Defect hidden danger data diagnosis method based on distribution network SVG single line diagram

Abstract

The invention discloses a defect hidden danger data diagnosis method based on a distribution network SVG single line diagram. The method comprises the steps that basic analysis is conducted on the input SVG single line diagram, basic geometric figure elements are extracted and normalized, a topological connection relation is preprocessed, and a multi-level index structure is constructed; a dynamic topology path reasoning mechanism is constructed, and reasoning and self-adaptive optimization of the topological relation of the electrical equipment are realized through path consistency calculation, topological entropy minimization and a dynamic topology updating strategy; a self-adaptive implicit relational knowledge graph is constructed, and semantic association between the electrical equipment is established through explicit topological relation fusion, implicit function dependence reasoning and dynamic semantic embedding optimization; and constructing a multi-modal time sequence feature vector, calculating the hidden danger risk of the equipment by adopting a space-time defect probability estimation model, and giving an alarm through a dynamic risk early warning strategy to realize prediction and early warning of the hidden danger of the defects of the electrical equipment. According to the method, the problems of semantic understanding, topological reasoning, state inference and defect prediction of the SVG single line diagram are effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution networks, and in particular to a method for diagnosing defect and hidden danger data based on a distribution network SVG single-line diagram. Background Art

[0002] The distribution network is a vital component of the power system, and its safe and stable operation is directly related to the reliable supply of electricity. To achieve refined management and intelligent operation and maintenance of distribution networks, power companies widely use single-line diagrams in the SVG (Scalable Vector Graphics) format to visualize grid structure, equipment status, and operating information. SVG single-line diagrams are scalable, editable, and interactive, accurately representing power equipment (such as transformers, circuit breakers, loads, busbars, etc.) and their electrical connections. They are widely used in distribution network dispatching, operation and maintenance, and other business scenarios.

[0003] In practical applications, the distribution network SVG single-line diagram not only contains a large number of graphic symbols, topological structures and equipment status information, but also integrates hierarchical relationships, operating logic and dynamic change characteristics, forming a highly complex multimodal data representation method. However, the current diagnosis of distribution network defects and hidden dangers mainly relies on manual inspections, rule matching or simple image processing methods, which have many limitations. For example, traditional methods often rely solely on shape features to identify devices, but are unable to understand the complex connection relationships and electrical topology information between devices, resulting in low fault diagnosis accuracy. In addition, because the operating status of electrical equipment has dynamic change characteristics (such as the opening and closing status of switches, the direction of current flow, etc.), existing technologies have difficulty in effectively modeling the timing characteristics of equipment, resulting in lags in hidden danger identification and difficulty in achieving accurate prediction and early warning.

[0004] Current image recognition technologies primarily rely on computer vision and deep learning, but most remain at the stage of simple shape recognition and target detection, making it difficult to deeply understand the semantic information and logical associations of graphic symbols in SVG single-line diagrams. For example, in complex power grid scenarios, accurately identifying the electrical connection relationships between different devices, automatically inferring the distribution network topology, and correctly judging the operating status of equipment (such as the open and closed status of switches) under complex working conditions remain key technical challenges that need to be addressed. Therefore, there is an urgent need to develop more advanced image semantic understanding algorithms, combine deep learning with knowledge graph technology, and construct an intelligent parsing method to achieve accurate parsing of SVG single-line diagrams, diagnose hidden dangers, and provide intelligent early warnings, thereby improving the safety and intelligence of the distribution network. Summary of the Invention

[0005] In order to solve the above problems, the present invention provides a method for diagnosing defect and hidden danger data based on the distribution network SVG single-line diagram.

[0006] To achieve the above object, the technical solutions adopted by the present invention are as follows:

[0007] The defect and hidden danger data diagnosis method based on the distribution network SVG single-line diagram includes:

[0008] Step 1: Perform basic parsing on the input SVG single-line diagram, extract basic geometric elements and normalize them, pre-process the topological connection relationship, and build a multi-level index structure;

[0009] Step 2: Build a dynamic topology path reasoning mechanism to achieve reasoning and adaptive optimization of the topological relationship of electrical equipment through path consistency calculation, topology entropy minimization and dynamic topology update strategy;

[0010] Step 3: Build an adaptive implicit relationship knowledge graph to establish semantic associations between electrical devices through explicit topological relationship fusion, implicit functional dependency reasoning, and dynamic semantic embedding optimization;

[0011] Step 4: Construct a multimodal time series feature vector, use the spatiotemporal defect probability estimation model to calculate the risk of equipment hidden dangers, and issue an alarm through a dynamic risk warning strategy to achieve prediction and early warning of electrical equipment defect hidden dangers.

