An image matching and anomaly detection method for cable layout in an energy storage battery cabin

Abstract

This invention discloses an image matching and anomaly detection method for cable layout inside an energy storage battery compartment, relating to the field of energy storage system technology. The method includes acquiring three-dimensional scanning data of the inside of the energy storage battery compartment, extracting an initial point set of cable layout using point cloud processing technology to obtain geometric feature points of the cable distribution, and performing feature matching based on these geometric feature points and a preset standard cable model to obtain an initial matching result. This image matching and anomaly detection method for cable layout inside an energy storage battery compartment significantly improves the safety, stability, and operational reliability of cable layout within the energy storage battery compartment.

Description

Technical Field

[0001] This invention relates to the field of energy storage system technology, specifically to an image matching and anomaly detection method for cable routing inside an energy storage battery compartment. Background Technology

[0002] In the field of energy storage and management, energy storage battery compartments are crucial components of power systems, with their internal cables playing a vital role in power transmission and signal exchange. As energy storage systems expand in scale and increase in system integration, the number of cables inside the battery compartments has significantly increased, and the structure has become increasingly complex. To ensure the safe and stable operation of the system, it is typically necessary to perform 3D scanning modeling of the battery compartment's internal structure and analyze the cable distribution based on the scan data to assess and manage the rationality of the cable layout.

[0003] In existing technologies, analysis methods for cable routing within battery compartments primarily rely on point cloud data for geometric feature extraction and modeling to obtain the spatial distribution structure of the cables. However, these methods often focus on the reconstruction and visualization of the cable structure, paying insufficient attention to the matching relationship between the extracted results and standard models. They lack effective image matching or feature matching mechanisms, making it difficult to accurately determine the deviation between the current cable routing and the expected design, thus affecting the accuracy of subsequent analyses. Furthermore, in the process of topology modeling and connection relationship analysis of cable structures, existing technologies typically rely solely on geometric proximity for connection inference, lacking a systematic verification of the rationality of connection relationships. This is especially problematic in complex wiring environments, making it difficult to promptly identify potential connection anomalies or structural deviations. Moreover, existing methods lack a unified judgment mechanism for the matching relationship of electrical characteristics between cable nodes, such as current and voltage parameters, leading to difficulties in timely detection of some abnormal states. During actual operation, cables within the battery compartment are also affected by changes in operating conditions, such as load fluctuations, temperature changes, and changes in spatial constraints. These factors can cause dynamic changes in the original connection relationships. However, existing technologies generally lack the ability to dynamically match and analyze cable connection relationships, making it difficult to continuously identify abnormal states during changes. At the same time, there is a lack of effective comprehensive evaluation and anomaly detection methods for problems such as unreasonable paths and local overload caused by matching deviations. Summary of the Invention

[0004] The purpose of this invention is to provide an image matching and anomaly detection method for cable layout inside an energy storage battery compartment, thereby solving the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an image matching and anomaly detection method for cable routing within an energy storage battery compartment, comprising: S1. Obtain 3D scanning data of the interior of the energy storage battery compartment, extract the initial point set of the cable layout through point cloud processing technology, obtain the geometric feature points of the cable distribution, and perform feature matching with the preset standard cable model based on the geometric feature points to obtain the initial matching result. S2. Use graph theory algorithms to perform connection analysis on geometric feature points, construct a preliminary model of the cable skeleton structure, determine the topological relationship of each cable segment, and perform consistency verification on the topological relationship based on the initial matching results; S3. The nodes in the preliminary model are grouped by clustering algorithm, it is determined whether the current matching between nodes meets the preset threshold, and potential abnormal nodes are identified by combining the matching deviation to obtain the optimized skeleton structure. S4. Based on the optimized skeleton structure, obtain the vector representation of the connection relationship. If the vector representation shows signs of connection abnormality or poor contact, apply the interpolation method to repair the connection points and determine a stable connection graph. S5. For stable connection diagrams, a dynamic adaptation attribute is introduced. The connection parameters are adjusted by simulating changes in working conditions to obtain an adaptive connection relationship. The connection relationship before and after adjustment is matched and analyzed to identify dynamic anomalies. S6. Extract voltage characteristic data from the adaptively enhanced connection relationship, determine the matching degree between voltage characteristics and spatial constraints. If the matching degree is lower than the threshold, it is determined to be a path abnormality and the cable path is reassigned to determine a collaborative layout scheme.

[0006] Preferably, S1 includes: Acquire panoramic 3D scan data of the interior of the energy storage battery compartment, and obtain a subset of point cloud to be processed by stripping background point cloud data of non-cable objects; Perform skeleton extraction on the subset of point cloud to be processed to generate an initial point set that reflects the three-dimensional spatial orientation and connectivity of the cable; Calculate the discrete curvature along the path direction of the initial point set. If the discrete curvature changes abruptly, it is identified as a key geometric feature point. The similarity between key geometric feature points and theoretical feature points in a pre-defined standard cable model is measured to construct a mapping matrix. The rotation and translation matrices are solved based on the mapping relationship matrix to obtain the initial matching result.

[0007] Preferably, S2 includes: Obtain the set of geometric feature points and calculate the Euclidean distance between point pairs. Construct a weighted adjacency matrix containing connection states and edge weights based on the Euclidean distance. The Prim algorithm is used to traverse and search the weighted adjacency matrix to generate a connected subgraph that is free of loops and has the minimum weight sum, thus obtaining a cable skeleton diagram that represents the cable geometry. The node degree of the nodes in the cable skeleton diagram is retrieved to determine the branch nodes. The paths connecting the branch nodes are extracted as cable segments, and the topological relationship of each cable segment is determined. Obtain a reference topology structure based on texture features, perform isomorphic mapping analysis between the topology relationship and the reference topology structure, and complete the consistency verification of the cable skeleton structure topology relationship based on the numerical value of the mapped structural deviation.

[0008] Preferably, S3 includes: The feature vectors of the nodes in the preliminary model are obtained and clustering calculations are performed to obtain the group sets belonging to each cluster center; The Euclidean distance between the node current values ​​and the mean within the group is calculated based on the group set, and a deviation vector representing the consistency of the current is generated. If the magnitude of the deviation vector is greater than the preset matching threshold, then the anomaly coefficient is calculated based on the deviation vector to mark potential abnormal nodes; Obtain the original adjacency matrix of the preliminary model, reset the connection weights of potential abnormal nodes in the original adjacency matrix to zero, and reconstruct the topological relationships to obtain the optimized skeleton structure.

[0009] Preferably, S4 includes: The optimized skeleton structure data is obtained and a topological adjacency matrix is ​​constructed. The joint connection features of the skeleton structure data are extracted through a graph convolutional network to obtain a spatial vector mapping that reflects the dependencies between nodes. The Euclidean distance and direction cosine between adjacent vectors are calculated based on the spatial vector mapping. The calculation results are compared with the preset geometric constraints to generate an anomaly detection score. If the anomaly detection score exceeds the preset threshold, an abnormal connection point with poor contact is identified in the spatial vector mapping. For abnormal connection points, temporal adjacent frame data and spatial neighbor node information are obtained. The cubic spline interpolation function is applied to fit the trajectory of the abnormal connection points to obtain a corrected vector that conforms to the continuity of motion. The corrected vectors are backfilled into the spatial vector map and the skeletal connectivity data is updated. The edge weights between nodes are reconstructed based on the repaired connectivity, thereby determining a stable connection graph.

