Power distribution network scene reconstruction method based on three-dimensional point cloud

Abstract

The invention discloses a three-dimensional point cloud-based power distribution network scene reconstruction method, and relates to the technical field of power distribution network scene modeling, and the method comprises the following steps: firstly, a laser radar carries out the data collection of a target region according to a preset scanning frequency, and guarantees the basic stability and consistency of data collection; according to the method, the key features of the regional complexity are extracted, and the pre-trained deep learning model is combined for intelligent evaluation, so that the laser radar can dynamically adjust the scanning strategy according to the real-time complexity evaluation index. For a common scene, a fixed scanning frequency is kept to save computing resources; for a complex scene, the laser pulse emission frequency is increased, the point cloud density is increased, and it is ensured that details are not lost. The adaptive adjustment optimizes the data acquisition efficiency and precision, avoids the generation of redundant data, and ensures the comprehensive reconstruction and accurate analysis of a power distribution network scene.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network scene modeling, and in particular to a distribution network scene reconstruction method based on three-dimensional point cloud. Background Art

[0002] "Distribution network scene reconstruction based on 3D point cloud" refers to the process of modeling and reconstructing the spatial scene of the distribution network using 3D point cloud data. Specifically, 3D point cloud data is obtained through LiDAR, photogrammetry or other sensors, and contains the precise spatial coordinate information of various objects in the distribution network environment (such as poles, wires, substation facilities, etc.). These data are usually discrete point sets and lack structured surface information. Surface reconstruction is to convert these discrete points into a continuous 3D surface model through mathematical algorithms and computer graphics technology, so as to obtain the complete 3D spatial structure of the distribution network. This process can help realize the visualization, planning, analysis and optimization of the distribution network, especially when conducting distribution network design, inspection, fault diagnosis or environmental simulation, it can provide accurate spatial data support and decision-making basis.

[0003] The prior art has the following deficiencies:

[0004] In the prior art, LiDAR usually uses a fixed scanning frequency to acquire distribution network scene data, which means that the number of times the LiDAR emits and receives laser pulses per second remains constant, thereby ensuring the stability of the scanning process and the consistency of data collection. However, when the distribution network scene is in a highly complex environment (such as underground distribution network facilities, crisscrossing lines, etc.), continuing to use a fixed scanning frequency may lead to serious consequences. Especially in complex environments, the distance between distribution network facilities is short, and cables and pipelines may be staggered and overlapped. Fixed scanning frequencies cannot provide sufficient point cloud density to capture all details, especially those small devices that are closely adjacent to other facilities (such as junction boxes, substations, underground cables, etc.). This low-density data will result in the loss of important detail information in the final three-dimensional point cloud, which will in turn affect subsequent scene reconstruction and analysis, and cannot achieve accurate detection, planning and maintenance of distribution network equipment.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a distribution network scene reconstruction method based on three-dimensional point cloud. By extracting the key features of regional complexity and combining it with a pre-trained deep learning model for intelligent evaluation, the laser radar can dynamically adjust the scanning strategy according to the real-time complexity evaluation index. For ordinary scenes, a fixed scanning frequency is maintained to save computing resources; for complex scenes, the number of laser pulse emissions is increased to increase the point cloud density and ensure that details are not lost. This adaptive adjustment optimizes data acquisition efficiency and accuracy, avoids the generation of redundant data, and ensures comprehensive reconstruction and accurate analysis of distribution network scenes to solve the problems in the above-mentioned background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a distribution network scene reconstruction method based on three-dimensional point cloud, comprising the following steps:

[0008] First, the LiDAR collects data from the target area at a preset scanning frequency to ensure the basic stability and consistency of data collection;

[0009] Preprocess the original point cloud data to improve data quality and prepare for subsequent feature extraction and analysis;

[0010] Extract key features reflecting regional complexity from preprocessed point cloud data, conduct in-depth analysis on the extracted key features within the detection window, and quantify regional complexity through the analyzed features;

[0011] The analyzed features are input into the pre-trained deep learning model, and the quantified complexity features are intelligently evaluated through the deep learning model;

[0012] Through the evaluation of complexity features by the deep learning model, the current scanning area is divided into two categories: "highly complex scenes" and "simple scenes";

[0013] For simple scenarios, the LiDAR continues to collect data at a preset scanning frequency to ensure the rational use of resources while maintaining the stability and reliability of data collection;

[0014] For highly complex scenes, the number of laser pulse emissions is dynamically increased to increase the point cloud density and ensure the precision and completeness of data collection.

[0015] Preferably, key features reflecting regional complexity are extracted from the preprocessed point cloud data, and the extracted features include the degree of geometric overlap between different facilities and the degree of fragmentation of power grid facilities in space. Within the detection window, the degree of geometric overlap between different facilities and the degree of fragmentation of power grid facilities in space are analyzed to generate geometric overlap reference values ​​and spatial fragmentation reference values, respectively. The spatial complexity and density of facility layout in the distribution network area are quantified by the geometric overlap reference values ​​and spatial fragmentation reference values.

