FA range intelligent analysis and labeling method and system based on mixed feature fusion and semantic check

Abstract

The invention discloses an intelligent FA range analyzing and labeling method and system based on mixed feature fusion and semantic checking, and belongs to the technical field of power distribution network automation and artificial intelligence. The method comprises the following steps: constructing a distribution network fault knowledge large model collected based on a multi-source information system; performing multi-dimensional fault probability calculation on FA related equipment by using a mixed feature fusion algorithm; an FA range accuracy checking method based on text vectorization and semantic similarity calculation is constructed, and accurate range correction is achieved; and generating FA range visual intelligent annotation and research and judgment interval recommendation. According to the method, electrical feature analysis, topological correlation calculation and text semantic understanding are fused, intelligent analysis, automatic checking and accurate labeling of the FA range are achieved, and the intelligent level and decision-making efficiency of fault handling are improved.

Description

Technical Field

[0001] This invention relates to the fields of power distribution network automation and artificial intelligence technology, specifically to a method and system for intelligent analysis and annotation of FA range based on hybrid feature fusion and semantic verification. Background Technology

[0002] Currently, the accuracy of the fault range (FA range) automatically determined by the distribution network feeder automation (FA) system after a fault occurs directly affects the efficiency of fault isolation and power restoration. Current FA range determination mainly relies on electrical quantity information, which is susceptible to interference and insufficient for identifying complex faults. Furthermore, there is a semantic gap between the textual descriptions provided by maintenance personnel on-site and the system's determination results. Traditional keyword matching verification methods have low accuracy, resulting in significant manual intervention still required to generate the FA range marked on the single-line diagram and the manually assessed interval, thus impacting the processing speed. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes an intelligent analysis and annotation method and system for FA range based on hybrid feature fusion and semantic verification.

[0004] The technical solution of this invention is: an intelligent analysis and annotation method for FA range based on hybrid feature fusion and semantic verification, comprising the following steps: Step 1): Construct a large-scale model of distribution network fault knowledge based on multi-source information system collection; Step 2): Based on the large knowledge model of distribution network faults, a hybrid feature fusion algorithm is used to calculate the multi-dimensional fault probability of FA-related equipment and generate a list of equipment probabilities. Step 3): Based on Step 2), the FA range is checked and corrected through text vectorization and semantic similarity calculation, and the final confirmed list of FA range devices is output. Step 4): Based on Step 3), generate FA range visualization intelligent annotation and judgment interval recommendation.

[0005] Step 1) includes: Multi-source heterogeneous data are extracted in real time or synchronized periodically from distribution network fault handling monitoring applications, power grid dispatching systems, production management systems, and mobile operation platforms. ETL tools are used to clean, align and integrate multi-source heterogeneous data to build a distribution network fault analysis knowledge base with "fault event ID" as the primary key. Subsequently, the Sentence-BERT model was used to convert the descriptions of typical fault cases, standard handling procedures, and on-site feedback texts in the knowledge base into semantic vectors and store them in a dedicated vector database. Finally, through efficient parameter fine-tuning technology, vectorized domain knowledge is injected into a general large language model, completing the construction of a large knowledge model for distribution network faults.

[0006] The multi-source heterogeneous data includes FA action signals, protection alarm information, power grid topology graphic files, and on-site maintenance work order texts.

[0007] Step 2) includes: Data standardization processing is performed on the multi-dimensional feature dataset of fault events output by the large-scale knowledge model of distribution network faults; Based on the processed data, the electrical fault probability and topological correlation degree are calculated respectively; Calculate the comprehensive failure probability based on electrical failure probability and topological correlation. Calculate the FA-related equipment and generate a list of equipment probabilities sorted in descending order of comprehensive failure probability.

[0008] The multi-dimensional feature dataset of the fault events includes electrical feature vectors and equipment topology adjacency matrices. The electrical characteristic vector includes impedance modulus and phase angle.

[0009] The electrical fault probability is: ; In the formula, S elec For electrical fault probability scores, To create an ensemble model of multi-classifiers that integrates decision trees, support vector machines, and neural networks, For electrical feature vectors; The topological correlation degree is: ; In the formula, S topo The topological correlation score. d i For equipment i The shortest path hop count to the fault core; The overall failure probability is: ; In the formula, S hybrid The overall failure probability score, and These are the weighting coefficients.