[0012] Further: Step 1 includes:

[0013] A hierarchical XML parsing method is used to parse SVG files, extracting a set of basic graphic elements, and a depth-first parsing algorithm is used to ensure complete extraction.

[0014] Normalize different types of geometric elements to form a standardized set of geometric elements;

[0015] Preprocess the topological connection relationship and construct a preliminary topological adjacency matrix by determining the common intersection of the geometric coordinate point set, the intersection relationship between the endpoint and the boundary, and the minimum Euclidean distance principle;

[0016] Construct a multi-level index structure, perform hierarchical storage according to device category, geometric shape and topological association, and form a data index.

[0017] Further: Step 2 includes:

[0018] Define a path consistency score to quantify the topological connection stability between devices by calculating the number of common neighbors between devices, the shortest path length, and the connection confidence based on physical electrical characteristics, and construct a PCS matrix;

[0019] The topological entropy is calculated using the connection probability between devices and used as the optimization objective function. Low-confidence connections are removed through the gradient descent method to optimize the topological structure.

[0020] Calculate the time-sensitive topology change rate and adopt different adaptive topology update strategies based on the size of the change rate.

[0021] Further: Step 3 includes:

[0022] Extract the topological connections and attribute information between electrical devices and convert them into a standardized electrical equipment triple representation. Use the topological adjacency matrix to clarify the explicit topological relationship, state reasoning relationship, and implicit functional dependency relationship between devices to form a preliminary electrical equipment knowledge graph.

[0023] Calculate the relationship confidence of triples in the knowledge graph, comprehensively consider the path consistency score, relationship semantic confidence, and device function similarity score, adopt a confidence-weighted knowledge completion strategy, generate or delete missing relationships through Bayesian reasoning, and optimize the accuracy and completeness of the knowledge graph;

[0024] A dynamic semantic embedding method is introduced to define the high-dimensional embedding vector of the equipment, and the adaptive attention mechanism is used to calculate the equipment relationship weight. The equipment representation is iteratively updated based on the graph neural network to ensure that the equipment semantics and topological structure are highly consistent. An adaptive electrical knowledge graph that integrates topological information and equipment functional semantics is generated to provide knowledge support for defect and hidden danger identification.

[0025] Further: Step 4 includes:

[0026] Integrate electrical topology features, equipment operating status, historical fault records, and environmental factors to construct a complete multi-modal time series feature vector;

[0027] A time-series Bayesian model is used to estimate the probability of device defects, and Markov chain Monte Carlo sampling is used to calculate the probability of future defects. Furthermore, a graph neural network is used to extract topological correlation risks, and a Gaussian kernel is used to calculate the topological impact factor. This considers the topological transmission effect between devices, updates the device's defect risk vector, and calculates the defect probability.

[0028] Define a risk level index, divide the risk level according to the equipment's influencing factors and defect probability, trigger different levels of early warning, use the hidden Markov model to predict future failure trends, and dynamically adjust the alarm level.

[0029] Compared with the prior art, the present invention has the following technical advances:

[0030] This paper addresses the intelligent parsing and defect diagnosis of distribution network SVG single-line diagrams. It innovatively proposes a series of new algorithms, including multi-scale graph parsing based on dynamic hierarchical decomposition, self-evolutionary reasoning of electrical topology states, adaptive implicit relationship knowledge graph construction, and a multimodal time-sensitive defect prediction network. These algorithms overcome the limitations of existing image recognition technology and enable deep semantic understanding, topological reasoning, and intelligent early warning of distribution network SVG single-line diagrams. Compared to traditional methods, this method offers the following advantages:

[0031] 1. Take into account both morphological information and topological semantics to achieve accurate device identification and electrical connection analysis

[0032] This invention utilizes a multi-scale graph parsing algorithm based on dynamic hierarchical decomposition, innovatively introducing an adaptive hierarchical decomposition strategy and multi-scale feature aggregation, effectively addressing the difficulty of parsing complex device structures and the fuzzy hierarchical relationships found in SVG single-line diagrams. Compared to traditional shape recognition methods, this invention not only accurately identifies the morphological features of electrical devices but also achieves high-precision recovery of topological relationships based on topological constraints and graph neural network inference of electrical connections between devices.