[0016] Preferably, the specific steps of analyzing the geometric overlap degree between different facilities under the detection window to generate the geometric overlap reference value are as follows:

[0017] Extract geometric information of different facilities in the scene, and convert each facility into a three-dimensional geometric body through the point cloud data collected by the LiDAR. In this process, the spatial geometry algorithm is used to identify the overlapping areas between facilities. The calculation expression is as follows:

[0018] D(F i , F j )=∫ Ω ||P i -P j ||dΩ

[0019] , where D(F i , F j ) is facility F i and facilities j The spatial distance between them, Ω is the spatial range occupied by facilities F1 and F2 respectively, and F i and F j are the i-th facility and the j-th facility in the distribution network, P i Facility F i At any point on P j Facility F j Any point on ||P i -P j || is the Euclidean distance between two points, indicating facility F i and facilities j The straight-line distance between any two points on

[0020] After detecting the overlapping areas between facilities, the volume of the overlapping areas is calculated. The calculation expression is as follows;

[0021]

[0022] , where V overlap (F i , F j ) is the overlapping volume, indicating that the facility F iand facilities j The overlapping space volume between Facility F i The spatial range, Facility F j The spatial range, Facility F i and facilities j The intersection of spatial ranges;

[0023] To quantify the overlap between facilities, the overlapping volume is compared with the total volume of the facilities to generate a geometric overlap reference value, which is generated using the following formula:

[0024]

[0025] , where G overlap is the geometric overlap reference value, is the volume of the ith facility, is the volume of the jth facility, w i , j Facility F i and facilities j The weight factor between .

[0026] Preferably, the specific steps of analyzing the fragmentation degree of power grid facilities in space under the detection window to generate a reference value of spatial fragmentation degree are as follows:

[0027] First, the distribution of power grid facilities in space is extracted from the preprocessed point cloud data. By segmenting and calibrating the point cloud data, the boundaries and contours of different facilities are extracted, and the occupied volume of each facility in space is calculated. The calculation expression is as follows:

[0028]

[0029] , where V facility is the space occupied by the power grid facilities, A facility is the projected area of ​​the facility, V is the volume, and dV is the differential element of the volume;

[0030] After analyzing the geometric characteristics of the facilities, the relative distances between the facilities and their spatial connectivity are further calculated. By calculating the distances between the facilities and combining the connectivity of the spatial layout, the relationship between the facilities is quantified. The calculation expression is as follows:

[0031]

[0032] , where is the weighted sum of the distances between facilities, d ij is the distance between facility i and facility j, γ is the exponential coefficient;

[0033] Further analysis of the spatial distribution of facilities shows that in areas with high fragmentation, the facility layout is scattered, and the distribution density of equipment in different areas varies greatly. The distribution density function is introduced to quantify the distribution of facilities in space, thereby evaluating the aggregation and sparseness of facilities in space. The calculation expression is as follows:

[0034]

[0035] , where is the weighted sum of facility density, A fragmented is the fragmentation area, ρ density is the facility number density, α is the exponential coefficient, d is the area, and dA is the differential area within the integration region;

[0036] The space volume V occupied by the integrated power grid facilities facility , the weighted sum of distances between facilities and the weighted sum of facility density Generate a reference value of spatial fragmentation, and the generation formula is as follows:

[0037]

[0038] , where F fragmented is the reference value of spatial fragmentation, is the facility volume occupancy ratio, β1 is the weighted sum of the distances between facilities The weight coefficient, β2 is the weighted sum of facility density The weight coefficient, β3 is the facility volume occupancy ratio The weight coefficient of .

[0039] Preferably, the analyzed geometric overlap reference value and spatial fragmentation reference value are input into a pre-trained deep learning model, and a complexity evaluation index is generated by the deep learning model. The regional complexity is intelligently evaluated through the complexity evaluation index.

[0040] Preferably, the complexity evaluation index generated by intelligently evaluating the quantified complexity features through a pre-trained deep learning model is compared and analyzed with a pre-set complexity evaluation index reference threshold, and the current scanning area is divided, and the division steps are as follows:

[0041] If the complexity evaluation index is greater than a preset complexity evaluation index reference threshold, the current scanning area is divided into a highly complex scene;

[0042] If the complexity evaluation index is less than or equal to a preset complexity evaluation index reference threshold, the current scanning area is divided into a simple scene.

[0043] Preferably, after the current area is determined to be a complex area, the complexity evaluation index Complexity Assessment Dynamically adjust the number of laser pulse emissions, calculate the new scanning frequency by comparing it with the preset scanning frequency, and increase the number of laser pulse emissions accordingly. The calculation expression is as follows:

[0044]

[0045] , where Complexity Assessment is the complexity evaluation index, Complexity ref is the reference threshold of complexity evaluation index, is the preset scanning frequency, ω1 is the adjustment coefficient, is the adjusted scanning frequency;

[0046] According to the adjusted scanning frequency Further increase the number of laser radar pulse transmissions to obtain higher density point cloud data. The calculation expression is as follows:

[0047]

[0048] , where is the number of initial laser pulses emitted, ω2 is the adjustment coefficient, which is used to control the degree of increase in the number of pulses. is the adjusted number of launches;

[0049] After increasing the scanning frequency and the number of laser pulses emitted, the LiDAR begins to collect point cloud data. During the collection process, adjustments should be made based on real-time feedback to ensure the integrity and accuracy of the data. The expression is as follows:

[0050]

[0051] , where Real-Time Feeback density is the point cloud density of real-time feedback, Target Density is the target point cloud density, ω3 is the real-time feedback adjustment coefficient, is the final adjusted number of laser pulse emissions.

