A method for automatic segmentation of 3D point cloud of power distribution network

Abstract

The present application relates to the technical field of point cloud segmentation, and provides a method for automatic segmentation of 3D point clouds of distribution networks. The method includes: interactively obtaining design information and dividing the local distribution network; obtaining the distribution network operation and maintenance platform; configuring a ground laser radar for the local distribution network, and randomly activating edge point cloud segmentation nodes; scheduling the ground laser radar to collect local laser point cloud data; the edge point cloud segmentation node receives the local laser point cloud data to perform segmentation, performs risk analysis based on the segmentation results, and obtains local distribution risks; the distribution network operation and maintenance platform performs risk assessment on local distribution risks, and performs global risk relief. The present application solves the technical problem of insufficient local feature extraction accuracy of PointNet, realizes the improvement of the efficiency and accuracy of 3D point cloud segmentation of the distribution network, and performs an external fault severity level analysis based on the power transmission relationship and external fault conditions in the distribution network, thereby improving the accuracy of external fault detection.

Description

Technical Field

[0001] The present application relates to the field of computer vision technology, specifically to the field of point cloud segmentation technology, and in particular to a method for automatic segmentation of 3D point clouds of power distribution networks. Background Art

[0002] With the rapid development of the power industry and the acceleration of urbanization, the distribution network, as an important part of the power system, undertakes the key task of directly supplying power to the majority of users. However, the distribution network has the characteristics of multiple voltage levels, complex network structure, diverse equipment types, many and wide operating points, and relatively poor safety environment, which brings great challenges to the operation and maintenance of the distribution network. Especially in recent years, with the development of modern technical means such as drones and digital twins, research on digital survey and three-dimensional visualization technology of distribution network non-stop operation has been carried out. However, when applying modern technology to the field of distribution network point cloud segmentation, there are many technical challenges. Due to the deep integration of the distribution network with the surrounding environment, the collected distribution network point cloud data is relatively fuzzy, and it is difficult to effectively identify and extract it through traditional point cloud segmentation algorithms. In addition, most of the existing technologies focus on the point cloud segmentation of large-scale and large target objects with obvious regional feature divisions. There is a lack of effective solutions for point cloud data with small ranges and fuzzy feature divisions such as distribution networks. Summary of the invention

[0003] This application provides a distribution network 3D point cloud automatic segmentation method, aiming to solve the technical problem of insufficient accuracy of local feature extraction by PointNet.

[0004] In view of the above problems, the present application provides a method for automatic segmentation of 3D point cloud of power distribution network.

[0005] The present application provides a method for automatic segmentation of 3D point clouds of a distribution network, the method comprising: interactively obtaining design information of a target distribution network, and dividing the target distribution network into M local distribution networks based on the design information, wherein M is a positive integer; obtaining a distribution network operation and maintenance platform, wherein the distribution network operation and maintenance platform is communicatively connected to a plurality of edge point cloud segmentation nodes; configuring M ground laser radars for the M local distribution networks, randomly activating M edge point cloud segmentation nodes at the plurality of edge point cloud segmentation nodes according to the M ground laser radars, and dividing the M ground The laser radar is communicatively connected to the M edge point cloud segmentation nodes; the M ground laser radars are dispatched to collect M local laser point cloud data of the M local power distribution networks; the M edge point cloud segmentation nodes receive the M local laser point cloud data to perform 3D point cloud automatic segmentation, and perform risk analysis based on the segmentation results to obtain M groups of local power distribution risks, wherein each group of local power distribution risks includes exogenous risks and endogenous risks; the distribution network operation and maintenance platform performs risk assessment on the M groups of local power distribution risks, and performs global risk removal of the target distribution network based on the assessment results.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The above-mentioned method for automatic segmentation of 3D point cloud of distribution network first understands the design information of the target distribution network, and divides the entire distribution network into M local distribution networks based on this information. Subsequently, a distribution network operation and maintenance platform is obtained, which can communicate with multiple edge point cloud segmentation nodes. These edge point cloud segmentation nodes are responsible for processing and analyzing the data collected from each local distribution network. After that, a ground laser radar is configured in each local distribution network, and a communication connection is established with the corresponding edge point cloud segmentation node. When the laser radar collects the data, it will transmit the data to the corresponding edge point cloud segmentation node. At these nodes, 3D point cloud automatic segmentation will be performed. Then, based on these segmentation results, risk analysis is performed to identify the exogenous risks and endogenous risks that each local distribution network may face, and these risks are classified into M groups of local distribution risks. Finally, the distribution network operation and maintenance platform will conduct a comprehensive evaluation of these M groups of local distribution risks, and formulate and implement a global risk relief strategy based on the evaluation results, to achieve intelligent operation and management of the distribution network and improve the efficiency and accuracy of 3D point cloud segmentation of the distribution network.

[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A schematic diagram of a flow chart of a method for automatic segmentation of 3D point clouds of a power distribution network in an embodiment;

[0011] Figure 2 A schematic diagram of a process of obtaining M edge point cloud segmentation nodes in a method for automatic segmentation of 3D point cloud of a power distribution network in an embodiment. DETAILED DESCRIPTION

[0012] The embodiment of the present application solves the technical problem of insufficient accuracy of local feature extraction by PointNet by providing a method for automatic segmentation of 3D point clouds of distribution networks.

[0013] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0014] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0015] Example

[0016] like Figure 1 As shown, the present application provides a method for automatic segmentation of 3D point cloud of power distribution network, the method comprising:

[0017] The design information of the target power distribution network is interactively obtained, and the target power distribution network is divided into M local power distribution networks based on the design information, where M is a positive integer.

[0018] In an embodiment of the present application, the system terminal interacts with the target distribution network to obtain the design information of the target distribution network. This information includes key data such as network layout and equipment location. Subsequently, based on the obtained design information, the system terminal subdivides the target distribution network into M smaller, independent local distribution networks. Wherein, M is a positive integer, representing the number of local distribution networks into which the target distribution network is subdivided. The purpose of this step is to enable subsequent point cloud data collection and processing to be more targeted. Each local distribution network can independently collect, segment and assess the risk of point cloud data, thereby improving the efficiency and accuracy of the entire distribution network management.

[0019] A power distribution network operation and maintenance platform is obtained, wherein the power distribution network operation and maintenance platform is communicatively connected to a plurality of edge point cloud segmentation nodes.