[0033] 2. Build a self-evolving electrical topology inference model to accurately analyze the power supply path and status of the equipment

[0034] In complex distribution network scenarios, traditional methods struggle to effectively model the dynamic connections and state changes between devices. This paper proposes a self-evolving inference model for electrical topology states. This model utilizes dynamic graph modeling, state propagation reasoning, and Bayesian inference to enable inference of electrical paths for key devices such as circuit breakers, busbars, and loads. Furthermore, combined with hidden Markov models, this paper can accurately determine the open and closed states of switchgear, thereby enhancing the intelligence of topology analysis.

[0035] 3. Combine knowledge graphs to build implicit device relationships and improve semantic understanding of complex scenarios

[0036] This invention innovatively introduces a method for constructing an adaptive implicit relational knowledge graph, overcoming the limitations of existing computer vision technologies that rely solely on morphological features and shallow topological information. By constructing a knowledge graph based on a multi-level attention mechanism, the present invention can automatically learn the implicit semantic relationships between devices and, combined with graph embedding and relational reasoning techniques, improve the ability to understand complex topological scenarios. For example, even in certain scenarios where some device information is missing, the present invention can still complete the missing relationships through knowledge graph reasoning, ensuring the robustness and generalization of the recognition results.

[0037] 4. Integrate multimodal time series information to achieve accurate defect and hidden danger prediction and intelligent early warning

[0038] Traditional methods for identifying defects and potential hazards rely primarily on static rule matching or single-point anomaly detection, failing to effectively capture long-term trends and equipment degradation patterns. This paper proposes a multimodal, time-sensitive defect prediction network that integrates topological evolution characteristics, electrical state time series data, historical fault records, and environmental information to construct a multimodal time series feature vector. This approach utilizes a graph neural network, time series Bayesian modeling, and deep LSTM to estimate potential defects. This method can detect potential defects in equipment early and predict future degradation trends, ensuring the foresight and operability of potential hazard management.

[0039] 5. Adopt adaptive dynamic early warning mechanism to improve the accuracy and practicality of defect warning

[0040] This invention utilizes a dynamic risk warning strategy, combined with the equipment's risk level index, to accurately warn of potential defects. Compared to traditional static warning methods based on fixed thresholds, this invention adaptively adjusts warning strategies based on historical trends, topological influences, and fault propagation effects, avoiding warning failures caused by false positives or missed alerts. Furthermore, by integrating the HMM hidden Markov model, this invention can predict future risk evolution trends, enabling intelligent decision support.

[0041] In summary, this invention, through the innovative integration of deep learning, graph neural networks, knowledge graphs, and time series modeling technologies, effectively addresses the challenges of semantic understanding, topological reasoning, state inference, and defect prediction for SVG single-line diagrams. This significantly improves the accuracy, real-time performance, and intelligence level of distribution network defect and hidden danger diagnosis. Ultimately, this method can be applied to multiple power industry scenarios, including intelligent operation and maintenance, hidden danger management, and distribution network scheduling optimization, providing a scientific basis and technical support for the efficient management and safe operation of distribution networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0043] In the attached figure:

[0044] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0045] The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. The embodiments of the present invention will be described below with reference to the accompanying drawings.

[0046] like Figure 1 As shown, the present invention discloses a method for diagnosing defect and hidden danger data based on a distribution network SVG single-line diagram, comprising:

[0047] Step 1: Perform basic parsing on the input SVG single-line diagram, extract basic geometric elements and normalize them, pre-process the topological connection relationship, and build a multi-level index structure;

[0048] Step 2: Build a dynamic topology path reasoning mechanism to achieve reasoning and adaptive optimization of the topological relationship of electrical equipment through path consistency calculation, topology entropy minimization and dynamic topology update strategy;

[0049] Step 3: Build an adaptive implicit relationship knowledge graph to establish semantic associations between electrical devices through explicit topological relationship fusion, implicit functional dependency reasoning, and dynamic semantic embedding optimization;

[0050] Step 4: Construct a multimodal time series feature vector, use the spatiotemporal defect probability estimation model to calculate the risk of equipment hidden dangers, and issue an alarm through a dynamic risk warning strategy to achieve prediction and early warning of electrical equipment defect hidden dangers.