[0010] Preferably, S5 includes: The initial stable connectivity map is obtained and time-varying perturbation factors are superimposed to obtain map data containing the characteristics of operating condition perturbation. A graph attention network is used to process graph data containing features of operating condition disturbances, and an adaptively enhanced connection topology is generated. The difference between the adaptively enhanced connectivity topology and the initial stable connectivity graph is calculated to construct a matching matrix, and the offset vector is extracted from the matching matrix; If the offset vector exceeds the fluctuation range, the location of the node that generated the offset vector is locked to identify dynamic anomalies.

[0011] Preferably, S6 includes: Analyze the connection topology data containing adaptive enhancement weights and calculate the voltage fluctuation characteristic sequence distributed along the connection path; The voltage fluctuation feature sequence is mapped to the geometric space model to construct a voltage space stress distribution map, and the matching degree between the voltage space stress distribution map and the preset spatial constraint model is calculated.

[0012] Preferably, S6 further includes: If the matching degree value is lower than the preset threshold, then search for alternative cable trajectories that avoid the stress area; The alternative cable trajectories are globally topologically integrated to determine a collaborative layout scheme.

[0013] Preferably, it also includes S7, which assesses overload risk indicators based on the coordinated layout scheme, reduces the risk value through iterative optimization based on the anomaly detection results, and outputs the cable anomaly detection results and the corresponding optimized cable layout model, specifically including: Obtain cable spatial coordinate data and rated current attributes in the collaborative cabling structure, and construct a thermal field distribution matrix that reflects the heat accumulation state of the cable. The set of overload risk values ​​is calculated based on the thermal field distribution matrix to quantify the potential failure probability.

[0014] Preferably, S7 further includes: The set of overload risk values ​​is input into the anomaly detection model to identify outliers and construct anomaly detection vectors that characterize the violation area. Based on the anomaly detection vector, the layout adjustment gradient data is calculated. The cable spatial coordinates are then iteratively updated using the layout adjustment gradient data until an optimized cable layout model with reduced risk value is output.

[0015] As can be seen from the above technical solution, the present invention has the following beneficial effects: This image matching and anomaly detection method for cable layout within an energy storage battery compartment achieves accurate identification and deviation analysis of cable layout status by introducing cable geometric feature extraction based on 3D point cloud data and a standard model feature matching mechanism. Simultaneously, it systematically constructs and verifies the cable topology by combining graph theory modeling and consistency verification methods, improving the reliability of connection relationships. Furthermore, it effectively identifies potential abnormal nodes through cluster analysis and current matching judgment, and enhances connection stability using vector representation and interpolation repair mechanisms. In addition, by introducing dynamic adaptive attributes and matching analysis, it achieves continuous monitoring and dynamic anomaly identification of cable connection relationships under changing operating conditions, and combines voltage characteristics and spatial constraint matching degree judgment to achieve accurate judgment and optimization adjustment of path anomalies. Finally, through comprehensive assessment and iterative optimization of overload risks, it not only outputs accurate anomaly detection results but also obtains optimized cable layout schemes, thereby significantly improving the safety, stability, and operational reliability of cable layout within the energy storage battery compartment. Attached Figure Description

[0016] Figure 1 This is a flowchart of the image matching and anomaly detection method of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] like Figure 1 As shown, the present invention provides a technical solution: an image matching and anomaly detection method for cable routing in an energy storage battery compartment, comprising: S1. Obtain 3D scanning data of the interior of the energy storage battery compartment, extract the initial point set of the cable layout through point cloud processing technology, obtain the geometric feature points of the cable distribution, and perform feature matching with the preset standard cable model based on the geometric feature points to obtain the initial matching result. S2. Use graph theory algorithms to perform connection analysis on geometric feature points, construct a preliminary model of the cable skeleton structure, determine the topological relationship of each cable segment, and perform consistency verification on the topological relationship based on the initial matching results; S3. The nodes in the preliminary model are grouped by clustering algorithm, it is determined whether the current matching between nodes meets the preset threshold, and potential abnormal nodes are identified by combining the matching deviation to obtain the optimized skeleton structure. S4. Based on the optimized skeleton structure, obtain the vector representation of the connection relationship. If the vector representation shows signs of connection abnormality or poor contact, apply the interpolation method to repair the connection points and determine a stable connection graph. S5. For stable connection diagrams, a dynamic adaptation attribute is introduced. The connection parameters are adjusted by simulating changes in working conditions to obtain an adaptive connection relationship. The connection relationship before and after adjustment is matched and analyzed to identify dynamic anomalies. S6. Extract voltage characteristic data from the adaptively enhanced connection relationship, determine the matching degree between voltage characteristics and spatial constraints. If the matching degree is lower than the threshold, it is determined to be a path abnormality and the cable path is reassigned to determine a collaborative layout scheme. S7. Based on the collaborative layout scheme, assess the overload risk indicators, and reduce the risk value through iterative optimization based on the anomaly detection results. Output the cable anomaly detection results and the corresponding optimized cable layout model.

[0019] This implementation addresses the technical challenges of complex cable layout, limited space, and dynamically changing operating states within energy storage battery compartments by constructing an anomaly detection method that integrates 3D perception, graph structure modeling, and multidimensional parameter analysis. First, high-precision 3D point cloud data of the energy storage battery compartment is acquired using laser scanning or structured light scanning equipment. Then, filtering, noise reduction, and region segmentation algorithms (such as point cloud processing methods based on RANSAC or Euclidean clustering) are used to extract point cloud sets related to the cables. Subsequently, curvature analysis and principal direction extraction are used to obtain the geometric feature points of the cables. Next, these geometric feature points are matched with a pre-established standard cable model (including standard routing, bending radius range, and connection node positions), using methods such as ICP (Iterative Closest Point) or feature descriptor matching to obtain initial matching results.

[0020] Building upon the above, geometric feature points are abstracted into graph nodes using graph theory methods. Edge connections are established based on spatial proximity and directional consistency to form a cable skeleton graph model. Simultaneously, the topological relationships between cable segments are determined, such as series, branch, or cross relationships. Using the initial matching results as constraints, the topology structure is validated for consistency, eliminating abnormal connections that do not conform to physical connection logic. Furthermore, nodes are grouped using clustering algorithms (such as K-means or DBSCAN), and combined with current distribution models or historical operating data, it is determined whether the current matching relationships between nodes meet preset thresholds, thereby identifying potential abnormal nodes and optimizing the skeleton structure.

[0021] Based on the optimized skeleton structure, the connection relationships are converted into vector representations, such as through node coordinate difference or graph embedding. When vector anomalies (such as abrupt changes in direction or distance anomalies) are detected, interpolation algorithms (such as spline interpolation or linear interpolation) are used to repair the connection points, generating a stable connection graph. Subsequently, by introducing dynamic adaptive attributes and combining simulated operating conditions such as temperature changes and load fluctuations, the connection parameters (such as impedance and current capacity) are dynamically adjusted, and dynamic anomalies are identified through before-and-after matching analysis.