[0052] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0053] The present invention extracts key features that reflect the complexity of the area (such as geometric overlap reference values ​​and spatial fragmentation reference values) and combines them with a pre-trained deep learning model for intelligent evaluation. This enables the lidar to dynamically adjust the scanning strategy based on the complexity evaluation index of real-time analysis. For ordinary scenes, a fixed scanning frequency is maintained to avoid excessive computational burden, while for complex scenes, the number of laser pulse emissions is increased to improve scanning accuracy and point cloud density. This adaptive adjustment not only improves data acquisition efficiency, but also ensures that details of high-density and complex areas can be fully captured, avoiding the problem of detail loss due to fixed frequency.

[0054] The present invention uses a deep learning model to evaluate the complexity of the area, so that the scanning area can be intelligently divided into "highly complex scenes" and "simple scenes". This division not only helps to adjust the working mode of the lidar in a targeted manner, but also can choose whether to improve the scanning accuracy according to actual needs, avoiding unnecessary high-frequency scanning in simple scenes, thereby effectively saving computing resources and data storage. This intelligent evaluation and dynamic adjustment process based on deep learning significantly improves scanning efficiency and accuracy, ensures a comprehensive and detailed reconstruction of the distribution network scene, and reduces the generation of redundant data. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0056] Figure 1 The present invention is a flow chart of a method for reconstructing a distribution network scene based on three-dimensional point cloud. DETAILED DESCRIPTION

[0057] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0058] The present invention provides Figure 1 The distribution network scene reconstruction method based on three-dimensional point cloud shown includes the following steps:

[0059] First, the LiDAR collects data from the target area at a preset scanning frequency to ensure the basic stability and consistency of data collection;

[0060] The preset scanning frequency refers to the number of laser pulses emitted and received by the laser radar per second when the laser radar collects data. This number determines the scanning density and the precision of data collection per unit time of the laser radar during the scanning process. Under the setting of a fixed scanning frequency, the number of laser pulse emissions is a fixed parameter to ensure that the laser radar can continuously and stably collect data from the scanning area within a specified time. Through this preset number, it can be ensured that the radar equipment can provide sufficient coverage of the target area in each time period, thereby providing a basis for subsequent data processing and analysis.

[0061] The LiDAR equipment collects comprehensive data on the entire scanning area of ​​the distribution network according to the predetermined scanning frequency and pulse emission number. The purpose of this stage is to obtain basic 3D point cloud data to ensure the stability of the scanning process and the consistency of the coverage. The preset number of laser pulse emissions needs to be reasonably set according to the overall complexity of the distribution network and the expected data density to ensure that sufficient basic data can be obtained in the initial stage, providing a reliable data source for subsequent preprocessing and feature extraction.

[0062] Preprocess the original point cloud data to improve data quality and prepare for subsequent feature extraction and analysis;

[0063] The preprocessing stage includes multiple sub-steps, such as noise filtering, point cloud downsampling, point cloud registration, and coordinate correction. First, the noise filtering algorithm is used to remove random noise and outliers in the point cloud data to ensure the purity of the data. Then, the downsampling technology is used to reduce the redundancy of the point cloud data and reduce the computational complexity while retaining key geometric features. Next, point cloud registration and coordinate correction are performed to ensure that the data under different scanning angles can be accurately aligned to form a unified three-dimensional spatial coordinate system. These preprocessing steps can significantly improve the quality and consistency of the data, providing a reliable data foundation for subsequent feature extraction and complexity analysis.

[0064] Extract key features reflecting regional complexity from preprocessed point cloud data, conduct in-depth analysis on the extracted key features within the detection window, and quantify regional complexity through the analyzed features;

[0065] Key features reflecting regional complexity are extracted from the preprocessed point cloud data. The extracted features include the degree of geometric overlap between different facilities and the degree of fragmentation of power grid facilities in space. Within the detection window, the degree of geometric overlap between different facilities and the degree of fragmentation of power grid facilities in space are analyzed to generate geometric overlap reference values ​​and spatial fragmentation reference values, respectively. The spatial complexity and density of facility layout in the distribution network area are quantified by the geometric overlap reference values ​​and spatial fragmentation reference values.

[0066] The large number of overlaps and intersections between different facilities does indicate that the current area is in a highly complex state. First, the overlap and intersection phenomenon means that the facilities in the distribution network (such as cables, pipelines, substation equipment, etc.) are close to each other or intertwined in a limited space. This dense layout usually occurs in environments with narrow space or a large number of equipment, especially in underground distribution networks and densely populated urban areas. The overlap of facilities not only increases the complexity of space use, but also makes it difficult for the lidar to accurately distinguish the shape and position of each independent device during scanning, resulting in overlapping or distortion of point cloud data. Secondly, this overlap and intersection will cause multiple reflections or occlusions in the point cloud data, affecting the integrity and accuracy of the data. Due to the extremely short distance between the facilities, the lidar cannot obtain sufficient sampling density on every detail, especially those small devices (such as junction boxes, sensors, etc.) may be blocked or missed. Finally, the highly complex facility layout not only increases the complexity of the scanned data, but also increases the difficulty of subsequent data processing and 3D reconstruction, especially when the geometric structure of these devices and pipelines is complex and the shape is irregular. Therefore, the overlap and intersection of facilities are important indicators of regional complexity, which can effectively reflect the challenges of spatial layout and data collection in the region, and require higher scanning accuracy and density to cope with this complex environment.

