling on the convolutional feature maps, the global response intensity of each channel is calculated and used as the channel saliency weight. Subsequently, the feature maps of each channel are weighted and fused with their corresponding saliency weights to form a two-dimensional spatial saliency map, which enhances the interpretability and discriminativeness of the feature regions and generates a structured point cloud site feature dataset.

[0086] S3. Calculate the optimal deployment area based on the point cloud site feature data to obtain the optimal deployment area data;

[0087] In one embodiment, for point cloud site feature data, a curvature anomaly region removal method is used for preliminary screening. The difference between the local principal curvature mean and the overall principal curvature mean is calculated. Based on a set principal curvature difference threshold, abnormal regions with obvious curvature abrupt changes are automatically removed. After removing abnormal regions, for the remaining site area, local slope entropy (used to characterize the complexity of slope changes), flatness index (calculated using the root mean square difference of point cloud elevations within the neighborhood), and topological defect rate (calculated based on neighborhood topological connectivity and defect index) are calculated region by region using a fixed neighborhood window (e.g., radius 0.5 meters). Combining these multi-dimensional indicators, a terrain stability weight field layer is assigned.

[0088] By combining highly saliency feature areas (including detected wheel rut areas, soft depression areas, and material accumulation areas) and steady-state occlusion areas, a feature avoidance zone field layer is constructed. Through feature map fusion (e.g., saliency feature mapping fusion and avoidance zone weight overlay), a complete risk avoidance layer is established. In the optimal deployment area stage, a field-guided spatial energy minimization optimization strategy is adopted. Through site risk calculation, the terrain stability field, feature avoidance zone field, and spatial accessibility field are jointly considered to dynamically optimize the spatial distribution of candidate deployment units, selecting optimal deployment area data with low overall risk and spatial connectivity.

[0089] S4. Perform intelligent deployment simulation based on the deployment optimization area data to obtain intelligent deployment data.

[0090] In one embodiment, for the preferred deployment area, equipment attitude simulation is performed. Simulation parameters include the equipment's horizontal rotation angle (typically within ±15°), front-to-back and left-to-right tilt angles, and the height adjustment of each support point. By simulating different attitude combinations, attitude feasibility calculations are performed for each combination. The calculation indicators are weighted based on equipment stability (e.g., overturning risk, support balance) and ground contact. Illumination and shadow evolution simulation is performed. Combining the geographical location of the deployment site, seasonal solar trajectory data, and steady-state shading layer information, ray tracing is used to simulate the equipment's illumination availability under different attitudes and at different time periods. The periodic cumulative effective illumination time index is calculated, and an illumination availability score is generated accordingly. Based on the above attitude feasibility score and illumination availability score, a multi-round iterative optimization strategy based on the guidance field is adopted to optimize and adjust the equipment attitude parameters to achieve a balance between maximizing illumination efficiency and optimizing attitude stability. After optimization, for scenarios involving the collaborative deployment of multiple devices, equipment collaborative deployment conflict detection is performed, including device boundary space conflict detection and inter-device travel path conflict detection. For conflicting device groups, a conflict-free intelligent deployment solution is generated through local fine-tuning optimization or dynamic reordering strategies. The output intelligent deployment data includes the deployment pose of each device (position coordinates, rotation angle, support height adjustment value), recommended deployment time window, and multi-device collaboration information. This data will be available for use by the on-site deployment terminal system to guide actual deployment operations.

[0091] Optionally, S1 includes:

[0092] S11. Control the terminal laser device to perform low resampling pre-scanning to obtain initial point cloud data;

[0093] In one embodiment, a mobile terminal laser scanning device, such as a typical 16-line or 32-line LiDAR module, is used, configured in a low line count, high scanning speed mode, to perform a preliminary full-field scan of the deployment site. During the scanning process, the scan line interval is set to be greater than 10 cm, and the point cloud spatial resolution is less than 5 points per square meter, to quickly complete the preliminary point cloud data acquisition within the site area. During the pre-scanning process, the moving speed of the control device is less than or equal to 3 km / h, and the scanning frame rate is greater than or equal to 10 Hz, thereby ensuring that the overall site scanning coverage reaches or exceeds 95%. The original echo intensity attribute, i.e., laser reflection intensity information, is retained in the initially acquired point cloud data.

[0094] S12. Perform preliminary estimation of surface carrying capacity on the initial point cloud data to obtain carrying capacity data;

[0095] In one embodiment, the collected initial point cloud data is projected onto a two-dimensional planar grid. Specifically, the point cloud is mapped onto the XY plane using orthogonal projection, dividing it into equally spaced two-dimensional grid cells. A grid size of 0.5 m × 0.5 m is recommended. For each grid cell, the point cloud density (i.e., the number of points per unit area), elevation variance (i.e., the variance of the point cloud height values ​​within the cell), and mean echo intensity (i.e., the average value of laser reflection intensity) are calculated. Based on these three indicators, a preliminary estimation model for surface carrying capacity is established. The carrying capacity of each grid cell is then calculated using a weighted summation method. ,in For the first The data carried by the grid cells For point cloud density, the weight data is mapped forward. This is the forward mapping data of point cloud density. For back-mapping weighted data of elevation variance, This is the inverse mapping data of elevation variance. For echo intensity, positively mapped weighted data. The echo intensity is mapped forward; the load-bearing capacity data layer is output and the high load-bearing, medium load-bearing, and low load-bearing areas are classified and labeled.