[0022] Furthermore, voltage distribution characteristics are extracted from the adaptively enhanced connection relationships, and the matching degree is calculated in conjunction with spatial constraints (such as minimum bending radius and obstacle avoidance requirements). When the matching degree is lower than a threshold, it is determined to be a path anomaly, and the cable path is replanned based on path optimization algorithms (such as A* or Dijkstra's algorithm) to form a collaborative layout scheme. Finally, by constructing an overload risk assessment model (such as an evaluation function based on current density and thermal effects), the risk of the overall layout is quantified, and the risk value is reduced through iterative optimization strategies (such as genetic algorithms or gradient optimization), outputting the final anomaly detection results and the optimized cable layout model.

[0023] S1 includes acquiring panoramic 3D scan data of the interior of the energy storage battery compartment, obtaining a subset of point clouds to be processed by stripping background point cloud data of non-cable objects; performing skeleton extraction on the subset of point clouds to be processed to generate an initial set of points reflecting the 3D spatial orientation and connectivity of the cables; calculating discrete curvature along the path direction of the initial set of points, and establishing key geometric feature points if there is a sudden change in discrete curvature; measuring the similarity between the key geometric feature points and theoretical feature points in a preset standard cable model to construct a mapping matrix; and solving the rotation and translation matrix based on the mapping matrix to obtain the initial matching result.

[0024] In one implementation, the interior of the energy storage battery compartment is first continuously scanned using a panoramic 3D scanning device to obtain raw point cloud data containing spatial coordinate information. The scanning device preferably employs a lidar or structured light scanning device, with a scanning resolution set to no less than 1 mm to ensure complete preservation of cable details. Subsequently, preprocessing operations are performed on the raw point cloud data, including denoising and smoothing. Denoising employs a statistical filtering method based on neighborhood distance, comparing the average distance of each point to the average distance of its 20 nearest neighbors. Points exceeding 1.5 times the global average distance are identified as noise points and removed. This 1.5 times factor is an empirical coefficient based on the upper limit of the noise distribution range statistically obtained from multiple sets of actual scanning data. Smoothing uses a moving least squares method, constructing a fitted surface within the local neighborhood to reduce measurement errors.

[0025] After preprocessing, a background stripping operation is performed on the point cloud data. Specifically, non-cable structures are identified based on geometric features. By analyzing the distribution of point normal vectors and curvature characteristics, planar structures (such as battery boxes and bulkheads) are distinguished from linear structures. The normal vectors are obtained by performing principal component analysis on 30 neighboring points of each point. When the direction corresponding to the minimum eigenvalue is stable and the curvature is less than 0.01, it is determined to be a planar structure and removed. This 0.01 threshold is determined based on the upper limit of curvature statistics for planar regions in multiple batches of engineering data. For the remaining point cloud, slender continuous structures are retained through connectivity analysis, forming a subset of the point cloud to be processed.

[0026] After obtaining the subset of point cloud to be processed, a skeleton extraction operation is performed. The specific process is as follows: First, an adjacency graph of the point cloud is constructed, and each point is connected to its 10 nearest neighbors. The number of these 10 neighbors is determined based on the cable diameter and point cloud density. Then, an iterative shrinkage method is used to gradually remove boundary points, retaining only the central path points, thus obtaining an initial point set reflecting the cable's central direction. During the shrinkage process, the change in distance from a point to its local centroid is assessed. Shrinkage stops when the distance change is less than 0.2 mm; this 0.2 mm threshold is determined based on the upper limit of the scanning accuracy.

[0027] Along the path direction of the initial point set, a local path segment is constructed for every three adjacent points, and the degree of path turning is calculated. Specifically, the directional change of two consecutive path segments is compared. When the angle corresponding to the directional change exceeds 15 degrees, it is determined as a curvature abrupt change point. This 15-degree threshold is based on the statistics of the minimum bending angle during actual cable laying. All points that meet this condition are marked as key geometric feature points, and their spatial coordinates and adjacent connection relationships are recorded.

[0028] In the feature matching stage, the aforementioned key geometric feature points are compared one by one with the theoretical feature points in the preset standard cable model. Specifically, this is achieved as follows: First, distance matrices are established for both sets of feature points, and the spatial distance between any two points is calculated. Then, for each actual feature point, candidate points in the standard model with a distance difference of less than 5 mm are selected as matching objects. This 5 mm threshold is set based on the allowable error range for cable installation. Further comparison of directional consistency is made within the candidate set. When the directional difference is less than 10 degrees, it is determined as a valid matching pair. This 10-degree threshold is determined based on construction deviation statistics. A mapping relationship matrix is ​​established through the above screening process. This matrix records the correspondence between actual feature points and standard feature points, as well as the matching weights.

[0029] Based on the mapping matrix, the spatial alignment relationship is further solved. The specific process is as follows: First, the centroid positions of all matching point pairs are calculated by averaging the coordinates of all matching points. Then, all points are transformed into a coordinate system with the centroid as the origin. By minimizing the sum of distances between corresponding points, the spatial orientation is gradually adjusted to obtain the rotation relationship. After determining the rotation relationship, the translation relationship is obtained by comparing the overall offset between matching point pairs. During the solution process, any matching point with an error exceeding 3 mm is discarded. This 3 mm threshold is determined based on the matching accuracy requirements to avoid local anomalies affecting the overall result. Finally, a stable spatial alignment result, i.e., the initial matching result, is obtained.

[0030] S2 includes obtaining a set of geometric feature points and calculating the Euclidean distance between point pairs; constructing a weighted adjacency matrix containing connection states and edge weights based on the Euclidean distance; using the Prim algorithm to traverse and search the weighted adjacency matrix to generate a connected subgraph without loops and with the minimum weight sum, thus obtaining a cable skeleton graph representing the cable geometry; retrieving the node degree of nodes in the cable skeleton graph to determine branch nodes, extracting the paths connecting the branch nodes as cable segments, and determining the topological relationship of each cable segment; obtaining a reference topology structure based on texture features, performing isomorphic mapping analysis between the topological relationship and the reference topology structure, and verifying the consistency of the cable skeleton structure topology relationship based on the numerical value of the mapped structural deviation.

[0031] In the specific implementation process, the acquired set of geometric feature points is first systematically organized, each feature point is assigned a unique number, and its position coordinates in three-dimensional space are recorded simultaneously. Then, using any given feature point as a reference point, all other feature points are traversed sequentially, and the spatial distance between each pair of points is calculated. During the calculation, the position differences between two feature points in three spatial directions are obtained, the differences in each direction are squared, summed, and then the square root of the sum is taken to obtain the actual spatial distance between the two points. The above calculation is performed on all feature point combinations one by one, ultimately forming a complete set of point-to-point distances.

[0032] After obtaining all point-to-point distance data, the data is filtered. When the distance between two points is greater than 50 mm, no connection is established. This 50 mm threshold is determined based on statistical results of the actual cable spacing within the energy storage battery compartment. Specifically, after statistically analyzing multiple sets of actual engineering wiring data, the upper limit of the distance distribution between adjacent cables is selected as the decision boundary to ensure that only point pairs with a potential connection are retained. When the distance between two points is less than or equal to 50 mm, a potential connection is determined between the two points, and this distance value is recorded as the connection weight.