[0067] The specific steps for analyzing the geometric overlap between different facilities under the detection window to generate the geometric overlap reference value are as follows:

[0068] Extract geometric information of different facilities in the scene, including modeling the spatial position, shape, size, etc. of the facilities. Through the point cloud data collected by the lidar, each facility is converted into a three-dimensional geometric body (such as a point set, grid, etc.). In this process, the spatial geometry algorithm is used to identify the overlapping areas between facilities. The calculation expression is as follows:

[0069] D(F i , F j )=∫ Ω ||P i -P j ||dΩ

[0070] , where D(F i , F j ) is facility F i and facilities j The spatial distance between them indicates their relative positions, Ω is the spatial range occupied by facilities F1 and F2, indicating the volume or surface area of ​​the two facilities, and F i and F j are the i-th facility and the j-th facility in the distribution network, P i Facility F i At any point on Pj Facility F j Any point on ||P i -P j || is the Euclidean distance between two points, indicating facility F i and facilities j The straight-line distance between any two points on the network can be calculated. By calculating the distance between these points, the geometric distance between facilities can be obtained and their relative spatial relationship can be evaluated.

[0071] Euclidean distance||P i -P j The role of || is to quantify the spatial distance between two facilities, that is, facility F i and facilities j Any two points P i and P j The straight-line distance between them. By calculating the Euclidean distance between these points, the spatial proximity between facilities can be measured. Euclidean distance is the most commonly used distance metric, which provides a simple and intuitive way to describe the relative position of two points in three-dimensional space. In the distribution network scenario, the calculation of Euclidean distance can help identify spatial overlap or proximity areas between different facilities, and then evaluate the complexity of facility layout. When the distance between two facilities is small, it indicates that they may overlap or intersect, which means that the area is geometrically complex. Conversely, a larger distance indicates that the distribution between facilities is sparse and the complexity of the area is low. Therefore, Euclidean distance is an important basis for calculating geometric overlap and spatial fragmentation, and it provides the necessary data support for the subsequent quantification of complexity.

[0072] After detecting the overlapping area between the facilities, the volume of the overlapping area is calculated. The volume of the overlapping area of ​​the facilities is an important factor in calculating the geometric overlap index. The calculation expression is as follows;

[0073]

[0074] , where V overlap (F i , F j ) is the overlapping volume, indicating that the facility F i and facilities j The overlapping space volume between Facility F i The spatial range, Facility F j The spatial range, Facility F i and facilities j The intersection of spatial ranges;

[0075] To quantify the overlap between facilities, the overlapping volume is compared with the total volume of the facilities to generate a geometric overlap reference value, which is generated using the following formula:

[0076]

[0077] , where G overlap is the geometric overlap reference value, is the volume of the ith facility, is the volume of the jth facility, w i , j Facility F i and facilities j The weight factor between is used to adjust the relative importance of overlap between facility pairs.

[0078] It can be seen from the geometric overlap reference value that the larger the geometric overlap reference value generated after analyzing the geometric overlap between different facilities under the detection window, the more complex the current area is. The geometric overlap reference value quantifies the degree of intersection and overlap between different facilities (such as cables, pipelines, substation equipment, etc.). The higher the overlap, the tighter the layout of these facilities in space, the closer the distance between the equipment, and even the intersection or overlap. This layout usually occurs in areas with narrow space and dense facilities, especially in underground distribution networks and complex urban environments, which increases the difficulty of laser radar data collection. In such an environment, the laser radar may not be able to accurately distinguish each device, resulting in overlapping, occluding or missing data, affecting subsequent three-dimensional reconstruction and analysis. Therefore, a larger performance value of the geometric overlap reference value indicates the complexity of the facility layout, indicating that the area has a high complexity in space utilization and data collection, and requires a more accurate data collection strategy. On the contrary, if the geometric overlap reference value is small, it means that the space between the facilities is relatively loose, there is less overlap, the complexity of the area is low, and the laser radar can easily capture the details of each independent device, making the data collection and analysis of the area simpler.

[0079] The significant increase in the fragmentation of power grid facilities in space usually indicates that the current area is in a highly complex state. This is because areas with high fragmentation mean that the layout of distribution network facilities is highly dispersed, scattered or irregular, the relative positions between devices are no longer continuous, and may consist of multiple small and independent components. This fragmentation phenomenon usually occurs in environments with limited space or dense facilities, such as underground distribution networks or areas where pipelines and cables are intertwined. Due to the irregular and scattered layout of facilities, the gaps and layout changes between devices are often large, which makes the collection and processing of point cloud data more complicated and difficult. Fragmented equipment is often difficult to be accurately captured by a unified scanning solution, which may lead to sparseness, overlap or missing data, and thus affect subsequent 3D reconstruction and analysis. In this case, traditional LiDAR acquisition methods may not provide sufficient accuracy and details to fully reflect the real situation in the area. Therefore, the significant increase in the fragmentation of power grid facilities not only reflects the complexity of equipment layout, but also means that the current area has higher requirements for data collection and analysis, usually requiring higher density scanning, higher frequency point cloud collection, and more complex post-processing algorithms to ensure accuracy and comprehensiveness.

