Power transformation loop reliability multi-source data fusion method, system and equipment based on natural language processing technology and medium

Abstract

The invention discloses a power transformation loop reliability multi-source data fusion method, system and equipment based on a natural language processing technology and a medium, and belongs to the technical field of power transformation loop reliability, and the method comprises the steps: carrying out the intelligent management of machine account data and operation data of a power transformation loop, constructing a standardized data base, and outputting a data set; based on the data set, realizing association and integration of cross-system data through multi-dimensional feature matching, and outputting a fused data set; on the basis of the fused data set, a natural language processing model is adopted to study and judge a shutdown event affiliation loop and an event type; and based on the event type research and judgment result, performing full-dimension analysis on the operation data, completing loop state classification and shutdown statistics, and generating loop reliability operation data. According to the method, by fusing natural language processing and graph neural network technologies, intelligent management and deep semantic fusion of the multi-source heterogeneous data of the power transformation loop are realized, and the problems of poor data quality, weak relevance and the like in a traditional method are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of substation circuit reliability technology, specifically to a method, system, device, and medium for multi-source data fusion of substation circuit reliability based on natural language processing technology. Background Technology

[0002] As a core component of the power system, the reliability of substation circuits directly affects the stability and security of power supply. With the advancement of information technology construction in the power industry, substation circuit-related data has experienced explosive growth. This data is distributed across multiple business systems such as the power grid management platform, the integrated power grid operation intelligent OCS system, and the OMS system, covering two core types of data: ledger data (substations, equipment, primary wiring diagrams, etc.) and operational data (production plans, dispatch records, work tickets, operation tickets, etc.).

[0003] However, existing technologies have the following prominent problems in the management and application of substation circuit reliability data: 1. Severe heterogeneity of multi-source data, forming "data silos"; 2. Outdated data governance methods, making it difficult to guarantee quality; 3. Low degree of integration of multi-source data, failing to release the associated value; 4. The judgment of outage events relies on manual labor, resulting in low accuracy and efficiency; 5. Lagging dynamic updates, failing to reflect real-time status. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention provides a method, system, device and medium for multi-source data fusion of substation circuit reliability based on natural language processing technology.

[0005] Therefore, the technical problem solved by this invention is that the current power system has a variety of heterogeneous data such as the ledgers and operation of substation circuits, which are stored in various business systems. This results in "data silos" with inconsistent data standards, poor quality, and weak correlation. Furthermore, traditional data governance methods are outdated and have a low degree of integration, which leads to the judgment of outage events relying on manual methods and the lag in dynamic updates. This makes it impossible to provide highly correlated data support for the reliability analysis of substation circuits.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a method for fusing multi-source data on substation circuit reliability based on natural language processing technology, comprising, Intelligent management is implemented for the ledger data and operational data of substation circuits to construct a standardized data foundation and output a dataset. Based on the dataset, cross-system data is correlated and integrated through multi-dimensional feature matching to output a fused dataset. Based on the fused dataset, a natural language processing model is used to determine the circuit to which the outage event belongs and the event type. Based on the event type determination results, the operational data is analyzed in all dimensions to complete the circuit status classification and outage statistics, generating circuit reliability operational data.

[0007] As a preferred embodiment of the multi-source data fusion method for substation circuit reliability based on natural language processing technology described in this invention, the step of intelligently managing the ledger data and operational data of the substation circuit and constructing a standardized data foundation includes: Regular expressions and ETL tools are used to standardize the ledger data, graph computing is used to achieve intelligent error correction of topology, and a rule engine is used to complete the ledger association verification.

[0008] Semantic analysis is performed on the task descriptions in the operational data to identify and correct errors in the data entry, and the operation time is verified through an intelligent verification algorithm.

[0009] As a preferred embodiment of the multi-source data fusion method for substation circuit reliability based on natural language processing technology described in this invention, the intelligent topology error correction includes: Identify device element information in a primary wiring diagram based on an XML algorithm.

[0010] A topology graph is generated by using BFS traversal, and the ledger data is compared with the topology graph using graph embedding technology to identify contradictions and trigger a closed-loop correction mechanism.

[0011] As a preferred embodiment of the multi-source data fusion method for substation circuit reliability based on natural language processing technology described in this invention, the semantic analysis includes: Natural language processing technology is used to segment and semantically analyze the job task descriptions in order to identify and correct errors in the data entry.