[0010] Step 3) includes: (1) Using the Sentence-BERT model, the "FA range description text" generated by the system is... T systemThe "fault description text" submitted by the on-site maintenance personnel. T field Convert them into range description vectors respectively and field fault description vector ; (2) Calculate cosine similarity Sim sem : ; Set dynamic similarity threshold ,like Sim sem ≥ The initial judgment range of the system is accurate; if Sim sem < If so, then mark it as the range to be corrected; (3) To determine the range to be corrected, initiate a vector database search to find the range that matches the target range. Most similar K Vector of historical failure cases and its corresponding equipment set ; Calculate the corrected weights for the current device set. W corr : ; In the formula, For indicator functions; (4) Generate the final device attention score S final : ; In the formula, γ is the correction coefficient; according to Sort and output the final confirmed list of FA range devices.

[0011] In step 4), when the FA range list is updated, it is automatically marked on the single-line graph with a highlighted color and a flashing border; Based on the final FA range, the algorithm automatically searches upstream for the nearest connection switch and downstream for all load nodes using a breadth-first search algorithm, generating a recommended judgment interval and rendering it with a semi-transparent overlay color.

[0012] A FA range intelligent analysis and annotation system based on hybrid feature fusion and semantic verification includes: The module is used to build a large-scale model of distribution network fault knowledge collected from multi-source information systems; The hybrid module is used to perform multi-dimensional fault probability calculations on FA-related equipment based on the large model of distribution network fault knowledge and to generate a list of equipment probabilities. The output module is used to verify and correct the FA range through text vectorization and semantic similarity calculation, and output the final confirmed list of FA range devices. The annotation recommendation module is used to generate intelligent annotations for FA range visualization and to recommend intervals for analysis.

[0013] In its operation, this invention constructs a large-scale knowledge model of distribution network faults through multi-source data fusion. Combining hybrid feature fusion and semantic similarity verification, it achieves intelligent analysis and accurate labeling within the fault location range. This invention significantly improves the accuracy of fault location and its consistency with actual on-site conditions, automatically generates and visualizes judgment intervals, effectively reduces the burden of manual judgment, optimizes fault handling processes, and promotes the transformation of distribution network fault handling towards intelligence, precision, and efficiency. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0015] 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 examples 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.

[0016] like Figure 1 As shown, this invention provides an intelligent analysis and annotation method for FA range based on hybrid feature fusion and semantic verification, including the following steps: Step 1): Construct a large-scale model of distribution network fault knowledge based on multi-source information system collection; Step 2): Based on the large knowledge model of distribution network faults, a hybrid feature fusion algorithm is used to calculate the multi-dimensional fault probability of FA-related equipment and generate a list of equipment probabilities. Step 3): Based on Step 2), the FA range is checked and corrected through text vectorization and semantic similarity calculation, and the final confirmed list of FA range devices is output. Step 4): Based on Step 3), generate FA range visualization intelligent annotation and judgment interval recommendation.

[0017] This invention first constructs a large-scale model of distribution network fault knowledge based on multi-source information system collection; then, it uses a hybrid feature fusion algorithm to calculate the multi-dimensional fault probability of FA-related equipment; next, it constructs an FA range accuracy verification method based on text vectorization and semantic similarity calculation to achieve accurate range correction; finally, it generates FA range visualization intelligent annotation and judgment interval recommendation.

[0018] This invention integrates electrical feature analysis, topology correlation calculation, and text semantic understanding to achieve intelligent analysis, automated verification, and accurate labeling of the fault location (FA) range, thereby improving the intelligence level and decision-making efficiency of fault handling. This invention relies on the data and functions of existing distribution network fault handling and monitoring applications to achieve accurate, rapid, and automatic labeling and analysis support for the FA range.

[0019] The specific process of this invention is as follows: Step 1): Construct a large-scale knowledge model of distribution network faults based on multi-source information systems. By aggregating FA action information, terminal alarm data, equipment topology and status data, and on-site operation and maintenance text reports from distribution network fault handling and monitoring applications, a unified distribution network fault analysis knowledge base for maintenance and management is built. This knowledge base adopts a fault scenario-based tagging system for knowledge classification and quality management, significantly improving the retrieval and standardization of fault cases.

[0020] The fault knowledge base content, which has been organized, filtered, and tagged, as well as the accumulated text information such as typical fault descriptions and historical handling records, are vectorized to form a high-dimensional semantic vector representation, which is then stored in a dedicated vector database.