[0080] The specific steps for analyzing the fragmentation degree of power grid facilities in space under the detection window to generate a reference value of spatial fragmentation degree are as follows:

[0081] First, the distribution of power grid facilities in space is extracted from the preprocessed point cloud data. The goal of this step is to identify the spatial relationship between facilities, including the layout of facilities, relative positions, distances, and connectivity between facilities. By segmenting and calibrating the point cloud data, the boundaries and contours of different facilities are extracted, and the occupied volume of each facility in space (i.e., the spatial range of each facility) is calculated. The calculation expression is as follows:

[0082]

[0083] , where V facility is the space occupied by the power grid facilities, A facility is the projected area of ​​the facility, V is the volume, and dV is the differential element of the volume, which represents a very small unit of volume;

[0084] By integrating the spatial occupancy of each point in the area, the geometric boundary of the facility can be effectively extracted.

[0085] After analyzing the geometric characteristics of the facilities, the relative distance between the facilities and their spatial connectivity are further calculated. Especially in fragmented areas, the spatial distribution of equipment is usually scattered, and there may be no direct connection between the facilities. By calculating the distance between the facilities and combining the connectivity of the spatial layout, the relationship between the facilities is quantified. The calculation expression is as follows:

[0086]

[0087] , where is the weighted sum of the distances between facilities, d ij is the distance between facility i and facility j, which is used to quantify the physical distance between two facilities. For fragmented areas, the distance between facilities is often large, indicating that the facilities are not densely distributed or independent of each other. γ is the exponential coefficient;

[0088] γ is an exponential coefficient used to weight the impact of the distance between facilities and determine the contribution of distance to the final calculation result. Specifically, the role of γ is to adjust the different effects of long-distance facilities and short-distance facilities on the fragmentation index.

[0089] Further analysis of the spatial distribution of facilities shows that in areas with high fragmentation, the facility layout is scattered, and the distribution density of equipment in different areas varies greatly. The distribution density function is introduced to quantify the distribution of facilities in space, thereby evaluating the aggregation and sparseness of facilities in space. The calculation expression is as follows:

[0090]

[0091] , where is the weighted sum of facility density, A fragmented is the fragmentation area, ρ density is the facility number density, α is the exponential coefficient, d is the area, and dA is the differential area within the integration region;

[0092] α is an exponential coefficient, which is used to weight the impact of grid facility density on fragmentation.

[0093] This step reflects the sparseness of the distribution of facilities in the fragmented region, with lower density indicating a more fragmented region.

[0094] The space volume V occupied by the integrated power grid facilities facility , the weighted sum of distances between facilities and the weighted sum of facility density Generate a reference value of spatial fragmentation, and the generation formula is as follows:

[0095]

[0096] , where F fragmented is the reference value of spatial fragmentation, is the facility volume occupancy ratio, β1 is the weighted sum of the distances between facilities The weight coefficient controls the contribution of the relative distance between facilities to the overall fragmentation, and β2 is the weighted sum of the facility density. The weight coefficient controls the contribution of facility distribution density to fragmentation, and β3 is the facility volume occupancy ratio The weight coefficient controls the relationship between the volume occupied by the facility in the space and the degree of fragmentation.

[0097] It can be seen from the spatial fragmentation reference value that the larger the spatial fragmentation reference value generated after analyzing the fragmentation degree of power grid facilities in space under the detection window, the more complex the current area is. The larger the spatial fragmentation reference value, the more scattered and discontinuous the layout of power grid facilities in the area is, and the relative positions of the equipment vary greatly. The distribution of facilities in space may be more scattered and irregular due to the staggered arrangement of pipelines, cables or other facilities. This fragmented layout makes the identification and data collection of equipment more difficult, and the point cloud data may appear sparse or incomplete, affecting the accuracy of three-dimensional reconstruction and subsequent analysis. Therefore, a larger spatial fragmentation reference indicates that the area is more complex in space and has stricter requirements for data collection. On the contrary, if the spatial fragmentation reference value is small, it means that the layout of facilities is more concentrated or regular, the overall environment is relatively simple, and data collection and analysis are also easier.

[0098] The analyzed features are input into the pre-trained deep learning model, and the quantified complexity features are intelligently evaluated through the deep learning model;

[0099] The analyzed geometric overlap reference value and spatial fragmentation reference value are input into a pre-trained deep learning model, and a complexity assessment index is generated through the deep learning model. The regional complexity is intelligently evaluated through the complexity assessment index.

[0100] A pre-learned deep learning model is a model that is trained using machine learning techniques (especially deep learning techniques) with a large amount of labeled historical data during the data preparation and training phase, so that it has the ability to automatically extract features from input data and perform intelligent reasoning. These deep learning models are usually designed and optimized based on neural network architectures (such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), or transformer models (Transformers). Through repeated learning of a large amount of training data, deep learning models can capture the potential patterns and complex relationships in the data, and ultimately output evaluation results that reflect the complexity of the region.

[0101] Specifically, the pre-learned deep learning model is a training process that associates various input parameters (such as geometric overlap reference values, spatial fragmentation reference values, etc.) with known annotated data (such as the actual complexity labels of each area). The model optimizes its output by continuously adjusting internal parameters (such as weights and biases) so that the model can accurately predict and evaluate the complexity of a given area. This process usually relies on large-scale data sets covering different types of regional complexity and equipment layouts, and is trained through supervised learning methods. When the model is fully trained, it has the ability to intelligently evaluate regional complexity from new input data.

[0102] In practical applications, the pre-learned deep learning model can generate a complexity evaluation index through intelligent reasoning based on the input geometric overlap reference value and spatial fragmentation reference value, combined with the complexity pattern in the historical data. This index can be used to quantify the complexity of the current area and provide a basis for subsequent decision-making. For example, for highly complex areas, the deep learning model may identify these areas with high equipment crossover, density and dispersion characteristics through learned features, thereby deriving a higher complexity evaluation index. This intelligent evaluation not only improves the efficiency of data processing, but also reduces the deviation and inconsistency of manual judgment through automation, greatly improving the accuracy and reliability of complexity evaluation in distribution network scenarios.