[0020] In one embodiment, during the process of automatic segmentation of the 3D point cloud of the distribution network, the system terminal obtains a pre-built distribution network operation and maintenance platform. This distribution network operation and maintenance platform is a core component of the automated segmentation process, including modules such as risk level evaluation module and risk distance evaluation module. And the distribution network operation and maintenance platform establishes a communication connection with multiple edge point cloud segmentation nodes to coordinate and manage the entire segmentation process. Among them, these edge point cloud segmentation nodes are built based on technologies such as PointNet architecture, HEA module, feature splicing algorithm, etc. These edge point cloud segmentation nodes play a vital role in the entire process because they will receive 3D point cloud data obtained from the distribution network and execute automated segmentation algorithms.

[0021] Further, if Figure 2 As shown, the present application provides a method for obtaining the M edge point cloud segmentation nodes, and the method also includes:

[0022] A global feature extraction branch is constructed based on the PointNet architecture; a local feature extraction branch is constructed based on the HEA module; and a feature fusion layer is constructed based on the feature splicing algorithm.

[0023] Preferably, in order to construct the edge point cloud segmentation node, the system terminal first constructs a global feature extraction branch based on the PointNet architecture. This global feature extraction branch is a network structure consisting of an input layer, a shared multi-layer perceptron (MLP), a feature extraction layer, and a maximum pooling layer. Among them, the constructed input layer can receive point cloud data (pcd), and the point cloud data format is n×3, where n is the number of points and 3 represents the spatial coordinates (x, y, z) of each point. The shared multi-layer perceptron (sharedMLP) is implemented by convolution with shared weights. The size of the first convolution kernel is 1x3. At this time, because each point is three-dimensional, that is, (x, y, z), the size of each remaining convolution kernel is 1×1. The feature extraction layer connects each point and then extracts 1024-dimensional features for each point to form an n×1024 feature matrix. Maximum pooling is to apply the maximum pooling function to aggregate the n×1024 feature matrix into a 1×1024 global feature vector, which contains the global information of the entire point cloud. Through the collaborative work of these four parts, the system terminal can build a global feature extraction branch.

[0024] For the local feature extraction branch, it is achieved by embedding the HEA module in the PointNet architecture. The HEA module takes the point cloud features extracted from the global feature extraction branch built on the PointNet architecture as input, and through multi-level encoding, it can capture local structural information of different scales and levels in the point cloud. These features reflect the detailed structure and properties of different regions in the point cloud.

[0025] For the feature fusion layer, the system terminal encapsulates the feature concatenation algorithm with the input layer and the output layer to form this feature fusion layer. The feature concatenation algorithm expands the generated global feature vector to match the number of rows of the local feature matrix to obtain the expanded global feature matrix. The expanded global feature matrix is ​​then concatenated with the local feature matrix in the feature dimension to obtain the fused feature matrix.

[0026] By building the above three parts, the system terminal can enable the edge point cloud segmentation node to extract global and local features of point cloud data and make comprehensive use of these features through feature fusion. These functions can enable the edge point cloud segmentation node to have better performance and robustness when processing complex point cloud data tasks such as classification, segmentation, detection, etc.

[0027] A plurality of sample point cloud monomer parameters and a plurality of sample point cloud classification labels are interactively obtained, and constructed based on the plurality of sample point cloud monomer parameters and the plurality of sample point cloud classification labels.

[0028] Preferably, when constructing the point cloud segmentation layer, the system terminal first collects a series of sample data, which include sample point cloud monomer parameters and corresponding sample point cloud classification labels. Point cloud monomer parameters refer to the detailed features of a single object in each point cloud data set, such as the three-dimensional coordinates, color, density, etc. of each point. The classification label is to label these point cloud monomers by category, such as buildings, trees, vehicles, etc. Subsequently, the system terminal extracts intermediate data of multiple sample point cloud monomer parameters through the constructed feature fusion layer, and combines the fusion results with multiple sample point cloud classification labels, and trains the point cloud segmentation layer constructed using the back propagation neural network to obtain the final point cloud segmentation layer. The point cloud segmentation layer is an important component of the edge point cloud segmentation node, and its main function is to divide a complete point cloud data set into multiple meaningful monomers or regions. These monomers or regions correspond to different objects or different parts of objects in reality.

[0029] Configure the input layer, cascade the input layer, the global feature extraction branch, the feature fusion layer and the point cloud segmentation layer, connect the output ends of the input layer and the global feature extraction branch to the input end of the local feature extraction branch, connect the output end of the local feature extraction branch to the input end of the feature fusion layer, and generate a standard edge segmentation node; copy the standard edge segmentation node to obtain the M edge point cloud segmentation nodes.

[0030] Preferably, the system terminal first sets the input layer of the standard edge segmentation node, which is used to receive the original point cloud data. The point cloud data contains a large number of points in three-dimensional space, and each point may also contain additional information such as color and density. The data of the input layer will be passed to the global feature extraction branch. This branch processes the entire point cloud data to extract global features that can describe the entire point cloud. At the same time, the data of the input layer will also be passed to the local feature extraction branch. The local feature extraction branch focuses on the local area or a single point in the point cloud and extracts features that can describe these local areas or points. These local features are very important for subsequent segmentation tasks because they can help identify the boundaries of different objects or different parts of objects in the point cloud. After the global feature extraction branch and the local feature extraction branch, a feature fusion layer is set. The function of this layer is to fuse the global features and the local features to generate a fused feature that contains both global and local information. Subsequently, the fused features are sent to the point cloud segmentation layer. This layer segments the point cloud based on the fused features and divides the points in the point cloud into different categories or regions. After cascading the input layer, global feature extraction branch, local feature extraction branch, feature fusion layer, and point cloud segmentation layer through the connection relationship described above, the system terminal obtains a complete standard edge segmentation node, which includes the entire processing flow from the input layer to the point cloud segmentation layer. In order to process more complex point cloud data and improve the efficiency of data processing, the system terminal copies the above standard edge segmentation node to obtain multiple edge point cloud segmentation nodes. These nodes can process different point cloud data segments in parallel to extract more diverse or finer features, thereby further improving the performance and accuracy of point cloud segmentation.