[0021] Based on this vector database, we deeply integrate and professionally empower the general pre-trained model with knowledge specific to the distribution network fault domain, injecting professional knowledge and semantic understanding capabilities in the fields of fault mechanism and handling strategy, thereby constructing a large-scale knowledge model of distribution network faults.

[0022] The specific implementation process is as follows: Through the data interface, FA action signals, protection alarm information, power grid topology graphic files and on-site operation and maintenance work order texts are extracted in real time or synchronized periodically from the distribution network fault handling monitoring application, power grid dispatching system (DMS), production management system (PMS) and mobile operation platform.

[0023] ETL tools were used to clean, align, and integrate multi-source heterogeneous data to build a distribution network fault analysis knowledge base with "fault event ID" as the primary key. This knowledge base is managed using a multi-level tagging system based on faulty equipment type, fault phenomenon, and handling stage.

[0024] Subsequently, the Sentence-BERT model was used to convert the descriptions of typical fault cases, standard handling procedures, and on-site feedback texts in the knowledge base into 768-dimensional semantic vectors, which were then stored in dedicated vector databases such as Milvus.

[0025] The optimized Sentence-BERT model can be used, with optimizations mainly focused on improving domain adaptability and semantic representation capabilities. Specifically, the original Sentence-BERT model is subjected to supervised domain-adaptive fine-tuning using professional text corpora in the field of power distribution network fault handling (including historical fault reports, operating procedures, equipment files, etc.), and a contrastive learning loss function is used to enhance the model's ability to capture key semantic features of fault scenarios.

[0026] In this way, the optimized model improves the accuracy of fault text semantic matching tasks, and can more accurately map professional expressions such as "switch tripping" and "line grounding" into vector representations with higher discriminative power, effectively supporting the accuracy of subsequent semantic similarity calculation and case retrieval.

[0027] Finally, through efficient parameter fine-tuning techniques such as LoRA, vectorized domain knowledge is injected into general-purpose large language models such as ChatGLM, completing the construction of a large-scale distribution network fault knowledge model, enabling it to possess professional fault semantic understanding and reasoning capabilities.

[0028] Step 2): The function of using a hybrid feature fusion algorithm to calculate the multi-dimensional failure probability of FA-related equipment. 21) Data Preprocessing Receive the multi-dimensional feature dataset of fault events output from the large-scale distribution network fault knowledge model constructed in step 1), including electrical feature vectors (impedance modulus, phase angle, etc.) and equipment topology adjacency matrix. Standardize the data (i.e., the multi-dimensional feature dataset of fault events): ; In the formula, X norm This represents the feature vector after standardization. These are the mean and standard deviation vectors of the feature dimensions, respectively.

[0029] The preprocessing employs the Z-score standardization method, subtracting the mean and dividing by the standard deviation for each feature dimension to eliminate the influence of dimensions, thus ensuring the comparability of different electrical parameters and topology indices. The resulting standardized feature matrix provides a unified input for subsequent calculations of electrical fault probability and topology correlation, ensuring the fairness of the fusion weights and forming the foundation of the entire hybrid feature fusion algorithm.

[0030] 22) Calculation of electrical fault probability Construct an electrical condition assessment model based on multi-classifier ensemble, and input electrical feature vectors. Output electrical fault probability score S elec : ; In the formula, This is an ensemble model for multi-classifiers that integrates decision trees, support vector machines, and neural networks.

[0031] 23) Calculation of topological correlation Construct a topology diagram with the FA initial action device as the core, for the device i Calculate its topological correlation score S topo : ; In the formula, d i For equipment i The shortest path hop count to the faulty core.

[0032] 24) Hybrid Probability Fusion Define the overall failure probability score of the equipment. S hybrid : ; In the formula, and Weighting coefficients Output the mixed failure probability vector for all devices. Up to step 3).

[0033] The specific implementation process is as follows: Upon receiving a specific fault event ID, the electrical characteristics and topology data of the relevant equipment are extracted from the large-scale knowledge model of distribution network faults. For switchgear with FA test data, the impedance modulus Z=85.1Ω and phase angle φ=-47.2° at a frequency of 1.05Hz are extracted as electrical feature vectors. The input is fed into a pre-trained multi-classifier ensemble model (e.g., an ensemble of random forest, SVM, and a three-layer BP neural network with weights of 0.3:0.3:0.4), and the output is the electrical fault probability score of the device. S elec =0.92.

[0034] Simultaneously, based on the power grid topology, the topological distance d=2 between the device and the initial operating switch of FA is calculated, thereby obtaining its topological correlation score. S topo =1 / (1+2)≈0.33.