[0096] S13. Perform key area stability extraction processing on the carrying capacity data to obtain key area stability data;

[0097] In one embodiment, based on the generated carrying capacity data layer, stability extraction processing is performed on key areas of the carrying capacity data. Local analysis is conducted on the carrying capacity layer using a sliding window with adjacent grids; a typical sliding window size is 3 rows. The system uses three grid cells, with each central grid cell as the core, and combines data from its eight neighboring grid cells to calculate a local stability index. This index is composed of the mean and variance of the carrying capacity scores within the neighborhood. For each central grid cell, the average carrying capacity score of its neighboring grid cells is calculated, and then the variance of the carrying capacity scores within the adjacent regions is subtracted to form the stability index. Based on a preset stability threshold, grid cell regions with stability indices higher than the threshold are selected to form a set of high-stability regions.

[0098] S14. Based on the stability data of key areas, control the terminal laser device to perform high-precision back-scan of the initial point cloud data in key areas to obtain point cloud data.

[0099] In one embodiment, based on the preferred area coordinate range determined by the stability data of the key area, the terminal laser scanning device is controlled to perform a high-density supplementary scanning operation on the key area. During the supplementary scanning process, the laser scanning device switches to a high line count mode, configured with 64 lines or more, or adopts a high frame rate scanning mode, with the frame rate set above 20 Hz. During the scanning path planning process, it is ensured that the point cloud sampling density of the key area is higher than 100 points per square meter, and the scanning coverage of a single area is higher than 99%, thereby ensuring the integrity of spatial detail information and the accuracy of the key area. The high-precision point cloud data collected during the supplementary scanning process will retain the same timestamp information as the initial scanning data.

[0100] Optionally, the point cloud site modeling includes:

[0101] S21. Match the scene type with the point cloud data and the preset scene library model to generate pseudo-historical site benchmark data.

[0102] In one embodiment, the pre-defined scene library model includes various typical rural site types, such as soft farmland, hardened ground, composite compacted sites, and mixed vegetation sites. Each scene model is associated with a corresponding feature vector, and the features include key spatial attribute indicators such as point cloud density, elevation variance, surface roughness index, and mean reflection intensity. For the currently collected point cloud data, the above feature indicators are extracted across the entire field, and a global feature vector for the current site is generated. A feature similarity measurement method is used, for example, by calculating the cosine similarity between the current feature vector and the feature vectors of each typical scene model in the scene library, to select the matching scene model with the highest similarity. The historical site benchmark data corresponding to the matched scene model will be used as the pseudo-historical benchmark data for the current site.

[0103] S22. Perform pseudo-physical field boundary fitting on the point cloud data to obtain optimized point cloud data;

[0104] In one embodiment, the acquired raw point cloud data is processed according to a preset method. First, the point cloud data is subjected to Delaunay triangulation to construct a spatial connectivity network; then, based on the parameters... A radius threshold is used to filter out triangular facets that meet the boundary features, and those that do not are gradually eliminated. The long side or thin slice of the constraint condition is used to extract the outer contour line with good spatial connectivity, forming the preliminary site boundary. The parameters are typically dynamically adjusted based on the average site dimensions, ranging from 1% to 5% of the average dimensions. In local boundary point processing, for each vertex on the boundary curve, a Laplace operator is used for smooth updates based on the positions of adjacent vertices. After these steps, denoised and boundary-optimized point cloud data is generated, serving as input data for the subsequent surface modeling module.

[0105] S23. Fit the surface morphology based on the point cloud optimization data and pseudo-historical site benchmark data to generate a surface model.

[0106] In one embodiment, surface morphology fitting is performed based on optimized point cloud data and matched pseudo-historical site benchmark data. Specifically, the optimized point cloud data and pseudo-historical benchmark surface data are bidirectionally fused. A weighted fusion strategy is used to integrate the elevation values ​​of both, with the fusion weight parameter set between 0.7 and 0.9. The currently acquired data is prioritized, while historical benchmark data serves as an auxiliary factor, enhancing the stability and continuity of the overall fitting. For low-density areas (i.e., areas where the point cloud density per unit area is below a preset threshold) identified during the fusion process, radial basis function (RBF) interpolation is used for spatial completion.

[0107] S24. Perform entropy filtering on the point cloud optimization data to obtain steady-state occlusion data;

[0108] In one embodiment, entropy filtering is used to extract steady-state occlusion information from the optimized point cloud data. Specifically, a local neighborhood is used as the analysis unit, with a neighborhood radius set between 0.5 meters and 1 meter. Spatial entropy and temporal entropy are calculated separately. Spatial entropy is calculated by statistically analyzing the spatial uniformity of the point cloud distribution within the neighborhood, based on the probability of local point density distribution. Temporal entropy is calculated by analyzing the time-series changes of the point cloud at the same location, calculating the height fluctuation of the point cloud in multiple frames at that location, and evaluating the frequency of dynamic changes. A weighted combination method is used to fuse the spatial and temporal entropies. The fusion coefficient is adjusted according to the dynamic characteristics of different sites, ranging from 0.6 to 0.8, with spatial entropy as the dominant factor. A filtering threshold is set based on the fused entropy value to remove high-entropy areas (i.e., dynamic occlusion areas), retaining only low-entropy steady-state occlusion points, forming a steady-state occlusion data layer.

[0109] S25. Calculate the slope risk of the surface model to obtain slope risk data;

[0110] In one embodiment, for the generated surface model, the first and second derivatives in the X and Y directions are calculated using numerical differentiation to obtain local slope values ​​and curvature change information. The first derivative is used to describe the local slope (gradient) change trend, and the second derivative is used to identify local curvature anomalies. Based on the changes in the second derivative, micro-mutation fields are extracted, and slope mutation points and areas of subtle topographic changes are calculated. The gradient magnitude is further calculated from the slope vector field formed by the first derivative to generate a slope gradient trend map, characterizing the spatial distribution characteristics of the slope change rate. Combined with boundary destabilization trend modeling, a stress propagation model is used to transmit stress from local high-risk slope areas to surrounding areas, simulating the regional destabilization trend caused by slope changes. In the stress propagation model, the propagation coefficient is dynamically adjusted according to the site type, and the propagation path is guided by the slope gradient trend map and micro-mutation fields. Through the above calculation process, a slope risk data layer is generated.