[0031] Furthermore, the present application provides interactively obtaining multiple sample point cloud monomer parameters and multiple sample point cloud classification labels, and constructing the point cloud segmentation layer based on the multiple sample point cloud monomer parameters and the multiple sample point cloud classification labels, the method comprising:

[0032] The multiple sample point cloud monomer parameters are input into the standard edge segmentation node to perform feature extraction, and intermediate data extraction is performed in the feature fusion layer to obtain multiple sample point cloud fusion features; the point cloud segmentation layer is constructed based on a back propagation neural network, and the point cloud segmentation layer is trained using the multiple sample point cloud fusion features and the multiple sample point cloud classification labels until the point cloud classification accuracy of the point cloud segmentation layer is higher than a preset accuracy.

[0033] Optionally, during the training of the point cloud segmentation layer, the system terminal inputs the monomer parameters of multiple sample point clouds into the constructed standard edge segmentation node. This node will perform a series of feature extraction operations, including global feature extraction and local feature extraction, and fuse these extracted global and local features together in the feature fusion layer to form the fusion features of each sample point cloud, thereby obtaining the fusion features of multiple sample point clouds. Subsequently, the system terminal constructs the number of neurons in the input layer, hidden layer, and output layer of the back propagation neural network (BPNN) according to actual needs. The weights and biases of the back propagation neural network are then initialized with random numbers. After that, the fusion features are input to the input layer of the point cloud segmentation layer. Through the calculation of the hidden layer, the information is transmitted layer by layer, and the predicted classification result is obtained at the output layer. The cross entropy loss function is then used to calculate the error between the predicted classification result and the classification label for the sample point cloud. Then, the gradient of the error to each layer weight is calculated according to the chain rule, and the weight and bias are updated using the gradient descent method to reduce the error. Repeat the process of forward propagation, error calculation, and back propagation until the point cloud classification accuracy of the back propagation neural network is higher than the preset accuracy. When the classification accuracy of the back propagation neural network is higher than the preset accuracy, the training process ends. At this time, the system terminal outputs the current back propagation neural network to generate a point cloud segmentation layer, which has learned how to accurately predict the category of the point cloud from the fusion features.

[0034] Furthermore, the present application provides a distribution network operation and maintenance platform, and the method also includes:

[0035] The distribution network operation and maintenance platform includes a risk level evaluation module, a risk distance evaluation module and a risk operation and maintenance analysis module, wherein the risk level evaluation module and the risk distance evaluation module are connected in parallel, and the output ends of the risk level evaluation module and the risk distance evaluation module are connected to the input end of the risk operation and maintenance analysis module;

[0036] Preferably, the distribution network operation and maintenance platform is a pre-built component that integrates risk identification, assessment and operation and maintenance analysis. The platform includes three core modules, namely, risk level evaluation module, risk distance evaluation module and risk operation and maintenance analysis module. The risk level evaluation module is mainly responsible for classifying the potential risks in the distribution network, and classifying the risks into different levels according to the severity and possibility of the risks. This module helps operation and maintenance personnel quickly understand which risks need to be prioritized. The risk distance evaluation module evaluates risks from another dimension, focusing on the distance between the location where the risk occurs and the distribution center. By calculating the risk distance, the operation and maintenance personnel can understand which risks are closer to themselves, so as to take targeted measures to prevent or respond. The two evaluation modules work in parallel, and they independently evaluate the risks in the distribution network, but with different focuses. The evaluation results of risk level and risk distance output by them will be passed to the risk operation and maintenance analysis module. The risk operation and maintenance analysis module is a comprehensive processing module. Its input end is connected to the output end of the risk level evaluation module and the risk distance evaluation module. It receives the output from the risk level evaluation module and the risk distance evaluation module to conduct in-depth operation and maintenance analysis. This module will generate a risk operation and maintenance coefficient to help operation and maintenance personnel formulate effective operation and maintenance plans and improve the stability and reliability of the distribution network.

[0037] Interactively obtain an exogenous risk level table and an endogenous risk level table, configure exogenous weights and endogenous weights for the exogenous risks and endogenous risks respectively, construct a risk level evaluation function based on the exogenous weights and endogenous weights, and synchronize the exogenous risk level table, endogenous risk level table and the risk level evaluation function to the risk level evaluation module.

[0038] Preferably, in order to more accurately assess the risk level, the system terminal interacts with the database of the distribution network operation and maintenance platform to obtain an exogenous risk level table and an endogenous risk level table. Exogenous risk refers to the risk caused by external environmental factors, such as the proximity of trees, buildings or other external objects to the distribution line, wire breakage, damaged insulators and other external facilities. Endogenous risk refers to the risk that may be caused by the internal equipment, system or operation of the distribution network, such as equipment aging, operational errors, etc. After obtaining these two risk level tables, the system terminal configures corresponding weights for exogenous risk and endogenous risk, namely, exogenous weights and endogenous weights. These weights represent the importance of different types of risks in the overall risk assessment, and the size of the weights is determined by factors such as expert experience and historical data. Subsequently, based on these weights, the system terminal constructs a risk level evaluation function. This function can combine the exogenous risk level table, the endogenous risk level table and the corresponding weights, and output a comprehensive risk level through weighted summation. This risk level reflects the overall risk level of the distribution network in its current state. Afterwards, the system terminal synchronizes the external risk level table, the internal risk level table and the constructed risk level evaluation function to the pre-constructed risk level evaluation module framework to form a risk level evaluation module. In this way, when the operation and maintenance personnel need to use the risk level evaluation module, it can quickly output an accurate risk level evaluation result based on the input risk information, combined with the external risk level table, the internal risk level table and the risk level evaluation function, to provide decision support for the operation and maintenance personnel.

[0039] A road traffic network is interactively obtained, and M operation and maintenance distance parameters between a distribution center and the M local distribution networks are obtained according to the design information of the target distribution network and the road traffic network, and the M operation and maintenance distance parameters are stored in the risk distance evaluation module.

[0040] Preferably, the system terminal interacts with the urban road platform to obtain information about the road traffic network. This road traffic network includes multiple factors such as roads and road conditions, which will directly affect the speed and convenience of the operation and maintenance personnel moving from the distribution center to each local distribution network. Subsequently, the M operation and maintenance distance parameters between the distribution center and the M local distribution networks are calculated by combining the design information of the target distribution network and the information of the road traffic network. These operation and maintenance distance parameters are the distance and time required for the operation and maintenance personnel to travel from the distribution center to each local distribution network in actual situations. Once these operation and maintenance distance parameters are obtained, the system terminal stores these parameters in the risk distance evaluation module. The main function of this module is to use the operation and maintenance distance as an important consideration when conducting risk assessment. For example, when a local distribution network fails, the risk distance evaluation module can evaluate the efficiency and difficulty of dispatching operation and maintenance personnel from the distribution center to the fault point based on the operation and maintenance distance parameters, thereby helping the operation and maintenance personnel to better formulate operation and maintenance strategies.