[0103] This pre-trained deep learning model has shown strong generalization ability and flexibility in practical applications, and can provide highly accurate prediction results under different environmental and data conditions. As new data is continuously introduced, the deep learning model can also be optimized and updated through incremental learning, so that the evaluation process gradually adapts to new environmental changes and challenges, thereby maintaining a high degree of real-time and adaptability.

[0104] The deep learning model is not limited here and can realize the geometric overlap reference value G overlap and the spatial fragmentation reference value F fragmented Perform comprehensive analysis to generate complexity evaluation index Complexity Assessment In order to realize the technical solution of the present invention, the present invention provides a specific implementation method;

[0105] Complexity Evaluation Index Complexity Assessment The generation formula is as follows:

[0106]

[0107] , where k1 and k2 are the overlap reference values ​​G respectively. overlap and the spatial fragmentation reference value Ffragmented The preset proportional coefficient, and k1 and k2 are both greater than 0.

[0108] It can be seen from the complexity evaluation index that the larger the geometric overlap reference value generated after analyzing the geometric overlap between different facilities under the detection window, the larger the spatial fragmentation reference value generated after analyzing the fragmentation degree of power grid facilities in space under the detection window, and the larger the complexity evaluation index generated by intelligently evaluating the quantified complexity features through the pre-trained deep learning model, it indicates that the current area is a highly complex area. Otherwise, it indicates that the complexity of the current area is low and it is a simple area.

[0109] Through the evaluation of complexity features by the deep learning model, the current scanning area is divided into two categories: "highly complex scenes" and "simple scenes";

[0110] The complexity evaluation index generated by intelligently evaluating the quantified complexity features through the pre-trained deep learning model is compared and analyzed with the pre-set complexity evaluation index reference threshold to divide the current scanning area. The division steps are as follows:

[0111] If the complexity evaluation index is greater than a preset complexity evaluation index reference threshold, the current scanning area is divided into a highly complex scene;

[0112] If the complexity evaluation index is less than or equal to a preset complexity evaluation index reference threshold, the current scanning area is divided into a simple scene.

[0113] Highly complex scenarios refer to areas with dense and complex distribution network facilities, which require higher point cloud density to capture details; simple scenarios refer to areas with relatively scattered facilities and a relatively simple environment, which can continue to use the preset fixed scanning frequency.

[0114] For simple scenarios, the LiDAR continues to collect data at a preset scanning frequency to ensure the rational use of resources while maintaining the stability and reliability of data collection;

[0115] The LiDAR collects data from the target area at a preset scanning frequency to ensure the basic stability and consistency of the data collection process. A fixed scanning frequency ensures that the number of laser pulses emitted and received per second remains constant, thereby maintaining consistent measurement accuracy and scanning density throughout the data collection process. This approach helps avoid fluctuations or inconsistencies in the data collection process, lays a solid foundation for subsequent point cloud processing, feature extraction, and scene reconstruction, and ensures that the final result has high reliability and accuracy.

[0116] For highly complex scenes, dynamically increase the number of laser pulse emissions and the point cloud density to ensure the precision and integrity of data collection;

[0117] For highly complex scenes, the specific steps to dynamically increase the number of laser pulses, increase the point cloud density, and ensure the precision and integrity of data collection are as follows:

[0118] After the current area is judged as a complex area, the complexity evaluation index Complexity Assessment Dynamically adjust the number of laser pulse emissions, calculate the new scanning frequency by comparing it with the preset scanning frequency, and increase the number of laser pulse emissions accordingly. The calculation expression is as follows:

[0119]

[0120] , where Complexity Assessment is the complexity evaluation index, Complexity ref is the reference threshold of complexity evaluation index, is the preset scanning frequency, ω1 is the adjustment coefficient, which controls the influence of the complexity evaluation index on the scanning frequency. is the adjusted scanning frequency;

[0121] According to the adjusted scanning frequency Further increase the number of laser radar pulse transmissions to obtain higher density point cloud data. In complex areas, especially in environments with dense underground facilities or narrow spaces, increasing the number of pulse transmissions can effectively compensate for the data loss caused by overlapping, occlusion and fragmentation of facilities in the area. The calculation expression is as follows:

[0122]

[0123] , where is the number of initial laser pulses emitted, ω2 is the adjustment coefficient, which is used to control the degree of increase in the number of pulses. It is the adjusted number of transmissions, which changes according to the scanning frequency adjustment;

[0124] This step increases the density of the point cloud by increasing the number of pulse transmissions, ensuring that every detail in complex scenes can be accurately captured.

[0125] After increasing the scanning frequency and the number of laser pulses, the LiDAR begins to collect point cloud data. During the collection process, adjustments should be made based on real-time feedback to ensure the integrity and accuracy of the data. The feedback can include parameters such as real-time point cloud density and facility overlap during the scanning process, and further guide whether the scanning frequency or pulse number needs to be adjusted again. The expression is as follows:

[0126]

[0127] , where Real-Time Feeback density is the point cloud density of real-time feedback, Target Density is the target point cloud density, ω3 is the real-time feedback adjustment coefficient, which is used to control the influence of real-time feedback on the number of pulse emissions and adjust the sensitivity of the laser radar to the point cloud density feedback. is the final adjusted number of laser pulse emissions.