[0033] Based on the filtered point-pair relationships, a weighted adjacency matrix is ​​constructed. Specifically, a matrix structure with the same number of rows and columns is created according to the feature point numbering order, with each position in the matrix corresponding to a pair of feature points. When a connection exists between corresponding point pairs, their spatial distance is entered as a weight in the corresponding position of the matrix; when no connection exists, a value much greater than 50 millimeters is entered as an unreachable indicator, selected based on the maximum possible distance range to ensure that it is not mistakenly selected as a valid connection in subsequent calculations. This method completes the matrix representation of all point-pair relationships.

[0034] After constructing the weighted adjacency matrix, the minimum connected structure generation process is executed. First, an arbitrary feature point is chosen as the starting node and added to the selected set. Then, among all the connections between selected and unselected nodes, the weight values ​​of each edge are compared, and the edge with the smallest weight is selected, with the corresponding unselected node added to the selected set. During each selection process, candidate edges with a weight greater than 30 mm are discarded. This 30 mm threshold is calculated based on the maximum allowable spacing for continuous cable laying in a local area, ensuring the generated structure has actual physical continuity. This process continues until all nodes are included in the set, resulting in a structure with the minimum total connection path length and no redundant connections; this structure is the cable skeleton diagram.

[0035] After obtaining the cable skeleton diagram, the number of connections for each node in the diagram is counted one by one. Specifically, each node is traversed, and the number of edges directly connected to that node is counted, with the result used as the node's degree of connection. When the degree of connection of a node is greater than or equal to 3, the node is identified as a branch node. This criterion is derived from the basic condition for branch formation in graph structures, namely, a node that connects to three or more paths simultaneously constitutes a branch structure. After identifying branch nodes, each branch node is used as the starting point, and the connection path is traced outwards step by step. The process terminates when the path extends to another branch node or an end node with a degree of connection of 1. All nodes traversed in the path are recorded in sequence to form a complete cable segment. The starting node, ending node, and total path length of the segment are also recorded, thereby establishing a complete cable topology.

[0036] After the topology relationships are established, a reference topology is obtained. This reference topology is formed based on extracted cable surface texture information. Specifically, an image of the cable's surface texture is acquired using an image acquisition device, and the image undergoes grayscale processing and edge enhancement to extract texture direction information. This information is then combined with the connection relationships at corresponding positions in historical cabling data to construct a standard topology model. Subsequently, a structural matching process is performed between the current cable skeleton structure and the reference topology.

[0037] During the matching process, the nodes in both sets of topologies are first uniformly numbered to provide a basis for one-to-one comparison. Then, the connection relationships between corresponding nodes are compared pair by pair. For any pair of nodes, if a connection exists in the current structure but not in the reference structure, this difference is recorded as a structural deviation; conversely, the same process is performed. Simultaneously, the lengths of each corresponding path are compared. When the path length difference is greater than 10 millimeters, it is also recorded as a deviation. This 10-millimeter threshold is statistically determined based on the actual cable installation error range and is used to distinguish between normal construction errors and abnormal deployment conditions.

[0038] After comparing all nodes and paths, the number of deviations is accumulated to obtain the overall structural deviation value. When this value is greater than 5, it is determined that the current cable skeleton structure is inconsistent with the reference topology. The criterion of 5 is derived from the statistical analysis results of multiple sets of normal operating samples, and it represents the maximum allowable range of structural deviation under normal conditions.

[0039] S3 includes obtaining the feature vectors of nodes in the preliminary model and performing clustering calculations to obtain group sets belonging to each cluster center; calculating the Euclidean distance between the node current value and the mean within the group based on the group set to generate a deviation vector representing the consistency of the current; if the magnitude of the deviation vector is greater than a preset matching threshold, calculating the anomaly coefficient based on the deviation vector to mark potential abnormal nodes; obtaining the original adjacency matrix of the preliminary model, resetting the connection weights of potential abnormal nodes in the original adjacency matrix to zero and reconstructing the topological relationship to obtain the optimized skeleton structure.

[0040] In the specific implementation process, data is first collected and organized for each node in the preliminary model to construct a complete feature description for each node. Specifically, the spatial coordinate information of each node is extracted, and its spatial position is determined by reading the node's position data in three-dimensional space; at the same time, the number of connections between the node and other nodes in the skeleton structure is counted, and the number of connecting edges is used as a structural feature; furthermore, the current measurement value of the corresponding cable of the node is obtained. The current data is collected by a current sensing device deployed on the cable and matched with the node position. After completing the above information collection, the spatial position, number of connections, and current value are arranged in a unified order to form the original feature data sequence of the node.

[0041] After obtaining the raw feature data of all nodes, a unified scaling process is performed on different data types. Specifically, the maximum and minimum values ​​of all nodes in the three dimensions of spatial location, number of connections, and current value are calculated. Then, the values ​​of each node in the corresponding dimension are mapped to a range, ensuring that all values ​​are distributed between 0 and 1. This range is determined based on the requirements of unified multi-dimensional feature processing and is used to eliminate the influence between different units of measurement. During the mapping process, for a certain dimension, when the difference between the maximum and minimum values ​​is less than 0.001, the values ​​of all nodes in that dimension are uniformly processed to 0.5. This 0.001 criterion is determined based on the lower limit of sensor measurement accuracy to avoid computational instability caused by excessive data concentration. After completing the above processing, the standardized node feature vector is obtained.

[0042] After constructing the feature vectors, clustering is performed on all nodes. The specific process is as follows: First, three nodes with significant feature differences are selected as initial classification centers. This number is set to three, based on the classification results of the cable operation status within the energy storage battery compartment, corresponding to normal, transitional, and abnormal states. Then, the remaining nodes are classified one by one, calculating the degree of difference between each node and each classification center. The difference between the node's feature value and the classification center's feature value is compared dimension by dimension, and all differences are summed. The classification center with the smallest difference is selected as the node's category. After one round of classification, the features of all nodes in each category are re-statistically analyzed, and the average value is calculated dimension by dimension to update the classification center positions. The above classification and center update process is repeated. Iteration stops when the change in the classification center is less than 0.01 in two consecutive updates. This 0.01 threshold is determined based on the stable range of feature changes after standardization, ensuring convergence of the classification results. Finally, multiple group sets are obtained, each corresponding to one type of operating state.

[0043] After obtaining the grouped sets, a consistency analysis is performed on the current data of the nodes within each group. Specifically, for each group, the current values ​​of all nodes within the group are first summed, and then the sum is divided by the number of nodes to obtain the average current value of the group. Subsequently, for each node within the group, the difference between its current value and the average current value of the group is calculated and recorded as the current deviation. To form a unified evaluation standard, the deviations of each node across all feature dimensions are summarized. Specifically, the deviation values ​​of each dimension are squared, summed, and then the square root of the sum is taken to obtain the overall deviation of the node.