[0041] Risk weights and distance weights are configured for risk levels and risk distances respectively, and a risk operation and maintenance evaluation function is constructed based on the risk weights and distance weights; a serialization engine is pre-built, and the serialization engine and the risk operation and maintenance evaluation function are synchronized to the risk operation and maintenance analysis module to complete the construction of the risk operation and maintenance analysis module.

[0042] Preferably, in the process of building the risk operation and maintenance analysis module, the system terminal first quantitatively evaluates the two key factors of risk level and risk distance. To achieve this, the system terminal configures risk weights and distance weights for them respectively. These weights represent the importance of risk level and operation and maintenance distance in the overall operation and maintenance evaluation, and they are determined based on historical data, expert experience, and specific business needs. The risk weight measures the impact of the risk level on operation and maintenance. If the risk level is high, the corresponding risk weight will also be greater, which means that the risk factor will occupy a more important position in operation and maintenance. Similarly, the distance weight reflects the impact of the operation and maintenance distance on operation and maintenance efficiency and cost. A longer operation and maintenance distance may require more time and resources, so its weight will also increase accordingly. Subsequently, based on these weights, the system terminal constructs a risk operation and maintenance evaluation function. This function takes the risk level and risk distance as inputs, and outputs a comprehensive operation and maintenance evaluation result through weighted summation. This result comprehensively considers the two factors of risk level and operation and maintenance distance, and provides a comprehensive basis for operation and maintenance decision-making for operation and maintenance personnel. In order to support the operation of the risk operation and maintenance evaluation function, the system terminal pre-builds a serialization engine. The main function of the serialization engine is to sort the operation and maintenance analysis results generated by the risk operation and maintenance analysis module in order to perform global risk removal of the target distribution network. The construction of the risk operation and maintenance analysis module is completed by synchronizing the serialization engine and the risk operation and maintenance evaluation function to the risk operation and maintenance analysis module.

[0043] M ground laser radars are configured for the M local power distribution networks, M edge point cloud segmentation nodes are randomly activated at the multiple edge point cloud segmentation nodes according to the M ground laser radars, and the M ground laser radars are communicatively connected to the M edge point cloud segmentation nodes.

[0044] In one embodiment, in order to improve the real-time monitoring and data analysis capabilities of the local power distribution network, the system terminal configures ground laser radar devices on M local power distribution networks. Ground laser radar (LiDAR) is a high-precision device that measures the distance between the ground and objects through a laser beam. It can quickly generate point cloud data, that is, a collection of a large number of three-dimensional coordinate points. These point cloud data can accurately reflect the three-dimensional structure of the surrounding environment. After the configuration is completed, the system terminal uses a random activation mechanism to randomly select and activate M nodes corresponding to M ground laser radars from multiple edge point cloud segmentation nodes. The edge point cloud segmentation node is a computing unit responsible for processing and analyzing point cloud data. It is deployed at the edge of the network to reduce the processing burden of the central server and improve the speed and efficiency of data processing. Subsequently, the M ground laser radars will be connected to their corresponding M edge point cloud segmentation nodes by communication. In this way, the point cloud data collected by each laser radar can be transmitted to the corresponding edge node in real time for processing and analysis. This distributed data processing method not only improves the speed of data processing, but also enhances the flexibility and scalability of the system.

[0045] The M ground laser radars are dispatched to collect M local laser point cloud data of the M local power distribution networks.

[0046] In one embodiment, in order to obtain detailed surrounding information of M local power distribution networks, the system terminal dispatches M ground laser radars for data collection. These M ground laser radars will scan the local power distribution networks they are responsible for respectively, and generate laser point cloud data of the local power distribution network. Laser point cloud data is a collection of a large number of three-dimensional coordinate points that can accurately reflect the three-dimensional structure of the surrounding environment, including terrain, buildings, vegetation, etc. In this way, the system terminal can obtain detailed surrounding information of M local power distribution networks in real time, providing support for subsequent operation and maintenance.

[0047] The M edge point cloud segmentation nodes receive the M local laser point cloud data to perform 3D point cloud automatic segmentation, perform risk analysis based on the segmentation results, and obtain M groups of local power distribution risks, wherein each group of local power distribution risks includes exogenous risks and endogenous risks.

[0048] In one embodiment, after obtaining M local laser point cloud data, the system terminal performs dimensionality reduction processing on these data to generate M local dimensionality reduction data. Subsequently, these dimensionality reduction data are transmitted to the corresponding M edge point cloud segmentation nodes. Once the M edge point cloud segmentation nodes receive the local dimensionality reduction data, the M edge point cloud segmentation nodes will immediately perform 3D point cloud automatic segmentation. This step is to use the global feature extraction branch, feature fusion layer and point cloud segmentation layer inside the edge point cloud segmentation node to divide the huge local dimensionality reduction data so as to more accurately identify and analyze different elements in the environment. Through automatic segmentation, the system terminal can quickly identify external factors that may pose a threat to the distribution network, such as the proximity of trees, buildings, and other external objects to the distribution lines. Based on the M segmentation results, the system terminal restores and generates M local restored networks, and generates M exogenous risks and M endogenous risks according to the preset exogenous risk distance and endogenous fault identification model. These two risks together constitute M local distribution risks. These risk information can help operation and maintenance personnel more accurately assess the security status of the current network and formulate corresponding operation and maintenance strategies.

[0049] Furthermore, the present application provides that the M edge point cloud segmentation nodes receive the M local laser point cloud data to perform 3D point cloud automatic segmentation, before which the method further includes:

[0050] A standard point cloud space is pre-constructed, and the first local laser point cloud data is fitted to the standard point cloud space to obtain a first point cloud space, wherein the first local laser point cloud data is any one of the M local laser point cloud data.