[0012] Clustering algorithms and Bayesian network regression were used to complete missing values.

[0013] As a preferred embodiment of the multi-source data fusion method for substation circuit reliability based on natural language processing technology described in this invention, wherein: the cross-system data association and integration are achieved through multi-dimensional feature matching, and the output fused dataset includes, The device text is divided into power industry terminology units using word segmentation technology. The matching degree between terminology units is calculated, with the smallest matching unit being power industry terminology.

[0014] The semantic feature vectors of the text are extracted using the ALBERT model, and the semantic similarity of the text is evaluated by calculating the cosine similarity between the vectors.

[0015] The beneficial effects of this preferred technical solution are that, through the dual verification mechanism of word segmentation matching based on power professional terms and ALBERT semantic vector similarity calculation, it not only ensures the accurate alignment of equipment text at the professional terminology level, but also achieves deep association at the contextual semantic level. This effectively solves the matching failure problem caused by non-standard naming and expression differences in traditional methods, and improves the accuracy and completeness of cross-system data fusion.

[0016] As a preferred embodiment of the multi-source data fusion method for substation circuit reliability based on natural language processing technology described in this invention, the step of using a natural language processing model to determine the circuit to which the outage event belongs and the event type includes... The ALBERT-Seq2Seq model is adopted, which encodes contextual semantics through bidirectional Transformer and extracts contextual features by combining Bi-LSTM encoder. An attention mechanism is introduced to weight and focus the equipment operation description, so as to classify multi-label loops and determine the event type of shutdown events.

[0017] The event types include scheduled shutdown, affected shutdown, planned shutdown, and forced shutdown.

[0018] The ALBERT-Seq2Seq model employs a transfer learning strategy, freezing the low-level parameters of the pre-trained model to retain general semantic understanding capabilities, and adjusting the upper-level network to adapt to entities and topological relationships in the power industry.

[0019] The beneficial effects of this preferred technical solution are that by combining the ALBERT-Seq2Seq model with the dual encoding mechanism of bidirectional Transformer and Bi-LSTM, and introducing an attention mechanism to focus on the operation description of key equipment, it accurately adapts to the characteristics of the power field while retaining the ability to understand general semantics. This achieves accurate classification of the circuit to which the outage event belongs and the event type by multiple labels, effectively solving the problem of insufficient accuracy of traditional methods in understanding complex power texts and judging multiple types of events.

[0020] As a preferred embodiment of the multi-source data fusion method for substation circuit reliability based on natural language processing technology described in this invention, the step of performing full-dimensional analysis of operational data, completing circuit status classification and outage statistics, and generating circuit reliability operational data includes: The operational data is filtered to remove data below 110kV and data related to non-substation circuits. Multiple work items are broken down by punctuation marks or serial numbers to extract key information.

[0021] The circuit status is divided into operating status and shutdown status. The shutdown status includes scheduled shutdown, affected shutdown, planned shutdown and forced shutdown. The status is determined based on whether the circuit can complete the power transmission function within a specific capacity range.

[0022] Based on the loss of circuit function, outage events are statistically analyzed. For different wiring methods of main transformer circuits, outgoing circuits, and bus circuits, corresponding conversion factors are determined according to the magnitude of the impact of outage on circuit function, generating circuit reliability operation data that includes circuit code, outage nature, conversion factor, and cause of responsibility.

[0023] The beneficial effects of this preferred technical solution are that by establishing a circuit state classification system based on functional loss and a dynamic conversion coefficient mechanism for different wiring methods, it achieves refined analysis and standardized statistics of operating data. This ensures that the circuit reliability assessment is highly consistent with the actual operating conditions of the power grid, and provides a unified and quantifiable reliability data analysis basis for substation circuits of different voltage levels and wiring types.

[0024] This invention provides a multi-source data fusion system for substation circuit reliability based on natural language processing technology.

[0025] To address the aforementioned technical problems, this invention provides the following technical solution: a multi-source data fusion system for substation circuit reliability based on natural language processing technology, comprising: a data governance and processing module, a semantic analysis and feature engineering module, an intelligent judgment and classification module, and an operational data analysis and statistics module.