[0111] S26. Based on steady-state shading data and slope risk data, spatial accessibility processing is performed to obtain spatial accessibility data;

[0112] In one embodiment, the steady-state occlusion data layer is used as the hard obstacle input, defined as an impassable area; areas with a risk level of 2 or higher in the slope risk data layer are used as the soft obstacle input, defined as traversable areas where path costs increase. The system uses a preset main access path (e.g., the path start and end points determined based on site entrances, road access points, etc.) as the path search baseline. During path accessibility analysis, each candidate deployment area unit is used as the path start point, and the main access path is used as the target. A step-by-step path expansion method is adopted, and the cumulative cost is evaluated according to the movement cost between adjacent units. Ordinary areas are assigned a standard cost, soft obstacle areas are assigned an increase coefficient (1.5 to 3 times the standard cost), and hard obstacle areas are set as impassable. By dynamically expanding path nodes, the current optimal cumulative cost is continuously updated until the path with the lowest cost from the candidate deployment area to the main access path is found. The accessibility index is calculated in reverse based on the cumulative path cost; the higher the score, the better the path accessibility. A spatial accessibility data layer is generated based on the accessibility scores of all candidate deployment areas.

[0113] S27. Based on the spatial accessibility data and the surface model, a point cloud site model is obtained by fusing them.

[0114] In one embodiment, elevation data (Z-value distribution), spatial accessibility data, and slope risk data from the surface model are fused to form a point cloud site model with a unified data structure. A unified spatial grid division method (e.g., 0.5 meters) is adopted. A spatial feature data unit containing multi-dimensional attributes is constructed by integrating elevation information, accessibility score, and slope risk score within each 0.5-meter raster cell.

[0115] Optionally, the entropy filtering process includes:

[0116] Dual-domain spatial entropy calculation is performed on the point cloud optimization data to obtain dual-domain spatial entropy data;

[0117] In one embodiment, for the optimized point cloud data, a local three-dimensional neighborhood is established within a radius of 0.5 meters to 1.0 meter for each point cloud sampling point, and dual-domain spatial entropy calculation is performed. The system calculates the geometric entropy value by dividing the elevation distribution of the point cloud within the neighborhood into M fan-shaped regions according to azimuth angle. The elevation variance of the point cloud within each fan-shaped region is statistically analyzed, and the proportion of the elevation variance of each region to the total variance is calculated. Based on this proportion, the geometric entropy value of the current point is calculated. The visibility entropy value is calculated by evaluating the occlusion relationship of the current point within the neighborhood. Using the current point as a reference, several rays (in directions 16 to 32) are emitted in all directions, and the obstruction of the rays by other points within the neighborhood is statistically analyzed along each direction. The occlusion rate in all directions is statistically analyzed, and the occlusion ratio, ranging from 0 to 1, is calculated to represent the spatial visibility level of the point. The visibility entropy value is calculated using the Shannon entropy formula based on this index. A weighted combination strategy is used to fuse the geometric entropy value and the visibility entropy value to form a combined entropy value. The combined weighting coefficients are dynamically adjusted based on the site type, and the sum of the two is 1. In steep terrain scenes, a geometric entropy weight is added to enhance sensitivity to changes in slope complexity; in flat scenes, a visibility entropy weight is added to highlight the impact of occlusion changes. Through the above process, a fused dual-domain spatial entropy data layer is obtained.

[0118] Based on the dual-domain spatial entropy data, preliminary marking of candidate occlusion points is performed on the point cloud optimization data to obtain candidate occlusion point marking data;

[0119] In one embodiment, preliminary labeling of candidate occlusion points is performed on the optimized point cloud data based on dual-domain spatial entropy data. A spatial entropy threshold is set as the basis for determining whether there is a high-complexity occlusion risk in the current point cloud region. Simultaneously, a low occlusion probability threshold is defined and set to 0.3. During the screening process, for each point cloud sampling point, if its combined entropy value is higher than the preset entropy threshold and its local visibility score is lower than the low occlusion probability threshold, then the point is initially determined to be a candidate occlusion point. The set of point cloud sampling points that meet the above conditions is labeled as candidate occlusion point labeling data, encoded in the form of a binary mask, with each point corresponding to a label of 1 (candidate occlusion point) or 0 (non-occlusion point).

[0120] Entropy gradient field data is generated by processing the candidate occlusion point marker data.

[0121] In one embodiment, for the candidate occlusion point marker data, a local neighborhood (e.g., a radius of 0.5 to 1.0 meters) is defined around each candidate occlusion point, and a local entropy gradient field is calculated. Using the numerical difference method, the first-order partial derivatives of the combined entropy value in the X and Y directions are calculated respectively to form a local entropy gradient vector, characterizing the trend and intensity of local entropy value changes. For regions with high entropy gradient magnitudes, located at the boundary of the candidate occlusion area, the system enhances the response weight of this region during the optimization process by weighting it, such as multiplying by 1.1 or 1.2. After completing the entropy gradient calculation, the entropy gradient field is standardized to generate a standardized entropy gradient layer. The entropy gradient magnitude is mapped to a standard interval of 0 to 1 to form a standardized entropy gradient field data layer.

[0122] Based on the entropy gradient field data, Markov correction modeling is performed to obtain the entropy gradient field model.