[0051] Preferably, in order to improve the efficiency and accuracy of point cloud data segmentation, the system terminal pre-constructs a standard point cloud space as a reference framework. This standard point cloud space represents an expected distribution network environment, and contains the spatial distribution information of elements such as the basic structure, equipment and obstacles of the network. When the actual M local laser point cloud data are obtained, the system terminal randomly extracts one from the M local laser point cloud data as the first local laser point cloud data, and fits it with the pre-constructed standard point cloud space, that is, the actually collected point cloud data is matched to the standard point cloud space to form a new point cloud space combined with the actual environment information, and the newly generated point cloud space is used as the first point cloud space. Through fitting, the first point cloud space not only contains the basic structural information in the standard point cloud space, but also incorporates the actual environment details in the first local laser point cloud data. This enables the first point cloud space to more completely and accurately reflect the actual distribution network environment.

[0052] A voxel resolution and a voxel adjustment step are preset, and based on the voxel resolution and the voxel adjustment step, data dimensionality reduction of the first local laser point cloud data is performed in the first point cloud space to obtain first local dimensionality reduction data.

[0053] Preferably, in order to more effectively process and analyze point cloud data, the system terminal introduces a method of data dimensionality reduction. This involves two key parameters, namely voxel resolution and voxel adjustment step size. Voxel resolution determines the fineness of the point cloud space being divided into three-dimensional voxel grids, while voxel adjustment step size is used to gradually adjust the size and shape of voxels during data dimensionality reduction. After obtaining the first point cloud space, the system terminal divides the space into three-dimensional voxel grids according to the preset voxel resolution. Each voxel grid is a small cubic space that contains all point cloud data in the area. In order to reduce the dimension of the point cloud data and reduce the amount of calculation, the system terminal uses the center point of all point clouds in each voxel grid to represent all points in the voxel, and the distance from the center point to all points in the voxel is roughly equal. In this way, the original large amount of point cloud data is simplified to a smaller number of center points, thereby achieving data dimensionality reduction. This data dimensionality reduction method not only retains the main features of the original point cloud data, but also greatly reduces the complexity and processing difficulty of the data. Based on the first local dimensionality reduction data after dimensionality reduction, the system terminal can perform subsequent segmentation work more efficiently.

[0054] By analogy, data dimensionality reduction is performed on the M local laser point cloud data to obtain M local dimensionality reduction data; the M edge point cloud segmentation nodes perform automatic 3D point cloud segmentation based on the M local dimensionality reduction data.

[0055] Preferably, for the M local laser point cloud data collected by the M local power distribution networks, the system will adopt the same steps as the aforementioned steps to obtain the first local dimensionality reduction data to obtain M local dimensionality reduction data, which are more concise and easier to process than the original point cloud data. Subsequently, the system terminal inputs the M local dimensionality reduction data into the M edge point cloud segmentation nodes to perform the aforementioned 3D point cloud automatic segmentation process. This process further divides the dimensionality-reduced point cloud data into smaller areas to facilitate more accurate identification and analysis of different elements in the environment.

[0056] Further, the present application provides a preset voxel resolution and a voxel adjustment step size, and performs data dimensionality reduction of the first local laser point cloud data in the first point cloud space based on the voxel resolution and the voxel adjustment step size to obtain first local dimensionality reduction data, and the method also includes:

[0057] An optimization frequency threshold is preset, and the optimization frequency threshold is used as a constraint, and the voxel adjustment step is used to perform the voxel resolution adjustment to obtain multiple alternative resolutions; the first point cloud space is divided using the voxel resolution to obtain a first voxel grid, and representative points of the first voxel grid are retained to obtain first reduced-dimensionality point cloud data.

[0058] Optionally, in order to find the most suitable voxel resolution to perform dimensionality reduction of point cloud data, the system terminal will preset an optimization frequency threshold. This threshold is set based on historical experience and is used to limit the number of times the voxel resolution is adjusted to ensure that the optimal resolution is found within a reasonable computational cost. Subsequently, the system terminal gradually adjusts the preset voxel resolution using the voxel adjustment step, and after each adjustment, a new voxel resolution is generated as an alternative. This process is repeated until the preset optimization frequency threshold is reached. The voxel resolution obtained after each adjustment is recorded to form multiple alternative resolutions. After that, the system terminal uses the preset resolution to divide the first point cloud space and generate a first voxel grid. This grid divides the point cloud space into multiple small cubes, each of which (voxel) contains a part of the original point cloud data. Then, the first voxel grid is subjected to a representative point retention operation. This operation is the same as the above, and the center point in each voxel is selected to represent all the points in the voxel, thereby achieving dimensionality reduction of the point cloud data to obtain the first reduced dimensionality point cloud data, which is more concise than the original point cloud data while retaining the main features of the original data.

[0059] Divide the first point cloud space based on the multiple candidate resolutions to obtain multiple candidate voxel grids, retain representative points of the multiple candidate voxel grids, and obtain multiple candidate reduced-dimensionality point cloud data; perform point cloud coincidence analysis on the first reduced-dimensionality point cloud data and the multiple candidate reduced-dimensionality point cloud data to obtain multiple point cloud coincidence coefficients;

[0060] Optionally, the system terminal uses the same method as the above-mentioned method for obtaining the first reduced dimension point cloud data, divides the first point cloud space using multiple alternative resolutions, and generates multiple alternative reduced dimension point cloud data. Subsequently, in order to evaluate the quality of these alternative reduced dimension point cloud data, the system terminal performs point cloud coincidence analysis on these alternative reduced dimension point cloud data and the first reduced dimension point cloud data. Point cloud coincidence analysis is a method for quantitatively evaluating the similarity between two point cloud data sets. Specifically, the system terminal randomly selects one from multiple alternative reduced dimension point cloud data, compares it with the first reduced dimension point cloud data, and for each point in the first reduced dimension point cloud data, calculates its Euclidean distance to the nearest point in the alternative reduced dimension point cloud data. Subsequently, all the calculated Euclidean distances are summarized and their average is calculated to obtain an overall indicator representing the distance between the two point clouds. The calculated average value is then reciprocally calculated to obtain a point cloud coincidence coefficient. Afterwards, the remaining alternative reduced dimension point cloud data are compared using the same method to obtain multiple point cloud coincidence coefficients. A higher point cloud overlap coefficient means that the alternative dimensionality reduction point cloud data is closer in structure to the first dimensionality reduction point cloud data, and therefore performs better in preserving the characteristics of the original point cloud data.