[0026] The data governance and processing module intelligently manages the ledger data and operational data of the substation circuit, constructs a standardized data foundation, and outputs datasets.

[0027] The semantic analysis and feature engineering module, based on the dataset, achieves semantic association and integration of cross-system data through multi-dimensional feature similarity evaluation, and outputs a fused dataset.

[0028] The intelligent analysis and classification module, based on the fused dataset, uses a deep semantic understanding model to analyze the circuit to which the outage event belongs and intelligently analyzes the event type.

[0029] The operational data analysis and statistics module performs a full-dimensional analysis of the operational data based on the event type judgment results, completes the circuit status classification and outage statistics, and generates circuit reliability operational data.

[0030] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the multi-source data fusion method for substation circuit reliability based on natural language processing technology.

[0031] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the multi-source data fusion method for substation circuit reliability based on natural language processing technology.

[0032] The beneficial effects of this invention are as follows: By integrating natural language processing and graph neural network technology, this invention realizes intelligent governance and deep semantic fusion of multi-source heterogeneous data of substation circuits, and constructs a precise judgment system for outage events based on the ALBERT-Seq2Seq model. This effectively solves the problems of poor data quality, weak correlation and reliance on manual judgment in traditional methods, and improves data accuracy, correlation utilization and operation and maintenance analysis efficiency. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 The above is a flowchart of a multi-source data fusion method for substation circuit reliability based on natural language processing technology, provided as an embodiment of the present invention.

[0035] Figure 2 This is an overall framework diagram of a multi-source data fusion system for substation circuit reliability based on natural language processing technology, provided as an embodiment of the present invention. Detailed Implementation

[0036] To make the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0037] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a method for fusing multi-source data on the reliability of substation circuits based on natural language processing technology, including: S1. Intelligent management of the ledger data and operation data of substation circuits, construction of a standardized data foundation, and output of datasets.

[0038] S2. Based on the dataset, cross-system data is associated and integrated through multi-dimensional feature matching, and a fused dataset is output.

[0039] S3. Based on the fused dataset, a natural language processing model is used to determine the circuit to which the outage event belongs and the event type.

[0040] S4. Based on the event type analysis results, perform full-dimensional analysis of the operation data, complete the circuit status classification and outage statistics, and generate circuit reliability operation data.

[0041] By constructing a full-link technology system from intelligent data governance and semantic-level fusion to intelligent analysis, the system has achieved standardized processing of multi-source heterogeneous data from substation circuits, deep cross-system correlation, and accurate classification of outage events. This has not only effectively solved core problems such as data silos, low quality, and low efficiency of manual analysis, but also improved the accuracy, timeliness, and automation level of circuit reliability analysis.

[0042] Example 2, an embodiment of the present invention, provides a method for fusing multi-source data on substation circuit reliability based on natural language processing technology, based on the previous embodiment, including: The intelligent management of substation circuit ledger data and operational data in S1, and the construction of a standardized data foundation, includes steps A1-A2: A1. The ledger data is standardized using regular expressions and ETL tools, intelligent error correction of topology is achieved through graph computing, and the ledger association verification is completed using a rule engine.

[0043] Specifically, a regular expression library is used to format fields such as device name, date, and voltage level to unify the data representation; ETL tools are used to batch detect and complete missing key fields, and intelligent completion is achieved by referring to historical data patterns to ensure field integrity and format consistency.

[0044] A graph rule engine incorporating power industry rules is constructed to verify equipment connection relationships, voltage level consistency, and other aspects. Cross-system cross-checks ensure data consistency across different platforms and intercept logically erroneous data.

[0045] A2. Perform semantic analysis on the task descriptions in the operation data to identify and correct errors in the data entry, and verify the operation time through an intelligent verification algorithm.

[0046] Intelligent verification algorithms are used to validate operation times. In power systems, operation times may involve verifying the time sequence of multiple events, checking whether the operation time is reasonable, and whether it conflicts with other events.

[0047] (1) Time consistency check: For each operation record, let the start time be... The end time is Then check ; Time conflict detection: For two operations A and B, if they involve the same device or loop, check if the time intervals overlap. If they overlap, time merging is required, and time complements are added.