[0128] Real-time feedback is the feedback information obtained during the data collection process, which is used to adjust the number of pulse transmissions in real time. Feedback regulation ensures that the lidar can dynamically respond to environmental changes and adjust scanning parameters in time to avoid incomplete or redundant data.

[0129] In highly complex scenarios, the main purpose of dynamically increasing the number of laser pulse emissions is to increase the point cloud density during the lidar scanning process, thereby ensuring the precision and integrity of data collection. In complex environments, such as underground distribution network facilities, crisscrossing lines, or areas with highly dense equipment layouts, ordinary fixed scanning frequencies often cannot provide a high enough point cloud density to capture all details. In complex scenarios, the distance between facilities is short, and there may be a large number of intersecting, overlapping, or fragmented facility layouts. The fixed-frequency laser pulse emission is not enough to ensure that every detail can be accurately scanned and recorded.

[0130] Dynamically increasing the number of laser pulse emissions can solve this problem by increasing the scanning frequency. The complexity of the current area is determined based on the complexity evaluation index of the area (such as the geometric overlap index and the spatial fragmentation index), and the scanning frequency is adjusted accordingly. Specifically, when the complexity evaluation index exceeds the preset threshold, the frequency of laser pulse emission is dynamically increased through an algorithm to increase the density of point cloud acquisition, so that the details of each device and facility in the area can be captured more accurately during scanning. This not only helps to improve the accuracy and completeness of the data, but also avoids information loss in data sparse or occluded areas.

[0131] The core role of this step is to ensure that in complex scenes, the LiDAR can obtain data comprehensively and finely, avoid the loss or misjudgment of facilities due to insufficient sampling density, and provide high-quality data support for subsequent 3D reconstruction, analysis, detection and maintenance. At the same time, this dynamic adjustment process helps to optimize data collection efficiency, avoid unnecessary overscanning and data redundancy, reduce computing burden, and improve the work efficiency of the entire system.

[0132] The present invention extracts key features that reflect the complexity of the area (such as geometric overlap reference values ​​and spatial fragmentation reference values) and combines them with a pre-trained deep learning model for intelligent evaluation. This enables the lidar to dynamically adjust the scanning strategy based on the complexity evaluation index of real-time analysis. For ordinary scenes, a fixed scanning frequency is maintained to avoid excessive computational burden, while for complex scenes, the number of laser pulse emissions is increased to improve scanning accuracy and point cloud density. This adaptive adjustment not only improves data acquisition efficiency, but also ensures that details of high-density and complex areas can be fully captured, avoiding the problem of detail loss due to fixed frequency.

[0133] The present invention uses a deep learning model to evaluate the complexity of the area, so that the scanning area can be intelligently divided into "highly complex scenes" and "simple scenes". This division not only helps to adjust the working mode of the lidar in a targeted manner, but also can choose whether to improve the scanning accuracy according to actual needs, avoiding unnecessary high-frequency scanning in simple scenes, thereby effectively saving computing resources and data storage. This intelligent evaluation and dynamic adjustment process based on deep learning significantly improves scanning efficiency and accuracy, ensures a comprehensive and detailed reconstruction of the distribution network scene, and reduces the generation of redundant data.

[0134] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0135] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0136] It should be noted that, in this article, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0137] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0138] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0139] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0140] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0141] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0142] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0143] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A distribution network scene reconstruction method based on three-dimensional point cloud, characterized in that: The following steps are involved: First, the LiDAR collects data from the target area at a preset scanning frequency to ensure the basic stability and consistency of data collection; Preprocess the original point cloud data to improve data quality and prepare for subsequent feature extraction and analysis; Extract key features reflecting regional complexity from preprocessed point cloud data, conduct in-depth analysis on the extracted key features within the detection window, and quantify regional complexity through the analyzed features; The analyzed features are input into the pre-trained deep learning model, and the quantified complexity features are intelligently evaluated through the deep learning model; Through the evaluation of complexity features by the deep learning model, the current scanning area is divided into two categories: "highly complex scene" and "simple scene"; For simple scenarios, the LiDAR continues to collect data at a preset scanning frequency to ensure the rational use of resources while maintaining the stability and reliability of data collection; For highly complex scenes, the number of laser pulse emissions is dynamically increased to increase the point cloud density and ensure the precision and completeness of data collection.

2. The method for reconstructing a distribution network scene based on three-dimensional point cloud according to claim 1, characterized in that: Key features reflecting regional complexity are extracted from the preprocessed point cloud data. The extracted features include the degree of geometric overlap between different facilities and the degree of fragmentation of power grid facilities in space. Within the detection window, the degree of geometric overlap between different facilities and the degree of fragmentation of power grid facilities in space are analyzed to generate geometric overlap reference values ​​and spatial fragmentation reference values, respectively. The spatial complexity and density of facility layout in the distribution network area are quantified by the geometric overlap reference values ​​and spatial fragmentation reference values.