[0061] Perform majority aggregation on the plurality of point cloud coincidence coefficients, and extract the first local dimensionality reduction data from the plurality of candidate dimensionality reduction point cloud data based on the aggregation result.

[0062] Optionally, when analyzing the overlap coefficients of multiple point clouds, the system terminal can effectively determine which candidate dimensionality reduction point cloud data can best represent the original dense point cloud distribution by using the majority aggregation method. Majority aggregation refers to selecting the value with the highest frequency of occurrence as the aggregation result. The system terminal obtains statistics on the multiple point cloud overlap coefficients to find the value with the highest frequency of occurrence, that is, the mode. Since the mode represents the point cloud overlap coefficient with the highest frequency of occurrence, the candidate dimensionality reduction point cloud data corresponding to this point cloud overlap coefficient is structurally closest to the original point cloud and can retain the distribution characteristics and detail information of the original point cloud to the greatest extent. Therefore, based on this aggregation result, we extract the corresponding point cloud data from multiple candidate dimensionality reduction point cloud data as the first local dimensionality reduction data.

[0063] Further, the present application provides a method for obtaining the M group of local power distribution risks, the method further comprising:

[0064] The M edge point cloud segmentation nodes perform 3D point cloud automatic segmentation based on the M local laser point cloud data to obtain M groups of point cloud segmentation results, wherein each group of point cloud segmentation results includes a distribution network point cloud set and an interference factor point cloud set; based on the M groups of point cloud segmentation results, a first group of point cloud segmentation results is obtained by calling, wherein the first group of point cloud segmentation results includes a first distribution network point cloud set and a first interference factor point cloud set.

[0065] Optionally, the system terminal inputs the M local laser point cloud data after dimensionality reduction into the corresponding M edge point cloud segmentation nodes, performs automatic segmentation of 3D point clouds, and obtains M groups of point cloud segmentation results. Among them, each group of point cloud segmentation results includes a distribution network point cloud set and an interference factor point cloud set. The distribution network point cloud set contains key points and structural information in the distribution network. The interference factor point cloud set contains information on objects such as trees and buildings. Subsequently, the system terminal selects the first group of point cloud segmentation results from these M groups of point cloud segmentation results as the starting point for further analysis. This group of point cloud segmentation results includes the first distribution network point cloud set and the first interference factor point cloud set.

[0066] Based on the restoration of the first distribution network point cloud set, a first local restoration network is generated; an exogenous risk distance is preset, and the first interference factor point cloud set is traversed to select and locate the first exogenous risk with the exogenous risk distance and the first local restoration network as constraints.

[0067] Optionally, the system terminal uses the Poisson surface reconstruction method to convert the first distribution network point cloud obtained into a grid structure, such as a triangular grid, a tetrahedral grid, etc. Subsequently, on the basis of the gridding, the topological relationship of the distribution network is constructed. This includes determining the location and connection relationship of network components such as nodes, branches, transformers, switches, etc. Then, attribute information such as voltage level, capacity, impedance, etc. is assigned to the network components. These attribute information are obtained based on the measurement values ​​in the original point cloud data. When the attribute information is assigned, the system terminal reconstructs and generates the first local restoration network. After that, the system terminal sets an exogenous risk distance, which is set according to expert advice and is used to define the minimum safe distance of external factors that may affect the distribution network. Then, using this exogenous risk distance and the first local restoration network that has been constructed as constraints, the first interference factor point cloud set is traversed. This process is to analyze each point in the interference factor point cloud set to determine whether the distance between them and the distribution network is less than or equal to the preset exogenous risk distance. Finally, when the system terminal finds objects that are too close to the distribution network, that is, less than or equal to the external risk distance, in the interference factor point cloud set, it selects and locates them to obtain the first external risk. These risk points may directly affect the normal operation of the distribution network, and corresponding measures need to be taken to prevent or deal with them.

[0068] An endogenous fault identification model is pre-constructed, and the endogenous fault identification model is used to detect the first local restoration network to obtain a first endogenous risk. The first endogenous risk and the first exogenous risk constitute a first group of local power distribution risks; and so on, the M groups of local power distribution risks are obtained.

[0069] Optionally, in order to accurately assess the endogenous risk, the system terminal pre-builds an endogenous fault identification model. Specifically, the system terminal collects historical distribution network fault data and corresponding label data, which includes detailed information of various types of endogenous faults, such as equipment aging, overload, short circuit, etc. Subsequently, the endogenous fault identification model is initialized using a back propagation neural network according to the complexity of the problem and the characteristics of the data. Then, using a method similar to the aforementioned training point cloud segmentation layer, the initialized endogenous fault identification model is trained using the historical distribution network fault data and the corresponding label data to obtain the final endogenous fault identification model. After the model is built, the system terminal uses this endogenous fault identification model to detect potential faults in the first local restoration network. Through model analysis, a first endogenous risk can be obtained, which represents a possible fault or abnormality in the local distribution network. Afterwards, the system terminal combines the first endogenous risk obtained with the first exogenous risk obtained previously to form a first group of local distribution risks. By analogy, the system terminal performs the same processing on the remaining local restoration networks to obtain M groups of local distribution risks. This risk information is crucial for the maintenance and management of distribution networks and can help timely detect and address potential faults and safety hazards.

[0070] The power distribution network operation and maintenance platform performs risk assessment on the M groups of local power distribution risks, and performs global risk elimination of the target power distribution network based on the assessment results.

[0071] In one embodiment, after the distribution network operation and maintenance platform receives M groups of local distribution risks, a series of risk assessment processes will be carried out. This process uses the internal risk level assessment module, risk distance assessment module and risk operation and maintenance analysis module to conduct risk assessment on the input M groups of local distribution risks, thereby obtaining an assessment result. This evaluation result contains K risk operation and maintenance coefficients. Subsequently, the serialization engine is activated to sort the local distribution network operation and maintenance of the K risk operation and maintenance coefficients, and the global risk relief of the target distribution network is performed according to the sorting results. The relief method includes emergency maintenance of high-risk local distribution networks, strengthening monitoring and early warning measures, etc.