[0048] Dynamic Time Warping (DTW) Algorithm: DTW is used to compare the similarity of two time series and detect temporal logical conflicts. Let two time series... and These represent timestamp sequences (e.g., operation start time or end time) from scheduling records, operation tickets, or work tickets. DTW is calculated through the following steps: ① Distance Matrix Construction: Calculate the Euclidean distance (or other distance metric) between every pair of points in the two sequences to form a distance matrix D, where the elements... .

[0049] in, For sequence The i-th data point in Let be the i-th data point in sequence Y.

[0050] ② Cumulative distance matrix calculation: Solve for the minimum cumulative path and the cumulative distance matrix using dynamic programming. The element calculation is as follows: in, express and, The distance between them This represents the cumulative distance traveled horizontally from the left. () represents the cumulative distance traveled vertically from above. This represents the cumulative distance traveled from the top left diagonal.

[0051] Initialization conditions: , (for ), (for ) ③DTW distance: The final DTW distance is If the DTW distance exceeds the threshold (obtained based on historical data training), a logical conflict in the time series is determined, triggering manual review and correction.

[0052] In this application embodiment, the graph computation in A1 is a graph neural network, which specifically includes: constructing a substation circuit topology model based on the graph neural network; realizing intelligent verification and error correction of equipment connection relationships through node embedding and relation reasoning; abstracting the ledger equipment as graph nodes and the electrical connection relationship as edges; and using the neighborhood information aggregation mechanism of the graph neural network to detect topological logic contradictions and generate correction suggestions.

[0053] In one alternative implementation, graph computation can be based on the inductive learning framework GraphSAGE. First, an adjacency matrix of device connectivity is constructed. Then, multi-hop neighborhood information is aggregated through GraphSAGE's message passing mechanism. Finally, topological inconsistencies are detected based on the similarity calculation of node embedding vectors, thereby achieving automatic topological verification of newly connected devices.

[0054] In another alternative implementation, graph computing can be a hybrid architecture combining a graph attention network (GAT) and a knowledge graph. This integrates professional knowledge such as wiring rules and equipment parameter constraints from power industry standards into the knowledge graph in the form of triples, constructing a power topology knowledge base. In practice, the GAT module is responsible for learning the dynamic dependencies between devices, while the knowledge graph module provides rule constraints; the two work collaboratively.

[0055] Furthermore, the intelligent topology correction in A1 includes steps A11-A12: A11. Identify device element information in a primary wiring diagram based on XML algorithm.

[0056] A12. A topological graph is generated by BFS traversal, and the ledger data is compared with the topological graph by graph embedding technology to identify contradictions and trigger a closed-loop correction mechanism.

[0057] Specifically, the device nodes and connection relationships of the primary wiring diagram in SVG / CAD format are extracted based on the XML parsing algorithm, and a standardized topology map is generated by BFS traversal. The topology map is compared with the ledger data through a graph neural network to identify contradictions and trigger a closed-loop correction mechanism to ensure that the topology relationship is consistent with the actual situation.

[0058] Furthermore, the semantic analysis in A2 includes steps A21-A22: A21. Use natural language processing technology to segment and semantically analyze the work task description in order to identify and correct errors in the data entry.

[0059] A22. Use clustering algorithm and Bayesian network regression to complete missing values.

[0060] Clustering algorithms are specifically as follows Clustering algorithm.

[0061] In this embodiment of the application, the use of natural language processing technology to segment and semantically analyze the work task description in A21 involves using Jieba segmentation combined with a power industry dictionary to segment the work task description, and using BERT semantic analysis technology to identify and correct errors in the data entry. Specifically, this includes first constructing a power industry dictionary covering substation equipment terminology, operational terminology, and standard and normative terms; then using the Jieba segmenter to load the dictionary and perform domain-adaptive segmentation on the work task description to ensure complete segmentation of professional vocabulary; then using a pre-trained BERT model to convert the segmented text into semantic vectors, and analyzing the deep semantics of the text based on an attention mechanism; and finally using a dynamic time warping algorithm to compare the timestamps in the scheduling operation records, operation tickets, and work tickets to identify time logic conflicts and prompt corrections, ensuring the consistency of the time series, and generating correction suggestions based on contextual semantics.