[0123] In one embodiment, a Markov random field model is constructed for the standardized entropy gradient field data, and spatial consistency correction of candidate occlusion point labels is performed using an energy optimization method. This is calculated using the overall energy function. ,in This is the global energy function value. Number the current point cloud sampling point. A univariate potential term describing the sampling points of the point cloud. The state deviates from the entropy threshold The cost, These are the weighting coefficients of the binary potential term. For the set of adjacent point pairs, This is an adjacency graph. This is a binary potential term that describes the cost of maintaining state consistency between adjacent points. For point The current state of the occluded label. For point The current state of the occluded label; It includes univariate and binary potential terms: The univariate potential term describes the cost of the current point cloud sampling point's state deviating from the entropy threshold, and is defined as the absolute value of the difference between the current point's combined entropy value and the preset entropy threshold, i.e. ,in A univariate potential term describing the sampling points of the point cloud. The state deviates from the entropy threshold The cost, For point The combined entropy value, A preset entropy threshold is used to divide the occluded and unoccluded regions; a binary potential term is used to describe the cost of state consistency between neighboring points, which is measured by the product of the difference in the normal vectors of neighboring points and the state difference index, i.e. ,in This is a binary potential term that describes the cost of maintaining state consistency between adjacent points. For point The normal vector, For point The normal vector, This is an indicator function; it takes a value of 1 when the two labels are different (i.e., their label values ​​are different), and a value of 0 when they are the same. For point The current state of the occluded label. For point The current occlusion label state is determined. The point cloud normal vector is pre-calculated using a local neighborhood fitting method. During energy optimization, the edge probability distribution of each point is initialized. Based on the adjacency graph, messages (i.e., potential functions) between adjacent points are iteratively passed to update the edge probability distribution of each point, gradually approaching the global optimal solution. This achieves the optimization of the overall energy function and yields the corrected occlusion label model.

[0124] Steady-state occlusion segmentation is performed based on the entropy gradient field model to obtain steady-state occlusion data.

[0125] In one embodiment, high-response regions in the entropy gradient field model are marked for connectivity. By merging point cloud sampling points with consistent occlusion labels, continuous spatial regions are identified, completing the initial extraction of steady-state occlusion areas. For the marked connected occlusion regions, a filtering process is performed based on region area, removing isolated occlusion regions with areas below a preset threshold (set to 0.25 square meters), retaining only the main steady-state occlusion regions with significant spatial continuity and occlusion stability. After the above processing, a steady-state occlusion data layer is output.

[0126] Optionally, the slope risk calculation includes:

[0127] Micro-mutation field extraction was performed on the surface model to obtain micro-mutation field data;

[0128] In one embodiment, a local sliding window is used to extract micro-mutation fields for the constructed surface model. A local neighborhood window with a radius of 0.5 to 1.0 meters is constructed, centered on each two-dimensional grid point. The point cloud density change within this neighborhood is calculated; that is, the difference between the point cloud density value of the current grid point and the average density value of its neighborhood is defined as the density gradient change rate. Based on the elevation function in the surface model, the second-order spatial partial derivatives of each grid point in the X and Y directions are calculated. The square root of the squares of the second derivatives in both directions is used to form the micro-mutation intensity factor. The micro-mutation value for each grid point is generated by weighted fusion of the density gradient change rate and the micro-mutation intensity factor. The fusion weight coefficients can be set according to the actual site characteristics, where the density term weight reflects the influence of point cloud aggregation characteristics, and the curvature term weight reflects the significance of terrain mutations. The micro-mutation field data layer is output, forming a spatially continuous two-dimensional risk index map.

[0129] Slope gradient trend map data is generated from micro-mutation field data.

[0130] In one embodiment, a spatial gradient calculation method is used to extract slope gradient trend map data from the generated micro-mutation field data layer. Specifically, based on two-dimensional spatial directions (X and Y directions), the first-order partial derivatives of the micro-mutation score values ​​in the X and Y axes are calculated to form a spatial gradient vector field, describing the spatial variation trend of the micro-mutation score values. By calculating the magnitude of the gradient vector at each location, the slope gradient trend index for that location is obtained. By calculating the slope gradient trend index for each grid cell across the entire field, a complete slope gradient trend map data layer is formed.

[0131] Surface stress propagation is calculated based on slope gradient trend map data to obtain boundary destability trend data.

[0132] In one embodiment, for the generated slope gradient trend map data, the slope gradient trend index is discretized and mapped to regular grid nodes, with each node corresponding to a spatial location unit. Based on the slope gradient similarity between nodes, a coupling relationship matrix between nodes is constructed to characterize the slope trend coupling strength between adjacent nodes. The dynamic diffusion process of slope trend in local areas of the land surface is simulated to obtain a surface stress propagation model. Within each iteration step, the surface stress state value of each node is updated. The update rule is that the stress state of the current node is a weighted accumulation of its previous state value and the slope trend difference of adjacent nodes. The weight coefficient (i.e., the propagation coefficient) is dynamically adjusted according to the soil's anti-slip capacity, and the parameter values ​​are referenced from the typical anti-slip capacity parameters of the corresponding landform type in the preset scenario library. Through the iterative propagation process, the stress state of each node is continuously updated until a stable convergence state is reached (e.g., the node state change is less than a preset threshold during continuous iteration). Based on the cumulative stress state of each node in the stable state, the destabilization trend index is extracted to obtain the boundary destabilization trend data layer.

[0133] Risk calculations are performed based on boundary destabilization trend data to obtain slope risk data.

[0134] In one embodiment, the boundary destabilization trend data layer, the slope gradient trend map data layer, and the micro-mutation field data layer are used as multiple input feature fields and jointly input into the feature fusion calculation module. The feature fusion calculation module employs a feature fusion network structure based on a self-attention mechanism, using a Transformer Encoder structure to jointly model the aforementioned spatial feature maps. By introducing a self-attention mechanism, the correlation of slope risk between different spatial locations can be effectively captured, and the dependencies between local risk features can be modeled. During the feature fusion process, each input feature field is encoded into a feature vector sequence, and feature interaction and information integration are performed through a multi-layer Transformer Encoder network to generate a slope risk score index corresponding to each spatial location.