[0072] Further, the present application provides that the distribution network operation and maintenance platform performs risk assessment on the M groups of local distribution risks, and performs global risk relief of the target distribution network based on the assessment results, and the method also includes:

[0073] Eliminate the empty set of the M groups of local power distribution risks to obtain K groups of local real-time risks, where K is a positive integer less than or equal to M; in the risk level evaluation module, use the K groups of local real-time risks to traverse the external risk level table and the internal risk level table for level matching, and calculate based on the risk level evaluation function to obtain K real-time risk levels,

[0074] Optionally, the system terminal performs an empty set elimination process on the M groups of local power distribution risks that have been identified and located. Empty set elimination is an important step used to exclude local power distribution areas that have neither endogenous risks nor exogenous risks. These areas are considered to be risk-free or extremely low risk in risk management, so no further risk assessment and relief measures are required. After the empty set elimination, the system terminal organizes the remaining local power distribution risks into K groups of local real-time risks, where K is a positive integer less than or equal to M. These real-time risks represent the risk points that really exist in the current distribution network and need to be paid attention to. Subsequently, in the risk level evaluation module, the system terminal uses the K groups of local real-time risks to traverse the exogenous risk level table and the endogenous risk level table. These two tables contain the level classification and evaluation criteria of various exogenous risks and endogenous risks, respectively. By traversing these tables, real-time risks can be matched with corresponding risk levels. Afterwards, based on the risk level evaluation function, the matched risk levels are calculated to obtain K real-time risk levels. These real-time risk levels reflect the severity and urgency of various risk points in the current distribution network, providing an important reference for subsequent risk relief and response measures.

[0075] In the risk distance evaluation module, K operation and maintenance distance parameters are obtained according to the K groups of local real-time risk scheduling; in the risk operation and maintenance analysis module, the K real-time risk levels and K operation and maintenance distance parameters are substituted into the risk operation and maintenance evaluation function to calculate and obtain K risk operation and maintenance coefficients; the serialization engine is activated to perform local distribution network operation and maintenance sorting with reference to the K risk operation and maintenance coefficients, generate a real-time operation and maintenance sequence, and execute global risk relief of the target distribution network based on the real-time operation and maintenance sequence.

[0076] Optionally, after K groups of local real-time risks are identified and their levels are determined, the system terminal will transmit the K groups of local real-time risks to the risk distance evaluation module. The main task of this module is to evaluate the time and distance required for operation and maintenance personnel to reach these risk points, that is, the operation and maintenance distance parameters. These parameters reflect the accessibility of operation and maintenance personnel to deal with risks. In the risk operation and maintenance analysis module, the system terminal will input K real-time risk levels and corresponding K operation and maintenance distance parameters into the risk operation and maintenance evaluation function. The function of this function is to comprehensively consider the risk level and operation and maintenance distance to calculate a risk operation and maintenance coefficient. This coefficient not only reflects the severity of the risk, but also takes into account the operation and maintenance resources and time required to deal with these risks. Once the K risk operation and maintenance coefficients are obtained, the system terminal will activate the sequence engine. This engine will refer to these coefficients to sort the risk points in the local distribution network and generate a real-time operation and maintenance sequence. This sequence is arranged according to the priority of the risk operation and maintenance coefficient, so that the operation and maintenance personnel can give priority to the risk points with the highest risk and the best operation and maintenance conditions. Afterwards, based on this real-time operation and maintenance sequence, the system terminal will guide the operation and maintenance personnel to perform global risk elimination of the target distribution network. This includes detailed inspection, repair, equipment replacement or other necessary measures at each risk point to ensure the stable and safe operation of the distribution network.

[0077] In summary, the embodiments of the present application have at least the following technical effects:

[0078] The embodiment of the present application interactively obtains design information, divides the target distribution network into multiple local networks, and uses the operation and maintenance platform and ground laser radar to collect point cloud data. Through the edge point cloud segmentation node, automatic segmentation is performed, and risk analysis is performed to obtain the local distribution risk. The method also involves feature extraction based on PointNet and HEA modules, as well as feature fusion and point cloud segmentation layer construction and training. Furthermore, the method also includes restoring the network, locating exogenous risks, detecting endogenous risks, and data dimensionality reduction and optimization. Finally, the operation and maintenance platform evaluates the risks and generates an operation and maintenance sequence through an evaluation module and a serialization engine to achieve global risk removal and optimize the operation and maintenance of the distribution network. These technical effects jointly solve the technical problem of insufficient local feature extraction accuracy of PointNet, and achieve the effect of improving the efficiency and accuracy of 3D point cloud segmentation of the distribution network.

[0079] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0080] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0081] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for automatic segmentation of 3D point cloud of power distribution network, characterized in that: The method comprises: interactively obtaining design information of a target power distribution network, and dividing the target power distribution network into M local power distribution networks based on the design information, where M is a positive integer; Acquire a power distribution network operation and maintenance platform, wherein the power distribution network operation and maintenance platform is communicatively connected to a plurality of edge point cloud segmentation nodes; Configure M ground laser radars for the M local power distribution networks, randomly activate M edge point cloud segmentation nodes at the multiple edge point cloud segmentation nodes according to the M ground laser radars, and connect the M ground laser radars to the M edge point cloud segmentation nodes for communication; Dispatching the M ground laser radars to collect M local laser point cloud data of the M local power distribution networks; The M edge point cloud segmentation nodes receive the M local laser point cloud data to perform 3D point cloud automatic segmentation, perform risk analysis based on the segmentation results, and obtain M groups of local power distribution risks, wherein each group of local power distribution risks includes exogenous risks and endogenous risks; The power distribution network operation and maintenance platform performs risk assessment on the M groups of local power distribution risks, and performs global risk elimination of the target power distribution network based on the assessment results; Wherein, the method further comprises: Build a global feature extraction branch based on the PointNet architecture; Construct a local feature extraction branch based on the HEA module; Construct feature fusion layer based on feature splicing algorithm; Interactively obtain multiple sample point cloud monomer parameters and multiple sample point cloud classification labels, and construct a point cloud segmentation layer based on the multiple sample point cloud monomer parameters and the multiple sample point cloud classification labels; Configure an input layer, cascade the input layer, the global feature extraction branch, the feature fusion layer and the point cloud segmentation layer, connect the output end of the input layer and the global feature extraction branch to the input end of the local feature extraction branch, connect the output end of the local feature extraction branch to the input end of the feature fusion layer, and generate a standard edge segmentation node; Copy the standard edge segmentation node to obtain the M edge point cloud segmentation nodes; The method of interactively obtaining a plurality of sample point cloud monomer parameters and a plurality of sample point cloud classification labels, and constructing the point cloud segmentation layer based on the plurality of sample point cloud monomer parameters and the plurality of sample point cloud classification labels, includes: Inputting the monomer parameters of the plurality of sample point clouds into the standard edge segmentation node to perform feature extraction, and performing intermediate data extraction in the feature fusion layer to obtain fusion features of the plurality of sample point clouds; The point cloud segmentation layer is constructed based on a back propagation neural network, and the point cloud segmentation layer is trained using the plurality of sample point cloud fusion features and the plurality of sample point cloud classification labels until the point cloud classification accuracy of the point cloud segmentation layer is higher than a preset accuracy; Wherein, the method further comprises: The M edge point cloud segmentation nodes perform 3D point cloud automatic segmentation based on the M local laser point cloud data to obtain M groups of point cloud segmentation results, wherein each group of point cloud segmentation results includes a distribution network point cloud set and an interference factor point cloud set; Based on the M groups of point cloud segmentation results, a first group of point cloud segmentation results is obtained, wherein the first group of point cloud segmentation results includes a first distribution network point cloud set and a first interference factor point cloud set; Generate a first local restored network based on the first distribution network point cloud set restoration; Preset an exogenous risk distance, and use the exogenous risk distance and the first local restoration network as constraints to traverse the first interference factor point cloud set to select and locate the first exogenous risk; Pre-constructing an endogenous fault identification model, and using the endogenous fault identification model to detect the first local restoration network to obtain a first endogenous risk, wherein the first endogenous risk and the first exogenous risk constitute a first group of local power distribution risks; By analogy, the M groups of local power distribution risks are obtained.