[0051] Specifically, step 1 includes:

[0052] In the first step of the multi-scale graph parsing algorithm based on dynamic hierarchical decomposition, it is necessary to perform basic parsing on the input SVG single-line diagram and extract the basic geometric elements to provide basic data support for subsequent hierarchical decomposition and topological structure identification. To this end, this step adopts SVG structure parsing rules, geometric feature extraction methods, and topological relationship preprocessing strategies to achieve effective parsing of SVG files. The specific implementation process is as follows:

[0053] 1.1 Parse the SVG file structure and extract the original geometric element set

[0054] First, the input SVG file S is parsed using a hierarchical XML parsing method to extract the basic graphic element set G. This set contains all basic geometric elements (such as polylines, rectangles, circles, paths, etc.) and is mathematically represented using the following mapping function:

[0055]

[0056] in:

[0057] g i Represents a single graphic element;

[0058] T i The graphic type (such as polyline, rectangle, circle, path, etc.);

[0059] P i The set of geometric coordinate points representing the element;

[0060] A i A collection of attributes for the element (such as color, line width, fill style, etc.).

[0061] The depth-first parsing (DFP) algorithm is used to traverse all elements layer by layer according to the DOM structure of SVG to ensure the complete extraction of various basic geometric elements.

[0062] 1.2 Analyze the geometric features of the primitives and establish a normalized representation

[0063] To ensure the universality of graphic parsing, it is necessary to normalize different types of graphic elements so that the representation of all graphics can be uniformly operated in subsequent calculations. First, the parsing strategy for different geometric elements is as follows:

[0064] For a polyline, extract its vertex coordinates:

[0065] P i ={(x1,y1),(x2,y2),...,(x m ,y m )}

[0066] And calculate its length L:

[0067]

[0068] For a rectangle (rect), extract its upper left corner coordinates (x, y) and width and height (w, h):

[0069] P i ={(x,y),(x+w,y),(x+w,y+h),(x,y+h)}

[0070] For a circle, extract its center coordinates (x c ,y c ) and radius r, calculate its normalized representation:

[0071] P i ={(x c ,y c ,r)}

[0072] For the path, the Bezier curve decomposition method is used to perform multi-point fitting on the complex path and convert it into an equivalent broken line form for subsequent processing.

[0073] After all geometric elements are normalized, they are converted into a standardized data structure to form a normalized geometric element set G':

[0074] G′={g′i |g′ i =(T i ,P′ i ,A′ i ),i=1,2,...,n}

[0075] Among them, P' i Represents the normalized coordinate point set, A' i Represents the normalized attribute set.

[0076] 1.3 Preprocessing topological connectivity and establishing a preliminary adjacency matrix

[0077] After extracting and normalizing the geometric elements, in order to perform the subsequent topological analysis, it is necessary to preliminarily build the topological connection relationship between devices. For any two geometric elements g' i and g' j , define their connection relationship matrix M as:

[0078]

[0079] The following rules are used to determine the topological relationship:

[0080] 1. If there is a common intersection in the geometric coordinate point sets of two devices, they are considered to be directly connected;

[0081] 2. If the endpoint of the polyline intersects the boundary of another device, it is considered connected to the device;

[0082] 3. Use the minimum Euclidean distance principle to calculate the shortest distance between devices that are not directly connected:

[0083]

[0084] If d ij If the value is less than the set threshold τ, the two are considered to have a potential connection relationship.

[0085] Finally, a preliminary topological adjacency matrix M is formed for subsequent topological hierarchical decomposition and optimization.

[0086] 1.4 Generate preliminary analysis results and build a multi-level index

[0087] After completing SVG structure parsing, geometric feature normalization, and topological connection relationship preprocessing, it is necessary to build an efficient data index structure to facilitate subsequent hierarchical decomposition operations. This step uses a multi-level index tree (MLIT) to perform hierarchical storage based on device category, geometric form, and topological association, forming the following index structure:

[0088] MLIT={L k|L k =(C k ,G k ,M k ), k=1,2,...,K}

[0089] in:

[0090] L k represents the kth level of the index;

[0091] C k Represents the equipment category set of this layer (such as transformers, circuit breakers, load equipment, etc.);

[0092] G k A collection of device geometry elements for this layer;

[0093] M k is the topological relationship submatrix of this layer.