[0044] During the deviation assessment process, the overall deviation of each node is compared with a preset matching threshold. When the overall deviation is greater than 0.2, the node is determined to be outside the normal range. This 0.2 threshold is determined based on the statistical upper limit of node deviation under a large number of normal operating conditions and is used to distinguish between normal fluctuations and abnormal fluctuations. For nodes that have been determined to be outside the range, their degree of abnormality is further calculated. Specifically, the ratio of the node's overall deviation to the average deviation of all nodes in its group is calculated. When this ratio is greater than 1.5, the node is marked as a potential abnormal node. This 1.5 criterion is derived from the boundary statistical results of the deviation ratio between abnormal samples and normal samples, and is used to improve the accuracy of abnormality identification.

[0045] After identifying potential anomalous nodes, the original adjacency matrix corresponding to the preliminary model is obtained. Specifically, the connection relationship data between nodes is read in order of node number, and the nodes marked as potential anomalous are located one by one. During processing, a connection relationship adjustment operation is performed on each potential anomalous node, setting all connection weight values ​​between it and other nodes to 0, indicating that the node no longer participates in the connection relationships of the current topology. After processing all anomalous nodes, the connection relationships of the remaining nodes are reorganized. Specifically, all valid connection relationships between nodes are traversed, and the connectivity structure is reconstructed according to the connection paths, ensuring that all remaining nodes remain connected, thereby forming a new cable skeleton structure.

[0046] S4 includes acquiring optimized skeleton structure data and constructing a topological adjacency matrix; extracting joint connection features from the skeleton structure data using a graph convolutional network to obtain a spatial vector mapping reflecting the dependencies between nodes; calculating the Euclidean distance and direction cosine between adjacent vectors based on the spatial vector mapping; comparing the calculation results with preset geometric constraints to generate anomaly detection scores; if the anomaly detection score exceeds a preset threshold, identifying abnormal connection points in the spatial vector mapping that show signs of poor contact; acquiring temporal adjacent frame data and spatial neighborhood node information for the abnormal connection points; applying a cubic spline interpolation function to fit the trajectory of the abnormal connection points to obtain corrected vectors that conform to motion continuity; backfilling the corrected vectors into the spatial vector mapping and updating the skeleton connectivity data; reconstructing the edge weights between nodes based on the repaired connectivity to determine a stable connection graph.

[0047] In the specific implementation process, the optimized skeleton structure data is first parsed node by node. All nodes are arranged in a unified numbering order, and the spatial coordinate information of each node and the connection relationships between nodes are extracted one by one. Based on the above data, a topological adjacency matrix is ​​constructed. Specifically, using the total number of nodes as the matrix dimension, any two nodes are traversed and judged. If there is a connection between the two nodes, it is marked as 1 in the corresponding matrix position; otherwise, it is marked as 0, thus forming a basic connection structure. At the same time, for each pair of nodes with a connection relationship, their spatial distance is calculated, and this distance is recorded as an additional weight in an independent data structure for subsequent feature calculation.

[0048] After constructing the topological adjacency matrix, initial feature data is extracted for each node. This feature data includes the node's spatial location and the distribution of its neighboring nodes. Then, multi-layer neighborhood aggregation processing is performed on the node features. Specifically: First, for each node, information on all its directly connected neighboring nodes is collected, and the feature data of these neighboring nodes is summarized. During the summarization process, each neighboring node is assigned a different weight, determined based on the spatial distance between nodes. A weight of 1 is assigned when the distance is less than 20 mm, and a weight of 0.5 is assigned when the distance is between 20 mm and 40 mm. This 20 mm and 40 mm division is determined based on the statistical distribution of distances between locally tight connections and general connections in cables. Then, the features of all neighboring nodes are weighted and summed, and then fused with the current node's own features to obtain updated node features. This process, as a feature aggregation layer, is executed three times. The three layers are determined based on the propagation range of cable connections, covering first-order, second-order, and third-order adjacency relationships to obtain a stable structural representation. Finally, each node forms a comprehensive spatial vector, representing its position and dependencies within the overall structure.

[0049] After obtaining the spatial vector mapping, each pair of nodes with connections is analyzed individually. Specifically, for any pair of adjacent nodes, their spatial vector data is read, and the numerical differences between the two vectors in each dimension are calculated. The squares of these differences are then summed and the square root is taken to obtain the vector distance between the two nodes. When this distance is greater than 0.3, it is considered an anomaly. This 0.3 threshold is determined based on the statistical maximum distance in the normal connection state within the normal vector distribution range. Simultaneously, the direction of change of the two node vectors in each dimension is compared one by one. When most dimensions show opposite trends, the direction consistency is considered low; when the direction consistency is less than 0.8, it is considered an anomaly. This 0.8 threshold is determined based on the statistical results of vector direction consistency under normal connection conditions.

[0050] After determining the distance and direction, an anomaly detection score is generated for each pair of nodes. Specifically: when both distance and direction anomalies are met, the score for that node pair is set to 2; when only one of the anomaly conditions is met, the score is set to 1; and when neither condition is met, the score is set to 0. Subsequently, all relevant scores for each connection point are accumulated. When the cumulative score is greater than or equal to 2, the connection point is determined to have signs of poor contact. This judgment criterion is determined based on statistical analysis of multiple sets of actual abnormal connection data to eliminate interference from occasional fluctuations.

[0051] After identifying the abnormal connection point, continuous data for that connection point in the time dimension is acquired. Specifically, two frames of data are extracted before and after the node, for a total of five frames. The number of five frames is determined based on matching the system sampling frequency with the cable change rate to fully reflect the short-term change process of the node. Simultaneously, a reference set of neighboring nodes within a spatial distance of no more than 40 millimeters is selected, with this 40-millimeter range determined statistically based on the influence range of the cable's local structure. The time-series data and spatial neighborhood data are jointly processed to form the data foundation for trajectory correction.

[0052] During trajectory fitting, the position data of the node in each frame is arranged chronologically, and the spatial distribution of neighboring nodes is combined to analyze the trend of node position changes segment by segment. Specifically, the position change between adjacent time points is calculated, and the change is smoothed to maintain the continuity of the overall trend. For time points with abnormal offsets, the position of the time point is corrected by referring to the position change trends of the preceding and following time points and the spatial distribution of neighboring nodes, so that its change amplitude is consistent with the overall trend, thus forming a continuous and smooth motion trajectory. Based on this trajectory, the spatial vector representation of the node is regenerated to obtain the corrected vector data.

[0053] After vector correction, the correction results are updated in the original spatial vector mapping, and the connection relationships of the corresponding nodes are updated synchronously. Specifically, the spatial distance between the node and its neighboring nodes is recalculated and used as the new connection weight; at the same time, a consistency check is performed on all related connection relationships to ensure that there are no broken or duplicate connections. Finally, based on the updated connection relationship data, the edge weight structure between nodes is reconstructed to form a stable connection graph.

[0054] S5 includes acquiring an initial stable connectivity graph and superimposing a time-varying disturbance factor to obtain graph data containing operating condition disturbance features; using a graph attention network to process the graph data containing operating condition disturbance features to generate an adaptively enhanced connectivity topology; calculating the difference between the adaptively enhanced connectivity topology and the initial stable connectivity graph to construct a matching matrix, and extracting an offset vector from the matching matrix; if the offset vector exceeds the fluctuation range, locking the node position that generated the offset vector to identify dynamic anomalies.