2. A method for automatic segmentation of 3D point cloud of power distribution network according to claim 1, characterized in that: The M edge point cloud segmentation nodes receive the M local laser point cloud data to perform 3D point cloud automatic segmentation, and before that, the method further includes: Pre-constructing a standard point cloud space, and fitting the first local laser point cloud data to the standard point cloud space to obtain a first point cloud space, wherein the first local laser point cloud data is any one of the M local laser point cloud data; Preset a voxel resolution and a voxel adjustment step size, and perform data dimensionality reduction of the first local laser point cloud data in the first point cloud space based on the voxel resolution and the voxel adjustment step size to obtain first local dimensionality reduction data; Similarly, data dimension reduction is performed on the M local laser point cloud data to obtain M local dimension reduction data; The M edge point cloud segmentation nodes perform automatic segmentation of 3D point clouds based on the M local dimensionality reduction data.

3. A method for automatic segmentation of 3D point cloud of power distribution network according to claim 2, characterized in that: Presetting a voxel resolution and a voxel adjustment step, and performing data dimensionality reduction of the first local laser point cloud data in the first point cloud space based on the voxel resolution and the voxel adjustment step to obtain first local dimensionality reduction data, the method further includes: Presetting an optimization frequency threshold, and taking the optimization frequency threshold as a constraint, performing the voxel resolution adjustment using the voxel adjustment step length to obtain a plurality of candidate resolutions; Dividing the first point cloud space by using the voxel resolution to obtain a first voxel grid, retaining representative points of the first voxel grid, and obtaining first reduced-dimensionality point cloud data; Divide the first point cloud space based on the multiple candidate resolutions to obtain multiple candidate voxel grids, retain representative points of the multiple candidate voxel grids, and obtain multiple candidate reduced-dimensionality point cloud data; Performing point cloud coincidence analysis on the first reduced-dimensionality point cloud data and the plurality of candidate reduced-dimensionality point cloud data to obtain a plurality of point cloud coincidence coefficients; Perform majority aggregation on the plurality of point cloud coincidence coefficients, and extract the first local dimensionality reduction data from the plurality of candidate dimensionality reduction point cloud data based on the aggregation result.

4. The method for automatic segmentation of 3D point cloud of power distribution network according to claim 1, characterized in that: The method further comprises: The distribution network operation and maintenance platform includes a risk level evaluation module, a risk distance evaluation module and a risk operation and maintenance analysis module, wherein the risk level evaluation module and the risk distance evaluation module are connected in parallel, and the output ends of the risk level evaluation module and the risk distance evaluation module are connected to the input end of the risk operation and maintenance analysis module; Interactively obtain an exogenous risk level table and an endogenous risk level table, and configure exogenous weights and endogenous weights for the exogenous risks and endogenous risks respectively, construct a risk level evaluation function based on the exogenous weights and endogenous weights, and synchronize the exogenous risk level table, the endogenous risk level table and the risk level evaluation function to the risk level evaluation module; Interactively obtain a road traffic network, and obtain M operation and maintenance distance parameters between a distribution center and the M local distribution networks according to the design information of the target distribution network and the road traffic network, and store the M operation and maintenance distance parameters in the risk distance evaluation module; configuring risk weights and distance weights for risk levels and risk distances respectively, and constructing a risk operation and maintenance evaluation function based on the risk weights and distance weights; A serialization engine is pre-built, and the serialization engine and the risk operation and maintenance evaluation function are synchronized to the risk operation and maintenance analysis module to complete the construction of the risk operation and maintenance analysis module.

5. A method for automatic segmentation of 3D point cloud of power distribution network according to claim 4, characterized in that: The power distribution network operation and maintenance platform performs risk assessment on the M groups of local power distribution risks, and performs global risk relief on the target power distribution network based on the assessment results. The method further includes: Eliminate the empty set of the M groups of local power distribution risks to obtain K groups of local real-time risks, where K is a positive integer less than or equal to M; In the risk level evaluation module, the K groups of local real-time risks are used to traverse the external risk level table and the internal risk level table for level matching, and K real-time risk levels are obtained based on the risk level evaluation function calculation; In the risk distance evaluation module, K operation and maintenance distance parameters are obtained according to the K groups of local real-time risk scheduling; In the risk operation and maintenance analysis module, the K real-time risk levels and the K operation and maintenance distance parameters are substituted into the risk operation and maintenance evaluation function to calculate and obtain K risk operation and maintenance coefficients; The serialization engine is activated to perform operation and maintenance sorting of the local distribution network with reference to the K risk operation and maintenance coefficients, generate a real-time operation and maintenance sequence, and perform global risk removal of the target distribution network based on the real-time operation and maintenance sequence.

Images