3. The method for reconstructing a distribution network scene based on three-dimensional point cloud according to claim 2, characterized in that: The specific steps for analyzing the geometric overlap between different facilities under the detection window to generate the geometric overlap reference value are as follows: Extract geometric information of different facilities in the scene, and convert each facility into a three-dimensional geometric body through the point cloud data collected by the LiDAR. In this process, the spatial geometry algorithm is used to identify the overlapping areas between facilities. The calculation expression is as follows: D(F i ,F j )=∫ Ω ||P i -P j ||dΩ, In the formula, D(F i , F j ) is facility F i and facilities j The spatial distance between them, Ω is the spatial range occupied by facilities F1 and F2 respectively, and F i and F j are the i-th facility and the j-th facility in the distribution network, P i Facility F i At any point on P j Facility F j Any point on ||P i -P j || is the Euclidean distance between two points, indicating facility F i and facilities j The straight-line distance between any two points on After detecting the overlapping areas between facilities, the volume of the overlapping areas is calculated. The calculation expression is as follows; ， Where V overlap (F i , F j ) is the overlapping volume, indicating that the facility F i and facilities j The overlapping space volume between Facility F i The spatial range, Facility F j The spatial range, Facility F i and facilities j The intersection of spatial ranges; To quantify the overlap between facilities, the overlapping volume is compared with the total volume of the facilities to generate a geometric overlap reference value, which is generated using the following formula: ， In the formula, G overlap is the geometric overlap reference value, is the volume of the ith facility, is the volume of the jth facility, w i，j Facility F i and facilities j The weight factor between .

4. The method for reconstructing a distribution network scene based on three-dimensional point cloud according to claim 2, characterized in that: The specific steps for analyzing the fragmentation degree of power grid facilities in space under the detection window to generate a reference value of spatial fragmentation degree are as follows: First, the distribution of power grid facilities in space is extracted from the preprocessed point cloud data. By segmenting and calibrating the point cloud data, the boundaries and contours of different facilities are extracted, and the occupied volume of each facility in space is calculated. The calculation expression is as follows: ， Where V facility is the space occupied by the power grid facilities, A facility is the projected area of ​​the facility, V is the volume, and dV is the differential element of the volume; After analyzing the geometric characteristics of the facilities, the relative distances between the facilities and their spatial connectivity are further calculated. By calculating the distances between the facilities and combining the connectivity of the spatial layout, the relationship between the facilities is quantified. The calculation expression is as follows: ， In the formula, is the weighted sum of the distances between facilities, d ij is the distance between facility i and facility j, γ is the exponential coefficient; Further analysis of the spatial distribution of facilities shows that in areas with high fragmentation, the facility layout is scattered, and the distribution density of equipment in different areas varies greatly. The distribution density function is introduced to quantify the distribution of facilities in space, thereby evaluating the aggregation and sparseness of facilities in space. The calculation expression is as follows: ， In the formula, is the weighted sum of facility density, A fragmented is the fragmentation area, ρ density is the facility number density, α is the exponential coefficient, d is the area, and dA is the differential area within the integration region; The space volume V occupied by the integrated power grid facilities facility , the weighted sum of distances between facilities and the weighted sum of facility density Generate a reference value of spatial fragmentation, and the generation formula is as follows: ， In the formula, F fragmented is the reference value of spatial fragmentation, is the facility volume occupancy ratio, β1 is the weighted sum of the distances between facilities The weight coefficient, β2 is the weighted sum of facility density The weight coefficient, β3 is the facility volume occupancy ratio The weight coefficient of .

5. The method for reconstructing a distribution network scene based on three-dimensional point cloud according to claim 2, characterized in that: The analyzed geometric overlap reference value and spatial fragmentation reference value are input into a pre-trained deep learning model, and a complexity assessment index is generated through the deep learning model. The regional complexity is intelligently evaluated through the complexity assessment index.

6. The method for reconstructing a distribution network scene based on three-dimensional point cloud according to claim 5, characterized in that: The complexity evaluation index generated by intelligently evaluating the quantified complexity features through the pre-trained deep learning model is compared and analyzed with the pre-set complexity evaluation index reference threshold to divide the current scanning area. The division steps are as follows: If the complexity evaluation index is greater than a preset complexity evaluation index reference threshold, the current scanning area is divided into a highly complex scene; If the complexity evaluation index is less than or equal to a preset complexity evaluation index reference threshold, the current scanning area is divided into a simple scene.

7. The method for reconstructing a distribution network scene based on three-dimensional point cloud according to claim 6, characterized in that: After the current area is judged as a complex area, the complexity evaluation index Complexity Assessment Dynamically adjust the number of laser pulse emissions, calculate the new scanning frequency by comparing it with the preset scanning frequency, and increase the number of laser pulse emissions accordingly. The calculation expression is as follows: ， In the formula, Complexity Assessment is the complexity evaluation index, Complexity ref is the reference threshold of complexity evaluation index, is the preset scanning frequency, ω1 is the adjustment coefficient, Adjusted is the adjusted scanning frequency; According to the adjusted scanning frequency Adjusted Further increase the number of laser radar pulse transmissions to obtain higher density point cloud data. The calculation expression is as follows: ， In the formula, is the number of initial laser pulses emitted, ω2 is the adjustment coefficient, which is used to control the degree of increase in the number of pulses, Pulse is the adjusted number of launches; After increasing the scanning frequency and the number of laser pulses emitted, the LiDAR begins to collect point cloud data. During the collection process, adjustments should be made based on real-time feedback to ensure the integrity and accuracy of the data. The expression is as follows: ， In the formula, Real-Time Feeback density is the point cloud density of real-time feedback, Target Density is the target point cloud density, ω3 is the real-time feedback adjustment coefficient, Pulse is the final adjusted number of laser pulse emissions.

Images