[0094] Index rules:

[0095] 1. Perform first-level indexing based on equipment categories (such as switches, loads, transformers, etc.);

[0096] 2. Perform second-level index optimization based on geometric features (such as rectangles, polylines, and curves);

[0097] 3. Based on the topological adjacency relationship, devices with high topological correlation are grouped into the third-layer index to form a topological subnet index.

[0098] Finally, the geometric information, topological relationships and hierarchical indexes of all devices have been established, laying the data foundation for the next step of spatial distribution density aggregation and dynamic hierarchical decomposition.

[0099] Specifically, step 2 includes:

[0100] In this step of the electrical topology state self-evolutionary inference model, a dynamic topology path inference mechanism is further constructed based on the preliminary topology adjacency matrix M and the multi-level index structure MLIT established in step 1. This allows for accurate inference of the topological relationships between different electrical devices and ensures the stability and adaptability of the topology in dynamic environments. This step proposes a self-evolutionary topology path construction method that ensures the accuracy and evolvability of the topology structure through path consistency calculation, dynamic topology optimization, and topology entropy constraint strategies.

[0101] 2.1 Calculating the path consistency score between devices

[0102] Before building an electrical topology path, it is necessary to determine whether the topological connection relationship between devices is consistent to avoid incorrect topological reasoning due to data noise or misconnection. To this end, a path consistency score (PCS) is defined to measure the consistency between any two devices g' i and g' j Topological stability between:

[0103] PCS(i,j)=α·CN(i,j)+β·SP(i,j)+γ·CC(i,j)

[0104] in:

[0105] CN(i,j) (Common Neighbors): computing device g' i and g' j The number of common neighbors between the two devices, that is, the number of other devices connected to both devices.

[0106] SP(i,j) (Shortest Path): Calculate g' i to g' j The shortest path length between them. If the path is too long, it may mean that the two devices are not directly related.

[0107] CC(i,j) (ConnectivityConfidence): uses the device connection confidence calculation method based on physical electrical characteristics to determine whether there is a strong physical connection between the two.

[0108] The constant parameters α, β, and γ can be optimized through reinforcement learning to adapt to different electrical network structures.

[0109] Finally, the path consistency score matrix PCS is constructed:

[0110]

[0111] This matrix is ​​used to guide the topology optimization and path construction process.

[0112] 2.2 Optimizing topological connections using the topological entropy minimization criterion

[0113] Since the topological structure in the SVG single-line diagram may contain redundant or incorrect connections, topological optimization is required before path construction. To this end, a topological entropy minimization criterion is proposed to ensure the orderliness and stability of the topological structure.

[0114] Define a device g' i The topological entropy of is:

[0115]

[0116] Among them, p ij Represents device g' i With equipment g' j The connection probability between them can be obtained by normalizing the PCS matrix:

[0117]

[0118] Finally, the objective function of topology optimization is:

[0119]

[0120] The topological entropy is optimized by the gradient descent method, and low-confidence connections that contribute significantly to the topological entropy are removed, making the topological structure clearer and more interpretable.

[0121] 2.3 Adopt dynamic topology update mechanism to ensure the self-evolution of topology

[0122] The topology of the distribution network may be continuously adjusted as the operating status changes. Therefore, this step adopts a dynamic topology update mechanism to achieve self-evolving topology construction through a time-sensitive topology correction strategy.

[0123] 1. Calculate the time-sensitive topology change rate (TCR)

[0124]

[0125] Among them, M(t) is the topological adjacency matrix at the current moment, and M(t-1) is the topological adjacency matrix at the previous moment.

[0126] 2. Set the topology update threshold θ and perform adaptive topology update strategies based on different change rates:

[0127] If TCR(t)<θ1, the current topology remains unchanged;

[0128] If θ1≤TCR(t)<θ2, incremental topology optimization is used to update only the local topology with large changes;

[0129] If TCR(t)≥θ2, global topology reconstruction is performed, the PCS matrix is ​​recalculated, and topological entropy is minimized.

[0130] 3. Update the topological adjacency matrix M(t) to form the evolved topological network structure.

[0131] Ultimately, the topological relationships between all devices have been optimized, forming an adaptively evolving topological network, laying the foundation for subsequent device status reasoning.