[0055] In the specific implementation process, an initial stable connectivity graph is first obtained. All nodes and their connections in this graph are then analyzed item by item, and a complete dataset of nodes and edges is established. Specifically, the spatial location of each node, the number of connected nodes, and the weight of the corresponding edge are read one by one. Simultaneously, the spatial distance and connection stability index of each edge are recorded. After completing the basic data processing, a time-varying perturbation factor is introduced to dynamically adjust the original connectivity relationships node by node and edge by edge.

[0056] During the disturbance superposition process, temperature factors are processed first. Specifically, real-time temperature data for each node's corresponding location is acquired, and the difference between the current temperature and the reference temperature is calculated. When this difference exceeds 5 degrees Celsius, the stability parameters of all connected edges related to that node are reduced by 0.1. The 5-degree Celsius threshold is determined based on the statistical upper limit of temperature fluctuations during the long-term operation of the energy storage battery compartment, and the 0.1 adjustment is determined based on experimental statistical results of the impact of temperature changes on the conductivity of materials. Next, current disturbances are processed. Specifically, the real-time current value of each cable is acquired and compared with its rated current. When the current change exceeds 10% of the rated value, the spatial distance of the corresponding connected edge is corrected by increasing the distance by 2 millimeters. The 10% value is determined based on the allowable fluctuation range for safe operation of the electrical system, and the 2-millimeter adjustment is determined based on statistical results of the impact of current changes on conductor thermal expansion. Further processing of vibration factors involves obtaining the vibration amplitude at the node location through a vibration sensing device. When the vibration amplitude exceeds 0.5 mm, the corresponding node location is disturbed and offset, causing it to shift randomly with the same amplitude from its original location. This 0.5 mm threshold is determined based on the equipment's vibration resistance design standards and long-term operating data.

[0057] After completing the multi-source perturbation superposition, a graph data containing the characteristics of the operating condition perturbation is formed. Then, a structural analysis based on an attention mechanism is performed on this graph data. Specifically, for each node, all its directly connected neighboring nodes are first identified, and the correlation degree between the current node and each neighboring node is calculated one by one. During the calculation, nodes with higher connection weights and smaller perturbation amplitudes are assigned higher attention values, while nodes with larger perturbation amplitudes have their attention values ​​reduced. Specifically, a weight of 1 is assigned when the connection edge weight is higher than 0.8 times the original weight; a weight of 0.6 is assigned when the connection edge weight is between 0.5 and 0.8 times; and a weight of 0.3 is assigned when the connection edge weight is lower than 0.5 times. The division between 0.8 and 0.5 is determined based on the statistical boundary of connection stability under normal and abnormal states. Then, the feature data of all neighboring nodes are weighted and accumulated according to the above weights, and then fused and updated with the current node's own features. This process is performed on all nodes one by one, ultimately obtaining a connection relationship topology structure adapted to the current perturbation state.

[0058] After obtaining the adaptively enhanced connectivity, it is compared item by item with the initial stable connectivity graph. Specifically, a matching matrix consistent with the number of nodes is established, and the connectivity changes of each node in the two graphs are calculated. First, the number of connecting edges is compared. When the change in the number of connections of a node exceeds 1, it is recorded as a structural offset. This criterion is determined based on the statistical results of the stability of connectivity under normal operating conditions. Second, the connection weights are compared. When the change in weight exceeds 0.2, it is recorded as a weight offset. This 0.2 threshold is determined based on the statistical analysis of the normal fluctuation range of connection weights under disturbance conditions. Third, the spatial positions of nodes are compared. When the position offset distance exceeds 3 mm, it is recorded as a position offset. This 3 mm threshold is determined based on a comprehensive consideration of equipment installation errors and operating vibration range.

[0059] After determining the three types of offsets mentioned above, the offset of each node is accumulated to form the offset vector corresponding to the node. Specifically, structural offset, weight offset, and position offset are counted separately, and the three results are summed to obtain the overall offset value of the node. Then, this value is judged within a range. When the offset value is greater than 1, it is determined to be outside the normal fluctuation range. The criterion of 1 is determined based on the maximum cumulative offset value in multiple sets of normal operation data, used to distinguish between normal disturbances and abnormal changes. For nodes determined to be outside the fluctuation range, their specific location in the overall connectivity diagram is further located and marked as dynamic abnormal nodes. Simultaneously, the disturbance source and offset type corresponding to this node are recorded.

[0060] S6 includes parsing the connection topology data containing adaptive enhancement weights, calculating the voltage fluctuation feature sequence distributed along the connection path; mapping the voltage fluctuation feature sequence to the geometric space model to construct a voltage space stress distribution map, calculating the matching degree value between the voltage space stress distribution map and the preset spatial constraint model; if the matching degree value is lower than the preset threshold, searching for alternative cable trajectories that avoid the stress area; and performing global topology fusion on the alternative cable trajectories to determine a collaborative layout scheme.

[0061] In the specific implementation process, the connection topology data containing adaptive enhancement weights is first processed through path-by-path parsing. Specifically, according to the connection relationships, starting from the starting node, the connection sequence recorded in the topology is traversed sequentially to the ending node. Nodes on each complete path are numbered sequentially, and their spatial locations and corresponding voltage values ​​are recorded. Voltage values ​​are acquired through voltage acquisition devices deployed at cable node locations and time-synchronized to ensure the comparability of voltage data from different nodes at the same time. During path traversal, the voltage values ​​between adjacent nodes are calculated segment by segment, i.e., the voltage value of the current node is subtracted from the voltage value of the previous node to obtain the voltage change of that path segment. These changes are recorded sequentially according to the path order, thus forming a complete voltage change sequence.

[0062] After obtaining the voltage change sequence, a smoothing process is performed. Specifically, three consecutive nodes are grouped together, and the voltage changes within each group are accumulated and averaged. This average value is then used as the smoothing result for the intermediate node. The processing range for these three nodes is determined based on the voltage sampling frequency and the system response time, used to eliminate interference from short-term fluctuations. For the starting and ending nodes of the path, the average of two adjacent nodes is used to ensure the continuity of boundary data. After the smoothing process, a stable voltage fluctuation characteristic sequence is obtained.

[0063] After obtaining a stable sequence, it is mapped onto a three-dimensional geometric space model. Specifically, the spatial coordinates of each node are used as a positioning reference, and the corresponding voltage change is assigned to that spatial location, thus forming discretely distributed voltage stress points in space. Then, the uncovered areas in the space are supplemented: for each spatial point, a set of nodes within a 30mm radius is searched, the voltage changes of these nodes are weighted and averaged, and then assigned to that spatial point, thus achieving a continuous spatial distribution. This 30mm range is determined based on the statistical results of the cable's local electrical influence radius. Through the above processing, a voltage spatial stress distribution map covering the entire deployment area is formed.