[0135] Optionally, the micromutation field extraction includes:

[0136] The density field of local neighborhood points is calculated on the surface model to obtain the density data of neighborhood points;

[0137] In one embodiment, the point cloud sampling point set corresponding to the surface model is used as the basis for calculation. For each sampling point, a local neighborhood radius is set, ranging from 0.5 meters to 1.0 meters, to construct a local neighborhood centered on the current point. Within this neighborhood, the number of valid point cloud sampling points is counted. The point cloud density value per unit area is calculated by dividing the number of point cloud points in the neighborhood by the corresponding projected area of ​​the neighborhood (i.e., the square of the neighborhood radius multiplied by pi). By performing the above local neighborhood density calculation on all sampling points in the surface model, a neighborhood point density field data layer is generated.

[0138] The neighborhood density change rate is calculated from the neighborhood point density data to obtain the neighborhood density change rate data;

[0139] In one embodiment, the neighborhood density change rate is calculated for the generated neighborhood point density data. Based on a spatial two-dimensional grid structure, a calculation window is constructed using each central grid cell as a foundation; the typical window size is 3 rows. 3 columns or 5 rows Five adjacent raster cells are used. Within each calculation window, the average point cloud density value of all adjacent raster cells within that window is calculated and used as the average density of the adjacent cells corresponding to the central raster. The difference between the current point cloud density value of the central raster cell and the average density of the adjacent cells is calculated and defined as the neighborhood density change rate. By performing the above calculation process sequentially on all raster cells within the overall spatial region, a neighborhood density change rate data layer is generated.

[0140] Sparse variability fields are generated based on neighborhood density change rate data to obtain locally sparse data;

[0141] In one embodiment, for the generated neighborhood density change rate data layer, the density change rate value of the current raster cell is normalized to the maximum absolute value of the density change rate in the entire spatial region to generate a normalized sparsity score. Based on the normalized sparsity score, a threshold screening method is used to extract sparse anomaly regions. During the screening process, a normalized sparsity score threshold is set, with a value ranging from 0.3 to 0.5. When the normalized sparsity score of a raster cell is higher than the preset threshold, the region is determined to have significant sparsity variability and is marked as a sparse variability anomaly region. By screening all raster cells in the entire region one by one, a local sparse variability field data layer is formed.

[0142] A local neighborhood topology map is generated based on the surface model to obtain local neighborhood map data;

[0143] In one embodiment, for the surface model, a local neighborhood topology is constructed using each sampling point as the center point. A neighborhood radius is set around this center point, ranging from 0.5 meters to 1.0 meter. All valid point cloud sampling points within this radius are selected as the node set of the topology subgraph. For the selected local neighborhood nodes, topological edges are established using spatial adjacency relationships. Adjacency edges can be formed in two ways: first, by constructing an optimal non-overlapping triangular network within the local neighborhood to form the edge set of the local topology; second, for each node, the k nearest adjacent nodes are selected, and topological connections are established using these. The value of k is dynamically adjusted based on the neighborhood density. Through these steps, a corresponding local neighborhood topology is generated sequentially for each sampling point in the surface model.

[0144] Local self-homeomorphic defects are calculated from local neighborhood graph data to obtain local self-homeomorphic defect data;

[0145] In one embodiment, for the generated local neighborhood topology map data, local topological structure indices are calculated, including the Eulerian characteristic number and the Betti number. The Eulerian characteristic number is used to describe the overall characteristics of connected regions and the number of holes in the local topology map; the Betti number is used to characterize the number of independent connected components and loops or holes in the topology map. The topological defect degree between the current topology map and the ideal smooth terrain reference model is calculated. Using the predefined reference Eulerian characteristic number and Betti number in the ideal smooth terrain model as benchmarks, the difference between the current topology map and the reference value is calculated respectively, and the sum is used to obtain the topological defect degree index of the local area. By calculating the above topological defect degree for each neighborhood topology map in the overall surface model, a local autosomal defect data layer is generated.

[0146] A self-homeomorphic defect map is generated from the local self-homeomorphic defect data to obtain point cloud defect map data;

[0147] In one embodiment, the calculated local self-homeomorphic defect data is mapped to a range of 0 to 1, resulting in a standardized topological defect index. The standardized defect index is obtained by dividing the current defect index value by the maximum value of the overall defect index. Based on the standardized topological defect score, a topological defect judgment threshold is set, ranging from 0.2 to 0.4. Regions with standardized scores higher than the preset threshold are identified as significant topological defect areas and explicitly marked. By judging each region individually, a spatially continuous point cloud defect map data layer is generated.

[0148] Slope risk data is obtained by performing masked multi-head self-attention calculation on point cloud defect map data based on local sparse data.

[0149] In one embodiment, a masked multi-head self-attention mechanism is used to jointly calculate the slope risk score for micro-mutation regions by combining the generated local sparse variability data and point cloud defect map data. A sparse mask layer is constructed based on the local sparse variability score. When the sparse variability score at a certain spatial location is higher than a preset threshold (range 0.3 to 0.5), it is determined to be a sparse anomaly region, and the corresponding mask value is set to 1; otherwise, it is set to 0. The point cloud defect map is used as feature input, combined with the sparse mask layer, and input to a feature fusion network based on the multi-head self-attention mechanism. The network dynamically adjusts the attention weight distribution by setting the point cloud defect map as the query, key, and value input, and the sparse mask as the attention mask. Through the above attention calculation process, a slope risk score layer containing spatial consistency and anomaly enhancement features is generated.

[0150] Optionally, the point cloud site feature extraction includes:

[0151] Dense height point cloud projection is performed based on the point cloud site model to obtain height map data;

[0152] In one embodiment, an orthogonal projection method is used to generate a dense height map for the constructed point cloud site model. The point cloud site model is projected vertically onto a horizontal two-dimensional plane. For each horizontal position cell, the maximum elevation value of the point cloud sampling point at that position is selected as the projected elevation value of the corresponding cell. This method forms a height projection layer. During the projection process, the spatial resolution of the projection grid cells is set, ranging from 0.1 meters to 0.2 meters, and can be dynamically adjusted according to the actual site size. For sparse or missing measurement areas, bilateral interpolation is used to complete the data, avoiding voids or discontinuities. A continuous and dense height map data layer is output.