[0132] Specifically, step 3 includes: integrating the semantic understanding of electrical equipment with knowledge graphs

[0133] In this step, based on the SVG primitive parsing and hierarchical topology construction results from step 1, combined with the electrical topology reasoning and state evolution model from step 2, an adaptive implicit relationship knowledge graph construction method is proposed. This method establishes deep semantic relationships between electrical devices, thereby enabling intelligent semantic parsing of distribution network topology and defect and hidden danger identification. This method ensures the accuracy, robustness, and adaptability of the knowledge graph through the fusion of explicit topological relationships, implicit functional dependency reasoning, and dynamic semantic embedding optimization.

[0134] 3.1 Forming electrical equipment triples and building a preliminary knowledge graph

[0135] Before building the knowledge graph, it is necessary to extract the topological connections and attribute information between all electrical devices based on the dynamic topological structure generated in step 2, and convert it into a standardized Electrical Equipment Triplet (EET) representation:

[0136] T={(h,r,t)|h∈E,t∈E,r∈R}

[0137] in:

[0138] E represents a collection of electrical equipment (such as transformers, circuit breakers, busbars, etc.);

[0139] R represents a set of topological relationships (such as "electrical connection", "control relationship", "power supply path", etc.);

[0140] h is the head entity, t is the tail entity, and r is the relationship between the two.

[0141] Specifically, using the topological adjacency matrix M(t) from step 2, triples are constructed according to the following rules:

[0142] 1. Explicit topological relationships

[0143] If the device e i With e j If there is a direct connection in M(t), a triple is formed:

[0144] (e i ,“direct connection”,e j )

[0145] If there is a shortest path SP(e i ,e j ) and its length is less than the threshold d max , then a triple is formed:

[0146] (e i,“Indirect connection”,e j )

[0147] 2. State Reasoning Relationship

[0148] Based on the device switch status, infer whether the devices are in the same power supply path. If so, construct:

[0149] (e i ,“Power supply related”,e j )

[0150] If there is a control relationship between devices (such as a switch controlling a transformer), then:

[0151] (e i ,“control”,e j )

[0152] 3. Implicit functional dependencies

[0153] Establish functional constraints for devices based on their functional categories, such as:

[0154] (Transformer, "Control", Busbar)

[0155] (Circuit Breaker, "Control", Load Equipment)

[0156] Finally, a preliminary electrical equipment knowledge graph G = (E, R, T) was constructed to provide data support for subsequent semantic optimization.

[0157] 3.2 Optimizing the Knowledge Graph Using an Adaptive Implicit Relationship Reasoning Model

[0158] Since the topological relationship between electrical equipment has strong implicit association characteristics, it is necessary to perform deep reasoning through an adaptive implicit relationship reasoning model to explore the potential semantic associations between devices.

[0159] 3.2.1 Calculating Relationship Confidence in Knowledge Graphs

[0160] The Relation Confidence Score (RCS) of any triple (h, r, t) is defined as follows:

[0161] RCS(h,r,t)=λ1·PCS(h,t)+λ2·SC(r)+λ3·FC(h,t)

[0162] in:

[0163] PCS(h,t): the path consistency score obtained from the topology reasoning model in step 2;

[0164] SC(r): The semantic confidence of relation r, calculated by statistically analyzing the distribution of device relations under different topologies:

[0165]

[0166] FC(h,t): device functional similarity score, which is calculated by the cosine similarity of the device's attribute vector:

[0167]

[0168] 3.2.2 Using confidence-weighted knowledge completion strategy

[0169] For missing triples with high confidence, a weighted knowledge completion strategy is adopted to generate missing relations using Bayesian reasoning:

[0170]

[0171] If a potential relationship (h, r', t) satisfies:

[0172] RCS(h,r',t)>τ add

[0173] Then add it to the knowledge graph; if the confidence of a relationship is lower than the deletion threshold τ remove , then delete it.

[0174] 3.3 Optimizing device semantic representation using dynamic semantic embedding

[0175] Since the semantic information of electrical equipment needs to be adjusted as the topological state changes, this step introduces a dynamic semantic embedding method to optimize the semantic representation of the equipment.

[0176] 1. Define the high-dimensional embedding vector of the device:

[0177] e i =f(T,M,A)

[0178] Among them, T is the knowledge graph, M is the topology matrix, and A is the device attribute set.

[0179] 2. Use the adaptive attention mechanism to calculate device relationship weights:

[0180]

[0181] 3. Optimize the device semantic embedding matrix and iteratively update the device representation based on the Graph Neural Network (GNN) to ensure that the device semantics are highly consistent with the topological structure.