[0064] After constructing the voltage spatial stress distribution map, it is compared and analyzed point by point with a preset spatial constraint model. The spatial constraint model predefines the allowable voltage stress range, minimum safe distance, and passable path area for each spatial location. Specifically, for each location point in the space, its voltage stress value is read and compared with the corresponding allowable range. When the voltage stress value exceeds the allowable range, the location is marked as a mismatch point. After traversing all spatial points, the number of mismatch points is counted, and their proportion in all spatial points is calculated to obtain the overall matching degree value. When the matching degree value is lower than 0.85, it is determined that the current layout does not meet the spatial constraint requirements. This 0.85 threshold is determined based on the statistical lower limit of matching degree under a large number of normal operating conditions and is used to distinguish between reasonable layout and layout states with risks.

[0065] When the matching degree is deemed insufficient, an alternative cable path search process is executed. Specifically, the starting and ending nodes of the original path are used as search boundaries to reconstruct the path in a 3D spatial model. During the search, areas where the voltage stress value exceeds the allowable range are first marked as impassable areas, and areas with a distance of less than 20 mm from other equipment or structures are also marked as restricted areas. This 20 mm threshold is determined according to electrical safety clearance standards. Subsequently, in the remaining passable space, candidate path nodes are gradually expanded outward from the starting node. In each step of expansion, nodes with lower voltage stress and consistent with the current path direction are preferentially selected as the next connection point, and the cumulative path length is recorded. When the path successfully connects to the ending node, a complete alternative path is formed.

[0066] After obtaining alternative paths, a global topology fusion process is performed. Specifically, all nodes in the alternative paths are uniformly numbered with nodes in the original topology, and the corresponding connection relationships are updated. For sections where the original and alternative paths overlap or conflict, their voltage stress values ​​and path lengths are compared one by one, prioritizing connections with lower voltage stress and smoother paths. Subsequently, the connection weights between nodes are recalculated for the updated topology. The weights are determined based on the spatial distance between nodes and the stability of voltage changes. A consistency check is performed on the overall topology to ensure that there are no broken connections or redundant paths. The final result is a collaborative cable layout scheme that meets spatial constraints and has a balanced voltage distribution.

[0067] S7 includes acquiring cable spatial coordinate data and rated current attributes in the collaborative cabling structure, constructing a thermal field distribution matrix reflecting the heat accumulation state of the cables; calculating an overload risk value set based on the thermal field distribution matrix to quantify the potential failure probability; inputting the overload risk value set into the anomaly detection model to identify outliers and constructing anomaly detection vectors representing non-compliant areas; calculating layout adjustment gradient data based on the anomaly detection vectors, iteratively updating the cable spatial coordinates using the layout adjustment gradient data, until an optimized cable layout model with reduced risk values ​​is output.

[0068] In the specific implementation process, all cables in the collaborative cabling structure are first processed segment by segment. Specifically, the three-dimensional spatial coordinate data of each cable is read sequentially, and the coordinate points are sorted according to the actual connection order of the cables to form a continuous spatial path. Then, the rated current attribute of each cable is obtained and matched with its spatial path segment by segment. In the path processing stage, each cable is discretized along its length, and the continuous path is divided into several discrete segments with a length of 10 mm. This 10 mm division scale is determined based on the balance between thermal diffusion spatial resolution and computational load. It is selected after analyzing multiple sets of actual cabling data, so that each discrete segment can reflect local thermal changes without causing excessive computational scale.

[0069] After discretization, the heat generation of each cable segment is calculated. Specifically, the actual operating current of the cable segment is read and compared with its rated current. The degree of difference between the two determines the heat generation intensity of the cable segment. When the actual current is close to the rated current, the cable segment is marked as a high-heat segment; when the actual current is below the rated current by a certain range, it is marked as a medium-heat segment or a low-heat segment. Subsequently, the heat generation intensity of each cable segment is associated with its spatial location and mapped onto a three-dimensional spatial grid. The spatial grid is divided into cubic units with a side length of 20 mm. This 20 mm size is determined based on the statistical results of the heat conduction influence range between cables, ensuring that the unit size can cover the thermal coupling area between adjacent cables. During the mapping process, the heat generation intensity of all cable segments located in the same grid unit is accumulated to obtain the heat value corresponding to each grid unit, ultimately forming a complete thermal field distribution matrix.

[0070] After obtaining the thermal field distribution matrix, statistical analysis is performed on the heat values ​​of all grid cells. Specifically, the heat values ​​of all grid cells are first summed and divided by the total number of grid cells to obtain the overall average heat level. Then, for each grid cell, the difference between its heat value and the overall average heat is calculated. When the heat value of a grid cell exceeds 1.5 times the average heat, the grid cell is marked as a high-heat region. This 1.5 times threshold is determined based on the statistical results of the upper limit of heat distribution under multiple sets of normal operating conditions and is used to distinguish between normal heat distribution and abnormal heat accumulation.

[0071] After identifying high-heat areas, an overload risk assessment is performed on each cable segment. Specifically, this involves determining the grid cell containing the cable segment, obtaining the heat level of that grid cell, and calculating the minimum spatial distance between the cable segment and other surrounding cable segments. When a cable segment is in a high-heat area and the distance to adjacent cables is less than 15 mm, its risk value is marked as high risk; this 15 mm threshold is determined based on cable safety spacing design standards. When only one condition is met, it is marked as medium risk; when neither condition is met, it is marked as low risk. By performing the above assessment on all cable segments one by one, a complete set of overload risk values ​​is formed.

[0072] After obtaining the risk value set, anomaly identification processing is performed. Specifically, for each cable segment, the five nearest cable segments in its spatial neighborhood are selected as a reference set. The number of these five neighboring cable segments is determined based on local wiring density statistics to reflect local environmental characteristics. Then, the difference between the risk value of this cable segment and the average risk value of its neighboring set is calculated. When this difference exceeds twice the average of the neighboring set, the cable segment is identified as an anomaly. This twice-threshold is determined based on the statistical results of abnormal heat accumulation samples. The spatial locations and corresponding risk levels of all anomalies are integrated to form an anomaly detection vector, which is used to characterize the distribution of areas with violation risks in space.

[0073] After anomaly detection, cable layout optimization is performed. Specifically, for each abnormal cable segment, its current position in space is first determined. Then, multiple candidate adjustment directions are searched within a 25mm radius around it. This 25mm range is determined based on statistical analysis of feasible local cabling adjustments. For each candidate direction, the heat change at the corresponding location is calculated, and the direction with the largest heat decrease is selected as the adjustment direction. Subsequently, the adjustment distance is determined based on the current risk level. When the cable segment is in a high-risk state, the adjustment distance is set to 5mm; when it is in a medium-risk state, the adjustment distance is set to 2mm. These 5mm and 2mm values ​​are determined based on the cable's movable range and structural constraints in the actual project.

[0074] After each position adjustment, the thermal field distribution is recalculated based on the updated cable spatial coordinates, and the risk values ​​of all cable segments are reassessed. This adjustment and assessment process is repeated until the overall risk value decreases by less than 0.05 in two consecutive iterations. This 0.05 threshold is determined based on the statistical results of the risk change convergence trend and is used to determine if the optimization process has reached a stable state. Finally, the updated cable spatial coordinate data and the corresponding layout model are output.