[0035] With fusion weights set α=0.7 and β=0.3, the overall failure probability score of the equipment is then calculated. S hybrid = 0.7*0.92 +0.3*0.33 ≈ 0.743. The system performs this calculation process on all relevant devices to generate the result.S hybrid A list of device probabilities sorted in descending order.

[0036] The purpose of the device probability list is as follows: Firstly, this list is the direct output of the hybrid feature fusion algorithm in step 2), which quantifies the probability of failure of each device and provides maintenance personnel with key points for preliminary judgment. Secondly, this list serves as the input basis for the semantic verification step 3). Its original ranking and device probability values ​​are the core basis for the subsequent semantic-driven correction mechanism—the correction weight W. corr The devices in this list are then assigned and calculated to generate a device attention score S that incorporates semantic information. final The output is then reordered and visualized in step 4).

[0037] Step 3): Construct an accuracy verification method for the FA range based on text vectorization and semantic similarity calculation to achieve precise range correction. For potential discrepancies between the initial FA range assessment and the on-site text description in the system (i.e., the distribution network fault handling and monitoring application constructed in this invention, which integrates a large distribution network fault knowledge model and a hybrid feature fusion algorithm), this solution designs a semantically driven consistency verification mechanism: First, the text information is mapped to a high-dimensional vector space using a sentence embedding model (i.e., the Sentence-BERT model); when the system determines that the semantic similarity between the range description vector and the on-site report vector is lower than a dynamic threshold, a correction suggestion mechanism based on case retrieval is initiated, effectively improving the consistency between the FA range and the actual situation. Specifically, this may include the following steps: (1) Text vectorization The Sentence-BERT model is used to process the "FA range description text" generated by the system. Tsystem The "fault description text" submitted by the on-site maintenance personnel. T field They are converted into high-dimensional semantic vectors respectively, namely: range description vector and field fault description vector .

[0038] (2) Semantic similarity decision Calculate the cosine similarity between two text vectors Sim sem : ; Set dynamic similarity threshold .like Sim sem ≥ The initial judgment range of the system is accurate; if Sim sem < If it is, then mark it as the range to be corrected.

[0039] (3) Case retrieval correction engine To determine the correction range, initiate a vector database search to find and match... Most similar K Vector of historical failure cases and its corresponding equipment set .

[0040] Calculate the corrected weights for the current device set. W corr : ; In the formula, This is an indicator function.

[0041] (4) Range correction and output By combining the adjusted weights with the mixed probability from step 2), the final device attention score S is generated. final : ; according to Sort and output the final confirmed list of FA range devices (step 4).

[0042] The ranking here serves the purpose of: based on the final device attention score S final A global reordering of all relevant candidate devices is performed to determine the final FA-range device list.

[0043] Specifically, the process involves combining all the initial devices and their mixed probabilities calculated in step 2) with the corrected weights calculated for each device through semantic verification in step 3) to generate a new, unified sequence of device priorities. This semantically enhanced global ranking result, as the final output of the entire analysis process, ensures that the device list submitted to the visualization stage simultaneously incorporates electrical characteristics, topological relationships, and textual semantic information, thereby optimizing and calibrating the analysis results.

[0044] Specific implementation examples are as follows: (1) The system initially determines the range of FA and generates descriptive text. T system = "The fault involves switches No. 15 and No. 16." On-site maintenance personnel reported the text via a mobile operation app. T field = "On-site inspection revealed burn marks on switch No. 15 and switch cabinet No. 17."

[0045] (2) After vectorizing the two text segments using the SBERT model, the cosine similarity is calculated to obtain... Simsem =0.68, which is lower than the preset threshold θ=0.75, so the system judges it as "to be corrected".

[0046] (3) Subsequently, the system retrieves data from the vector database that is related to... The three historical cases with the most similar semantics have case vectors that are similar to... The similarity scores were [0.91, 0.85, 0.82], and the corresponding sets of faulty devices were [{15th switch, 17th switch}, {15th switch}, {17th switch, 18th switch}].

[0047] (4) Calculate the equipment correction weight: Switch No. 15 W corr =0.91 + 0.85 = 1.76, switch number 17 W corr =0.91 + 0.82 = 1.73. Assume switch number 15 was originally... S hybrid =0.82, original value of switch No. 17 S hybrid =0.35, take the correction factor γ=0.1, then the final score S final (15) = 0.82 + 0.1 * 1.76 = 0.996, S final (17) = 0.35 + 0.1 * 1.73 = 0.523.