[0153] Convolutional network feature extraction is performed on the height map data to obtain convolutional network feature data;

[0154] In one embodiment, a multi-layer convolutional neural network (CNN) structure is used for multi-level feature extraction on the generated dense heightmap data. The heightmap data is used as the input image and fed into the CNN to extract spatial feature maps at different scales layer by layer. During feature extraction, the feature maps of the third and fourth layers of the CNN are selected as key feature layers, representing the spatial feature expression capabilities at medium and large scales, respectively. The feature maps of the two layers are fused through feature concatenation to form a composite-scale convolutional feature representation. The fused feature map is then normalized (using layer normalization or batch normalization methods). Multi-channel convolutional feature map data is output, with the feature channel dimension set to 64 to 128 dimensions.

[0155] Feature map data is obtained by performing feature map saliency mapping and reorganization on the feature data of the convolutional network.

[0156] In one embodiment, global average pooling is used to calculate the global response intensity of each channel in the convolutional feature map, which serves as the saliency weight. The calculated saliency weight for each channel is multiplied by the corresponding channel's convolutional feature map to form a weighted single-channel saliency feature map. All weighted single-channel saliency feature maps are then weighted and fused to generate a two-dimensional spatial saliency map. Through this mapping and recombination process, a feature map data layer is output.

[0157] Feature entropy modulation mapping is performed on the feature data of the convolutional network to obtain feature entropy data;

[0158] In one embodiment, for each spatial location (pixel) of the convolutional feature map, standardization is performed along the channel dimension, normalizing the feature intensity of each channel into a probability distribution. For the standardized feature probability distribution, the corresponding feature entropy value is calculated pixel-by-pixel based on the Shannon entropy calculation method. High-entropy regions correspond to areas with high feature complexity and drastic spatial variations; low-entropy regions correspond to areas with simple features and stable structures. In the entropy threshold setting process, a global statistical analysis method is used to dynamically set the high-entropy judgment threshold based on the overall feature entropy value distribution. The value range can be set to the upper quantile interval of the global entropy value distribution (e.g., above the 75th percentile). High-entropy regions are explicitly marked, and a feature entropy layer is output.

[0159] Feature-guided field generation is performed based on feature map data and feature entropy data to obtain point cloud site feature data.

[0160] In one embodiment, a feature-guided field generation module is used to fuse multi-source feature information to generate point cloud site feature data, based on the generated feature map data and feature entropy data. The feature-guided field generation module can adopt two typical structures: First, based on a multilayer perceptron (MLP) structure, it jointly encodes the salient feature map and feature entropy map, fusing spatial feature complexity and salient information through nonlinear mapping to extract a guided feature field with stronger feature representation capabilities; Second, it employs a feature fusion network based on a self-attention mechanism, where feature entropy values ​​are used as guiding factors for attention weights in self-attention calculation, dynamically adjusting the intensity of feature interactions between spatial locations. In the self-attention mechanism, the feature map is used as the input for query, key, and value, and an entropy weight guiding matrix is ​​generated for the corresponding location based on the feature entropy map, adjusting the weight distribution in the attention matrix. By enhancing the feature representation capability of high-entropy and high-complexity regions and suppressing feature redundancy in low-complexity regions, the spatial resolution capability of the overall feature map is improved. The point cloud site feature data output by the feature-guided field generation module is shown.

[0161] Optionally, S3 includes:

[0162] S31. Based on the point cloud site feature data, remove the site curvature anomaly area to obtain the initial site screening data.

[0163] In one embodiment, anomaly curvature areas are removed from the extracted point cloud site feature data. Based on a local neighborhood fitting surface method, the principal curvature values ​​of the surface at each spatial location are extracted, including the maximum and minimum principal curvatures. The average principal curvature value at the current spatial location is calculated. Within the overall spatial range, the average principal curvature values ​​at all locations are statistically analyzed, and the mean and standard deviation of the overall curvature distribution are calculated as the overall curvature benchmark index. Based on the statistical results, a curvature anomaly judgment threshold is set, calculated as the curvature anomaly threshold equal to the mean plus 1.5 or 2 times the standard deviation. For each spatial location, it is determined whether the absolute value of its average principal curvature value is lower than the preset anomaly judgment threshold. If it is lower than the threshold, it is judged as a normal curvature area and retained; if it is higher than the threshold, it is removed and marked as an anomaly area. Through the above filtering process, a preliminary site screening data layer is output.

[0164] S32. Calculate the terrain stability weights on the initial site screening data to obtain terrain stability data;

[0165] In one embodiment, within the area of ​​the initial site screening data, the following stability indices are calculated: calculating the local slope smoothness index, i.e., taking the absolute value of the surface gradient and its negative value; calling the sparse variability data generated by the aforementioned micro-mutation field; calling the topological defect degree data in the defect layer; and fusing the above three indices in a weighted manner to calculate the terrain stability weight. ,in For terrain stability data, For local slope smoothness weighting data, This is local slope smoothness data. For sparse, variable weighted data, For sparse and variable data, This is the topology defect weight data. This is topological defect data.

[0166] S33. Based on the point cloud site feature data and terrain stability data, perform feature avoidance zone field modeling to obtain deployment optimization area data.