[0075] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for image matching and anomaly detection of cable routing within an energy storage battery compartment, characterized in that, include: S1. Obtain 3D scanning data of the interior of the energy storage battery compartment, extract the initial point set of the cable layout through point cloud processing technology, obtain the geometric feature points of the cable distribution, and perform feature matching with the preset standard cable model based on the geometric feature points to obtain the initial matching result. S2. Use graph theory algorithms to perform connection analysis on geometric feature points, construct a preliminary model of the cable skeleton structure, determine the topological relationship of each cable segment, and perform consistency verification on the topological relationship based on the initial matching results; S3. The nodes in the preliminary model are grouped by clustering algorithm, it is determined whether the current matching between nodes meets the preset threshold, and potential abnormal nodes are identified by combining the matching deviation to obtain the optimized skeleton structure. S4. Based on the optimized skeleton structure, obtain the vector representation of the connection relationship. If the vector representation shows signs of connection abnormality or poor contact, apply the interpolation method to repair the connection points and determine a stable connection graph. S5. For stable connection diagrams, a dynamic adaptation attribute is introduced. The connection parameters are adjusted by simulating changes in working conditions to obtain an adaptive connection relationship. The connection relationship before and after adjustment is matched and analyzed to identify dynamic anomalies. S6. Extract voltage characteristic data from the adaptively enhanced connection relationship, determine the matching degree between voltage characteristics and spatial constraints. If the matching degree is lower than the threshold, it is determined to be a path abnormality and the cable path is reassigned to determine a collaborative layout scheme.

2. The image matching and anomaly detection method for cable layout in an energy storage battery compartment according to claim 1, characterized in that: S1 includes: Acquire panoramic 3D scan data of the interior of the energy storage battery compartment, and obtain a subset of point cloud to be processed by stripping background point cloud data of non-cable objects; Perform skeleton extraction on the subset of point cloud to be processed to generate an initial point set that reflects the three-dimensional spatial orientation and connectivity of the cable; Calculate the discrete curvature along the path direction of the initial point set. If the discrete curvature changes abruptly, it is identified as a key geometric feature point. The similarity between key geometric feature points and theoretical feature points in a pre-defined standard cable model is measured to construct a mapping matrix. The rotation and translation matrices are solved based on the mapping relationship matrix to obtain the initial matching result.

3. The image matching and anomaly detection method for cable routing in an energy storage battery compartment according to claim 1, characterized in that: S2 includes: Obtain the set of geometric feature points and calculate the Euclidean distance between point pairs. Construct a weighted adjacency matrix containing connection states and edge weights based on the Euclidean distance. The Prim algorithm is used to traverse and search the weighted adjacency matrix to generate a connected subgraph that is free of loops and has the minimum weight sum, thus obtaining a cable skeleton diagram that represents the cable geometry. The node degree of the nodes in the cable skeleton diagram is retrieved to determine the branch nodes. The paths connecting the branch nodes are extracted as cable segments, and the topological relationship of each cable segment is determined. Obtain a reference topology structure based on texture features, perform isomorphic mapping analysis between the topology relationship and the reference topology structure, and complete the consistency verification of the cable skeleton structure topology relationship based on the numerical value of the mapped structural deviation.

4. The image matching and anomaly detection method for cable layout in an energy storage battery compartment according to claim 1, characterized in that: S3 includes: The feature vectors of the nodes in the preliminary model are obtained and clustering calculations are performed to obtain the group sets belonging to each cluster center; The Euclidean distance between the node current values ​​and the mean within the group is calculated based on the group set, and a deviation vector representing the consistency of the current is generated. If the magnitude of the deviation vector is greater than the preset matching threshold, then the anomaly coefficient is calculated based on the deviation vector to mark potential abnormal nodes; Obtain the original adjacency matrix of the preliminary model, reset the connection weights of potential abnormal nodes in the original adjacency matrix to zero, and reconstruct the topological relationships to obtain the optimized skeleton structure.

5. The image matching and anomaly detection method for cable layout in an energy storage battery compartment according to claim 1, characterized in that: S4 includes: The optimized skeleton structure data is obtained and a topological adjacency matrix is ​​constructed. The joint connection features of the skeleton structure data are extracted through a graph convolutional network to obtain a spatial vector mapping that reflects the dependencies between nodes. The Euclidean distance and direction cosine between adjacent vectors are calculated based on the spatial vector mapping. The calculation results are compared with the preset geometric constraints to generate an anomaly detection score. If the anomaly detection score exceeds the preset threshold, an abnormal connection point with poor contact is identified in the spatial vector mapping. For abnormal connection points, temporal adjacent frame data and spatial neighbor node information are obtained. The cubic spline interpolation function is applied to fit the trajectory of the abnormal connection points to obtain a corrected vector that conforms to the continuity of motion. The corrected vectors are backfilled into the spatial vector map and the skeletal connectivity data is updated. The edge weights between nodes are reconstructed based on the repaired connectivity, thereby determining a stable connection graph.

6. The image matching and anomaly detection method for cable routing in an energy storage battery compartment according to claim 1, characterized in that: S5 includes: The initial stable connectivity map is obtained and time-varying perturbation factors are superimposed to obtain map data containing the characteristics of operating condition perturbation. A graph attention network is used to process graph data containing features of operating condition disturbances, and an adaptively enhanced connection topology is generated. The difference between the adaptively enhanced connectivity topology and the initial stable connectivity graph is calculated to construct a matching matrix, and the offset vector is extracted from the matching matrix; If the offset vector exceeds the fluctuation range, the location of the node that generated the offset vector is locked to identify dynamic anomalies.

7. The image matching and anomaly detection method for cable layout in an energy storage battery compartment according to claim 1, characterized in that: S6 includes: Analyze the connection topology data containing adaptive enhancement weights and calculate the voltage fluctuation characteristic sequence distributed along the connection path; The voltage fluctuation feature sequence is mapped to the geometric space model to construct a voltage space stress distribution map, and the matching degree between the voltage space stress distribution map and the preset spatial constraint model is calculated.

8. The image matching and anomaly detection method for cable routing in an energy storage battery compartment according to claim 7, characterized in that: S6 further includes: If the matching degree value is lower than the preset threshold, then search for alternative cable trajectories that avoid the stress area; The alternative cable trajectories are globally topologically integrated to determine a collaborative layout scheme.

9. The image matching and anomaly detection method for cable layout in an energy storage battery compartment according to claim 1, characterized in that, It also includes S7, which assesses overload risk indicators based on the collaborative layout scheme, and reduces the risk value through iterative optimization based on the anomaly detection results, outputting the cable anomaly detection results and the corresponding optimized cable layout model, specifically including: Obtain cable spatial coordinate data and rated current attributes in the collaborative cabling structure, and construct a thermal field distribution matrix that reflects the heat accumulation state of the cable. The set of overload risk values ​​is calculated based on the thermal field distribution matrix to quantify the potential failure probability.

10. The image matching and anomaly detection method for cable layout in an energy storage battery compartment according to claim 9, characterized in that: The S7 also includes: The set of overload risk values ​​is input into the anomaly detection model to identify outliers and construct anomaly detection vectors that characterize the violation area. Based on the anomaly detection vector, the layout adjustment gradient data is calculated. The cable spatial coordinates are then iteratively updated using the layout adjustment gradient data until an optimized cable layout model with reduced risk value is output.

Images