[0182] This step ultimately generates an adaptive electrical knowledge graph G = (E, R, T) that integrates topology information and equipment functional semantics, providing high-quality knowledge support for subsequent intelligent defect diagnosis and hidden danger analysis.

[0183] Specifically, step 4 includes:

[0184] This step builds on the SVG primitive parsing and topology structure extraction results from step 1, combined with the electrical topology state inference model from step 2 and the adaptive implicit relationship knowledge graph from step 3. A multimodal time-sensitive defect prediction network is proposed to achieve high-precision prediction and early warning of potential defects in electrical equipment. This method ensures the accuracy, real-time nature, and interpretability of prediction results through time series data fusion modeling, dynamic defect probability estimation, and adaptive warning strategy optimization.

[0185] 4.1 Constructing multimodal time series feature vectors to form defect prediction input data sets

[0186] Since electrical equipment defect hazards are affected by multiple factors such as topology, operating status, historical faults and environmental factors, it is necessary to fuse data from multiple modalities to construct a complete defect prediction input dataset.

[0187] 4.1.1 Defining Multimodal Feature Vectors

[0188] Assume that an electrical device e i The eigenvector at time t is It is composed of the following sub-feature vectors:

[0189]

[0190] in:

[0191] Based on the electrical topology reasoning in step 2, the topological characteristics of the device are extracted, including device degree, topological centrality, power supply path length, etc.

[0192] Based on the knowledge graph in step 3, extract the current operating status of the device, including switch status, current, voltage, temperature and other physical parameters;

[0193] Based on historical fault records, we construct time series defect features and define the time window w to calculate defect trends:

[0194]

[0195] in, for e i A binary variable indicating whether a fault occurs at time k;

[0196] Taking environmental factors (temperature, humidity, load fluctuation, etc.) into consideration, LSTM is used to model their temporal evolution characteristics.

[0197] Finally, construct the feature matrix for all devices:

[0198]

[0199] 4.2 Using spatiotemporal defect probability to estimate hidden danger risks of computing equipment

[0200] In order to accurately predict the defect risk of equipment, a spatiotemporal defect probability estimation model is proposed. The defect probability of each device is calculated through time series modeling, spatial correlation reasoning and adaptive dynamic updating.

[0201] 4.2.1 Using Time Series Bayesian Modeling to Model Defect Probability

[0202] For a device i , assuming that the probability of defect occurrence obeys the time-varying Bayesian model:

[0203]

[0204] in, represents the event that a defect occurs in the equipment at time t. Markov Chain Monte Carlo sampling is used to estimate the conditional probability and calculate the probability of defect occurrence at future times.

[0205] 4.2.2 Using Graph Neural Network (GNN) to Extract Topological Correlation Risk

[0206] Since defects and hidden dangers in electrical equipment often have a topological transmission effect, that is, the failure of a certain device may affect its adjacent devices, graph neural networks are used to calculate topological risk propagation:

[0207]

[0208] For device e i Feature representation at time t;

[0209] For device e i The set of adjacent devices;

[0210] α ij is the topological impact factor based on knowledge graph reasoning, calculated using the Gaussian kernel:

[0211]

[0212] where d ij For device e i and e j The shortest path length in the topology graph.

[0213] Finally, GNN is used to update the defect risk vector H of all devices (t) , and calculate the final defect probability:

[0214]

[0215] 4.3 Adopting dynamic risk warning strategies for intelligent alerting

[0216] After calculating the probability of equipment defects and hidden dangers, an adaptive early warning strategy needs to be implemented based on the risk level. This step proposes a dynamic risk early warning strategy that ensures the accuracy of the early warning system through adaptive adjustment of risk levels, fault propagation prediction, and alarm optimization.

[0217] 4.3.1 Calculation of risk level index

[0218] Define the Risk Evaluation Index (REI) to measure the risk level of the equipment:

[0219]

[0220] Among them, Impact i The impact factor of the device can be calculated by topological centrality and device function weight to set the risk threshold:

[0221] like Then trigger the advanced warning;

[0222] like A medium-level warning is triggered;

[0223] like Keep monitoring at a low level.

[0224] 4.3.2 Using Hidden Markov Model to Predict Future Failure Trends

[0225] To further optimize early warning effectiveness, a hidden Markov model (HMM) is used to predict future fault trends and dynamically adjust the alert level based on the prediction results. Ultimately, this forms an intelligent alert mechanism to ensure that potential defects are detected and measures can be taken in advance.