[0167] In one embodiment, feature avoidance field modeling is performed on the generated point cloud site feature data and terrain stability data to select preferred deployment areas. A saliency score layer is extracted from the point cloud site feature data. Locations with scores higher than a set threshold (ranging from 0.6 to 0.8) are selected as high-saliency feature areas, initially serving as avoidance mask layers. Multiple risk layers are jointly summarized, including high-saliency feature areas, areas with high sparsity variability (sparse scores higher than a preset threshold), steady-state occlusion areas, and curvature anomaly areas (previously removed), to construct a complete avoidance field layer. In the avoidance field layer, an avoidance weight coefficient is assigned to each spatial location, ranging from 0.7 to 1.0; a higher weight indicates a higher avoidance priority. A preferred deployment index is calculated to quantify the deployment suitability of spatial locations. The deployment index is calculated by multiplying the terrain stability weight of the current area by the complementary term of the avoidance field weight coefficient to obtain the deployment suitability index. A higher index indicates better regional deployment suitability. ,in To deploy suitability index data, The terrain stability weights are obtained by standardizing terrain stability data. To avoid biasing the weighted coefficient data, preset data is used. Based on the deployment index, regions with scores higher than the preferred deployment threshold (ranging from 0.6 to 0.8) are selected as preferred deployment regions.

[0168] Optionally, S4 includes:

[0169] S41. Perform deployment attitude simulation based on the deployment preferred area data to obtain deployment attitude data;

[0170] In one embodiment, deployment posture simulation analysis is performed on the selected preferred deployment area data. The preferred deployment area is divided into several candidate deployment units according to spatial rules. The unit area can be set according to the device type, with a value ranging from 2 square meters to 5 square meters, or flexibly adjusted according to preset parameters.

[0171] For each candidate deployment unit, equipment attitude simulation is performed to simulate various possible deployment states of the equipment at that location. During the simulation, the following attitude parameters are mainly adjusted: the horizontal rotation angle of the equipment, ranging from -15° to +15°; the height adjustment of the equipment support structure, ranging from 0 to the maximum support height; and the forward / backward tilt angle and left / right tilt angle, respectively simulating the pitch and roll attitude changes of the equipment relative to the ground. Based on the parameter adjustments, according to the equipment's physical stability model and the current terrain stability weights, each attitude combination is calculated. Higher stability indicates a more feasible deployment attitude. For each candidate deployment unit, the highest combination of attitude parameters is selected as the optimal deployment attitude for that unit, generating a deployment attitude data layer.

[0172] S42. Perform illumination and shadow evolution simulation on the deployment attitude data to obtain illumination simulation data;

[0173] In one embodiment, based on the geographical location parameters, seasonal variation cycle, and time period characteristics of the deployment area, and combined with typical sunshine data (including average sunshine duration, sunshine incidence angle, and sunshine intensity) from the local meteorological database, a multi-period (daily, weekly, and monthly) sunshine trajectory model is generated to dynamically describe the sunshine evolution characteristics during equipment deployment. During the illumination simulation, based on ray tracing methods, the spatial occlusion situation under the current equipment posture is dynamically simulated, and the effective illumination time per unit area is calculated for each time period. By accumulating the effective illumination time for each simulation period, the illumination value corresponding to the current deployment posture is calculated; a higher value indicates better light utilization efficiency under that posture.

[0174] S43. Optimize the deployment posture data based on the lighting simulation data to obtain optimized deployment posture data;

[0175] In one embodiment, deployment posture optimization is performed on the generated illumination simulation data and preliminary deployment posture data. The optimization objective is the product of the physical stability score and the illumination availability score of the current deployment posture. The optimization direction is to maximize the calculation result, ensuring that the device operates under a combination of posture parameters that combines high stability and excellent illumination conditions. Specifically, the objective function is to maximize the product of the stability score and the illumination score. That is, in the process of adjusting the combination of posture parameters (including horizontal rotation angle, tilt angle, support height adjustment, etc.), the optimal posture configuration is sought to achieve the best deployment effect. During the optimization process, a global optimization algorithm is used, typically including a genetic algorithm or particle swarm optimization algorithm, to iteratively optimize the search space, gradually approaching the global optimum and improving the overall performance of the deployment posture.

[0176] S44. Perform device collaborative deployment conflict detection on the deployment posture optimization data to obtain intelligent deployment data.

[0177] In one embodiment, conflict detection for collaborative device deployment is performed based on the optimized deployment attitude data. Within the global deployment area, device deployment simulation is conducted based on the actual physical boundary dimensions of the devices and the preset minimum safety distance, initially generating a global device deployment plan. During the deployment simulation, for any two device instances, the minimum distance between their spatial boundaries is calculated. If the minimum distance is lower than the safety distance threshold (ranging from 0.3 meters to 0.5 meters), a spatial conflict is determined to exist. For the detected spatially conflicting device units, the following optimization strategies are applied step by step: attempting local position optimization and adjusting the device deployment position; if position optimization is insufficient to eliminate the conflict, fine-tuning of the device attitude is performed, optimizing the device rotation angle or support height parameters; if the conflict still exists, a deployment order adjustment strategy is adopted to reorder the deployment priority. Through the above conflict detection and optimization process, a conflict-free intelligent deployment plan is output. Simultaneously, combined with the analysis of collaborative power generation efficiency between devices, an optimized deployment plan for collaborative power generation timing is added as intelligent deployment data for use by the on-site deployment execution system.

[0178] Optionally, this application also provides a smart management system for rural photovoltaic, energy storage, and charging stations that incorporates point cloud data, for executing the smart management method for rural photovoltaic, energy storage, and charging stations that incorporates point cloud data as described above. The smart management system for rural photovoltaic, energy storage, and charging stations that incorporates point cloud data includes:

[0179] The temporary site point cloud data acquisition module is used to acquire point cloud data of temporary deployment sites;

[0180] The point cloud site modeling and feature extraction module is used to perform point cloud site modeling based on point cloud data to obtain a point cloud site model; and to extract point cloud site features based on the point cloud site model to obtain point cloud site feature data.