[0048] (5) After correction, the ranking of switch No. 17 was significantly improved (i.e. the ranking moved significantly forward, exceeding the range included threshold set by the system or entering the top N positions of the list (the specific threshold or position can be set according to the actual operation and maintenance strategy)), and was successfully included in the final FA range, effectively correcting the initial omission of the system.

[0049] The revised final FA range device list includes switches 15, 16, and 17. Switch 16, as a device in the initial system judgment, has a higher original hybrid probability (Shybrid). After semantic correction calculation in step 3), its final attention score (S... final The value remained at a high level and was therefore retained in the final output list after reordering. The correction involved re-evaluating and reordering the entire candidate device set, adding switch number 17 to the final range, resulting in a more complete final FA range list containing three devices.

[0050] Step 3): Combining the adjusted weights with the mixed probabilities from Step 2), recalculate and generate the final device attention score for all relevant devices; sort all devices uniformly based on this score, and output the final confirmed list of devices within the FA range to Step 4).

[0051] Step 4): Based on the final FA range device list output in Step 3), the distribution network topology connection relationship is integrated and visualized using the graphics rendering engine of the distribution network fault handling monitoring application. When the FA range list is updated, it is automatically marked on the single-line diagram with a highlighted color and a flashing border. According to the final FA range, the nearest tie switch is automatically searched upstream and all load nodes are searched downstream based on the breadth-first search algorithm to generate a recommended judgment interval, which is rendered with a semi-transparent overlay color. At the same time, the FA analysis conclusion, semantic verification score, and recommended judgment interval topology are updated synchronously on the handling process interface, providing comprehensive fault handling decision support for operation and maintenance personnel.

[0052] The specific implementation is as follows: (1) The system pushes the final FA range device list (such as [switch 15, switch 16, switch 17]) confirmed in step 3) to the front-end graphical interface of the distribution network fault handling monitoring application in real time via the WebSocket protocol.

[0053] (2) After receiving the data, the front end immediately fills the corresponding device elements on the single-line diagram with a bright orange (#FFA500) according to the predefined rendering rules, and adds a red (#FF0000) flashing border to realize the intelligent labeling of the fault range.

[0054] (3) At the same time, the backend service uses graph theory algorithm to perform breadth-first search (BFS) starting from the devices in the list: search upstream to the nearest normally open tie switch S101, and traverse all connected load nodes (such as L1, L2, L3) downstream to automatically generate the judgment interval [S101, L1, L2, L3].

[0055] (4) The recommended interval is rendered on the single-line graph as a semi-transparent blue (#1E90FF, 40% transparency) covered area and is automatically added to the system's "fault judgment interval" list for maintenance personnel to adopt with one click.

[0056] A FA range intelligent analysis and annotation system based on hybrid feature fusion and semantic verification includes: The module is used to build a large-scale model of distribution network fault knowledge collected from multi-source information systems; The hybrid module is used to perform multi-dimensional fault probability calculations on FA-related equipment based on the large model of distribution network fault knowledge and to generate a list of equipment probabilities. The output module is used to verify and correct the FA range through text vectorization and semantic similarity calculation, and output the final confirmed list of FA range devices. The annotation recommendation module is used to generate intelligent annotations for FA range visualization and to recommend intervals for analysis.

[0057] This invention constructs a large-scale knowledge model of multi-source distribution network faults, combining a hybrid feature fusion algorithm with a semantic similarity verification mechanism to achieve intelligent analysis and accurate labeling of fault location ranges. This invention significantly improves fault location accuracy and range verification efficiency. Furthermore, it automatically generates visual annotations and recommended judgment intervals based on the full-process analysis results.

[0058] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A method for intelligent analysis and annotation of FA range based on hybrid feature fusion and semantic verification, characterized in that, Includes the following steps: Step 1): Construct a large-scale model of distribution network fault knowledge based on multi-source information system collection; Step 2): Based on the large knowledge model of distribution network faults, a hybrid feature fusion algorithm is used to calculate the multi-dimensional fault probability of FA-related equipment and generate a list of equipment probabilities. Step 3): Based on Step 2), the FA range is checked and corrected through text vectorization and semantic similarity calculation, and the final confirmed list of FA range devices is output. Step 4): Based on Step 3), generate FA range visualization intelligent annotation and judgment interval recommendation.