[0226] This step ultimately generates equipment defect hidden danger prediction results and intelligent early warning strategies Provide a scientific basis for fault prediction, hidden danger control and intelligent operation and maintenance of the power distribution system.

[0227] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A method for diagnosing defect and hidden danger data based on a distribution network SVG single-line diagram, characterized in that: include: Step 1: Perform basic parsing on the input SVG single-line diagram, extract basic geometric elements and normalize them, pre-process the topological connection relationship, and build a multi-level index structure; Step 2: Build a dynamic topology path reasoning mechanism to achieve reasoning and adaptive optimization of the topological relationship of electrical equipment through path consistency calculation, topology entropy minimization and dynamic topology update strategy; Step 3: Build an adaptive implicit relationship knowledge graph to establish semantic associations between electrical devices through explicit topological relationship fusion, implicit functional dependency reasoning, and dynamic semantic embedding optimization; Step 4: Construct a multimodal time series feature vector, use the spatiotemporal defect probability estimation model to calculate the risk of equipment hidden dangers, and issue an alarm through a dynamic risk warning strategy to achieve prediction and early warning of electrical equipment defect hidden dangers.

2. The method for diagnosing defect and hidden danger data based on the distribution network SVG single-line diagram according to claim 1 is characterized in that: The step 1 comprises: A hierarchical XML parsing method is used to parse SVG files, extracting a set of basic graphic elements, and a depth-first parsing algorithm is used to ensure complete extraction. Normalize different types of geometric elements to form a standardized set of geometric elements; Preprocess the topological connection relationship and construct a preliminary topological adjacency matrix by determining the common intersection of the geometric coordinate point set, the intersection relationship between the endpoint and the boundary, and the minimum Euclidean distance principle; Construct a multi-level index structure, perform hierarchical storage according to device category, geometric shape and topological association, and form a data index.

3. The defect and hidden danger data diagnosis method based on the distribution network SVG single line diagram according to claim 2 is characterized in that: The step 2 includes: Define a path consistency score to quantify the topological connection stability between devices by calculating the number of common neighbors between devices, the shortest path length, and the connection confidence based on physical electrical characteristics, and construct a PCS matrix; The topological entropy is calculated using the connection probability between devices and used as the optimization objective function. Low-confidence connections are removed through the gradient descent method to optimize the topological structure. Calculate the time-sensitive topology change rate and adopt different adaptive topology update strategies based on the size of the change rate.

4. The method for diagnosing defect and hidden danger data based on the distribution network SVG single-line diagram according to claim 3 is characterized in that: The step 3 includes: Extract the topological connections and attribute information between electrical devices and convert them into a standardized electrical equipment triple representation. Use the topological adjacency matrix to clarify the explicit topological relationship, state reasoning relationship, and implicit functional dependency relationship between devices to form a preliminary electrical equipment knowledge graph. Calculate the relationship confidence of triples in the knowledge graph, comprehensively consider the path consistency score, relationship semantic confidence, and device function similarity score, adopt a confidence-weighted knowledge completion strategy, generate or delete missing relationships through Bayesian reasoning, and optimize the accuracy and completeness of the knowledge graph; A dynamic semantic embedding method is introduced to define the high-dimensional embedding vector of the equipment, and the adaptive attention mechanism is used to calculate the equipment relationship weight. The equipment representation is iteratively updated based on the graph neural network to ensure that the equipment semantics and topological structure are highly consistent. An adaptive electrical knowledge graph that integrates topological information and equipment functional semantics is generated to provide knowledge support for defect and hidden danger identification.

5. The method for diagnosing defect and hidden danger data based on the distribution network SVG single-line diagram according to claim 4 is characterized in that: The step 4 comprises: Integrate electrical topology features, equipment operating status, historical fault records, and environmental factors to construct a complete multi-modal time series feature vector; A time-series Bayesian model is used to estimate the probability of device defects, and Markov chain Monte Carlo sampling is used to calculate the probability of future defects. Furthermore, a graph neural network is used to extract topological correlation risks, and a Gaussian kernel is used to calculate the topological impact factor. This considers the topological transmission effect between devices, updates the device's defect risk vector, and calculates the defect probability. Define a risk level index, divide the risk level according to the equipment's influencing factors and defect probability, trigger different levels of early warning, use the hidden Markov model to predict future failure trends, and dynamically adjust the alarm level.

Images