[0181] The deployment optimization region calculation module is used to calculate the deployment optimization region based on the point cloud site feature data to obtain the deployment optimization region data.

[0182] The intelligent deployment simulation module is used to simulate intelligent deployment based on the data of the preferred deployment area, and obtain intelligent deployment data.

[0183] Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended application documents rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of the equivalents of the application documents be incorporated into the invention.

[0184] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. 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 the invention. Therefore, the present invention 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 of the invention herein.

Claims

1. A rural light storage and charging station intelligent management method combining point cloud data, characterized in that, The method comprises: S1, acquiring point cloud data of a temporary deployment site; S2, performing point cloud site modeling according to the point cloud data to obtain a point cloud site model; performing point cloud site feature extraction according to the point cloud site model to obtain point cloud site feature data; S3, performing deployment optimization area calculation according to the point cloud site feature data to obtain deployment optimization area data; S4, performing intelligent placement simulation according to the deployment optimization area data to obtain intelligent placement data; The point cloud site modeling comprises: performing scene type matching according to the point cloud data and a preset scene library model to generate pseudo historical site reference data; performing pseudo physical field boundary fitting on the point cloud data to obtain point cloud optimization data; performing ground surface form fitting according to the point cloud optimization data and the pseudo historical site reference data to generate a ground surface model; performing entropy filtering processing on the point cloud optimization data to obtain steady-state occlusion data; performing slope risk calculation on the ground surface model to obtain slope risk data; performing space accessibility processing according to the steady-state occlusion data and the slope risk data to obtain space accessibility data; performing fusion according to the space accessibility data and the ground surface model to obtain the point cloud site model.

2. The method of claim 1, wherein, S1 comprises: controlling a terminal laser device to perform low resampling pre-scanning to obtain initial point cloud data; performing ground surface bearing capacity preliminary estimation processing on the initial point cloud data to obtain bearing capacity data; performing key area stability extraction processing on the bearing capacity data to obtain key area stability data; controlling the terminal laser device to perform key area high-precision back scanning on the initial point cloud data according to the key area stability data to obtain the point cloud data.

3. The method of claim 1, wherein, The entropy filtering processing comprises: performing double-domain space entropy calculation on the point cloud optimization data to obtain double-domain space entropy data; performing candidate occlusion point preliminary marking on the point cloud optimization data according to the double-domain space entropy data to obtain candidate occlusion point marking data; performing entropy gradient field generation on the candidate occlusion point marking data to obtain entropy gradient field data; performing Markov correction modeling according to the entropy gradient field data to obtain an entropy gradient field model; performing steady-state occlusion segmentation according to the entropy gradient field model to obtain the steady-state occlusion data.

4. The method of claim 1, wherein, The slope risk calculation comprises: performing micro-mutation field extraction on the ground surface model to obtain micro-mutation field data; performing slope gradient trend graph generation on the micro-mutation field data to obtain slope gradient trend graph data; performing ground surface stress propagation calculation according to the slope gradient trend graph data to obtain boundary destabilization trend data; performing risk calculation according to the boundary destabilization trend data to obtain slope risk data.

5. The method of claim 4, wherein, The micro-mutation field extraction comprises: performing local neighborhood point density field calculation on the ground surface model to obtain neighborhood point density data; performing neighborhood density change rate calculation on the neighborhood point density data to obtain neighborhood density change rate data; performing sparse variability field generation according to the neighborhood density change rate data to obtain local sparse data; performing local neighborhood topology graph generation according to the ground surface model to obtain local neighborhood graph data; performing local self-homeomorphism defect calculation on the local neighborhood graph data to obtain local self-homeomorphism defect data; performing self-homeomorphism defect graph generation on the local self-homeomorphism defect data to obtain point cloud defect graph data; According to the local sparse data, the point cloud defect map data is masked and multi-head self-attention calculation is performed to obtain slope risk data.

6. The method of claim 1, wherein, The point cloud site feature extraction includes: According to the point cloud site model, a dense height point cloud projection is performed to obtain height map data. Convolution network feature extraction is performed on the height map data to obtain convolution network feature data. Feature map data is obtained by performing feature map saliency mapping and reorganization on the convolution network feature data. Feature entropy data is obtained by performing feature entropy modulation mapping on the convolution network feature data. According to the feature map data and the feature entropy data, a feature guide field is generated to obtain point cloud site feature data.

7. The method of claim 1, wherein S3 It includes: According to the point cloud site feature data, the site curvature anomaly area is removed to obtain site preliminary screening data. Topographic stability data is obtained by performing topographic stability weight calculation on the site preliminary screening data. According to the point cloud site feature data and the topographic stability data, a feature avoidance area field modeling is performed to obtain deployment optimization area data.

8. The method of claim 1, wherein, S4 includes: According to the deployment optimization area data, a deployment posture simulation is performed to obtain deployment posture data; Light shadow evolution simulation is performed on the deployment posture data to obtain light simulation data; According to the light simulation data, the deployment posture data is optimized to obtain deployment posture optimization data; Device coordination placement conflict detection is performed on the deployment posture optimization data to obtain intelligent placement data. 9.A rural light storage and charging station intelligent management system combined with point cloud data, characterized in that, The rural light storage and charging station intelligent management system combined with point cloud data for performing the rural light storage and charging station intelligent management method of claim 1 includes: A temporary site point cloud data acquisition module is used to acquire point cloud data of a temporary deployment site. A point cloud site modeling and feature extraction module is used to model a point cloud site according to the point cloud data to obtain a point cloud site model; and extract point cloud site features according to the point cloud site model to obtain point cloud site feature data. A deployment optimization area calculation module is used to calculate a deployment optimization area according to the point cloud site feature data to obtain deployment optimization area data. An intelligent placement simulation module is used to simulate intelligent placement according to the deployment optimization area data to obtain intelligent placement data.

Images