2. The intelligent analysis and annotation method for FA range based on hybrid feature fusion and semantic verification according to claim 1, characterized in that, Step 1) includes: Multi-source heterogeneous data are extracted in real time or synchronized periodically from distribution network fault handling monitoring applications, power grid dispatching systems, production management systems, and mobile operation platforms. ETL tools are used to clean, align and integrate multi-source heterogeneous data to build a distribution network fault analysis knowledge base with "fault event ID" as the primary key. Subsequently, the Sentence-BERT model was used to convert the descriptions of typical fault cases, standard handling procedures, and on-site feedback texts in the knowledge base into semantic vectors and store them in a dedicated vector database. Finally, through efficient parameter fine-tuning technology, vectorized domain knowledge is injected into a general large language model, completing the construction of a large knowledge model for distribution network faults.

3. The intelligent analysis and annotation method for FA range based on hybrid feature fusion and semantic verification according to claim 2, characterized in that, The multi-source heterogeneous data includes FA action signals, protection alarm information, power grid topology graphic files, and on-site maintenance work order texts.

4. The intelligent analysis and annotation method for FA range based on hybrid feature fusion and semantic verification according to claim 1, characterized in that, Step 2) includes: Data standardization processing is performed on the multi-dimensional feature dataset of fault events output by the large-scale knowledge model of distribution network faults; Based on the processed data, the electrical fault probability and topological correlation degree are calculated respectively; Calculate the comprehensive failure probability based on electrical failure probability and topological correlation. Calculate the FA-related equipment and generate a list of equipment probabilities sorted in descending order of comprehensive failure probability.

5. The intelligent analysis and annotation method for FA range based on hybrid feature fusion and semantic verification according to claim 4, characterized in that, The multi-dimensional feature dataset of the fault events includes electrical feature vectors and equipment topology adjacency matrices. The electrical characteristic vector includes impedance modulus and phase angle.

6. The intelligent analysis and annotation method for FA range based on hybrid feature fusion and semantic verification according to claim 4, characterized in that, The electrical fault probability is: ； In the formula, S elec For electrical fault probability scores, To create an ensemble model of multi-classifiers that integrates decision trees, support vector machines, and neural networks, For electrical feature vectors; The topological correlation degree is: ； In the formula, S topo The topological correlation score. d i For equipment i The shortest path hop count to the fault core; The overall failure probability is: ； In the formula, S hybrid The overall failure probability score, and These are the weighting coefficients.

7. The intelligent analysis and annotation method for FA range based on hybrid feature fusion and semantic verification according to claim 1, characterized in that, Step 3) includes: (1) Using the Sentence-BERT model, the FA range description text generated by the system is used. T system Fault description text submitted by on-site maintenance personnel T field Convert them into range description vectors respectively and field fault description vector ; (2) Calculate cosine similarity Sim sem : ； Set dynamic similarity threshold ,like Sim sem ≥ The initial judgment range of the system is accurate; if Sim sem < If so, then mark it as the range to be corrected; (3) For the range to be corrected, initiate a vector database search to find the most similar vector. K Vector of historical failure cases and its corresponding equipment set ; Calculate the corrected weights for the current device set. W corr : ； In the formula, For indicator functions; (4) Generate the final device attention score S final : ； In the formula, γ is the correction coefficient; according to Sort and output the final confirmed list of FA range devices.

8. The intelligent analysis and annotation method for FA range based on hybrid feature fusion and semantic verification according to claim 1, characterized in that, In step 4), when the FA range list is updated, it is automatically marked on the single-line graph with a highlighted color and a flashing border; Based on the final FA range, the algorithm automatically searches upstream for the nearest connection switch and downstream for all load nodes using a breadth-first search algorithm, generating a recommended judgment interval and rendering it with a semi-transparent overlay color.

9. A FA range intelligent analysis and annotation system based on hybrid feature fusion and semantic verification, characterized in that, Applied to the FA range intelligent analysis and annotation method based on hybrid feature fusion and semantic verification as described in any one of claims 1-8, include: The module is used to build a large-scale model of distribution network fault knowledge collected from multi-source information systems; The hybrid module is used to perform multi-dimensional fault probability calculations on FA-related equipment based on the large model of distribution network fault knowledge and to generate a list of equipment probabilities. The output module is used to verify and correct the FA range through text vectorization and semantic similarity calculation, and output the final confirmed list of FA range devices. The annotation recommendation module is used to generate intelligent annotations for FA range visualization and to recommend intervals for analysis.

Images