[0062] In one alternative implementation, the word segmentation and semantic analysis of the work task description using natural language processing technology can be based on a sequence labeling architecture using an ELECTRA pre-trained model and a Conditional Random Field (CRF): ELECTRA's efficient semantic understanding capabilities are used to extract text features, and then the CRF layer is used for sequence labeling to accurately identify key elements in the work task description, such as equipment entities, operational actions, and status parameters, thereby achieving structured parsing and error detection of the reported content.

[0063] In another alternative implementation, the word segmentation and semantic analysis of the task description using natural language processing technology can also be a hybrid analysis method combining knowledge graph and semantic role labeling (SRL): first, the rationality of the equipment relationships and operational logic in the text is verified based on the knowledge graph of the power industry, and then the predicate-argument structure in the sentence is analyzed through semantic role labeling, thereby identifying deep semantic errors such as missing information and contradictory descriptions, and improving the coverage and accuracy of error identification.

[0064] In this embodiment, S2, achieving cross-system data association and integration through multi-dimensional feature matching, specifically involves achieving semantic association and integration of cross-system data through multi-dimensional feature similarity evaluation. This includes improving the Levin-Stokes distance algorithm, using power industry terminology as the smallest matching unit, to calculate the matching degree of equipment names and circuit identifiers, thus achieving accurate association between equipment and circuits. An ALBERT pre-trained model is used to extract semantic feature vectors from long texts such as outage event descriptions and fault reports. Cosine similarity is calculated to evaluate the semantic association of the texts and identify synonymous events.

[0065] Improved Levenst Distance Algorithm: The basic idea is to address the shortcomings of the Levenst distance algorithm by using electrical engineering terminology as the smallest matching unit, rather than calculating individual characters. ① Preprocessing: Word segmentation of power industry terminology. First, a power industry dictionary is constructed. It contains standardized vocabulary units such as equipment terminology, voltage levels, and wiring methods, and the input text is segmented based on this dictionary; ② Levinst distance calculation at the lexical level: The calculation is the edit distance between two word sequences. .

[0066] Editing operations include: inserting a word, deleting a word, and replacing a word.

[0067] Key improvement: The cost of replacing a word is no longer a fixed 1, but is intelligently determined based on the semantics of electricity. If two words are exactly the same (e.g., "primary variable" and "primary variable"), =0.

[0068] If the two words are synonyms for electricity (such as "circuit breaker" and "switch"), =0.1 (a very low cost).

[0069] If the two words are completely different (such as "circuit breaker" and "transformer"), =1.

[0070] ③ Calculate the final matching degree Match score = 1 - (Total editing cost / Length of the longer of the two sequences) Where m and n are the lengths of sequences A and B, respectively, and the closer the score is to 1, the more similar the two device names are.

[0071] In one alternative implementation, the association and integration of cross-system data through multi-dimensional feature matching can be achieved using a multimodal semantic fusion method based on Sentence-BERT. Specifically, the Sentence-BERT model is used to semantically encode multi-source heterogeneous data such as equipment ledger text, operation logs, and historical work orders to generate a unified semantic vector representation; a semantic association network between equipment entities and operation events is established by calculating cross-modal semantic similarity; and cross-source information propagation is carried out by combining graph neural networks.

[0072] In another alternative implementation, cross-system data association and integration can be achieved through multi-dimensional feature matching by combining a multi-source alignment framework that integrates knowledge graph embedding and attention mechanisms: First, a power domain knowledge graph containing entities such as equipment, circuits, and fault types is constructed, and embedding algorithms such as TransE are used to learn vector representations of entities and relationships; then, a cross-system attention mechanism is introduced to dynamically calculate the association weights of entity descriptions in different data sources; finally, vector space alignment and semantic disambiguation are performed.

[0073] In S2, cross-system data association and integration are achieved through multi-dimensional feature matching, and the output fused dataset includes steps B1-B2: B1. The device text is split into power industry terminology units using word segmentation technology, and the matching degree between terminology units is calculated. The smallest matching unit is power industry terminology.

[0074] B2. Use the ALBERT model to extract semantic feature vectors of the text, and evaluate the semantic similarity of the text by calculating the cosine similarity between the vectors.

[0075] Specifically, a "device-loop-event" ternary relationship model is constructed based on knowledge graph technology. The ledger data and operation data are fused through similarity association results to generate a unified dataset containing a 16-bit unique loop code, thereby achieving the organic integration of multi-source data.

[0076] In this embodiment, S3, which uses a natural language processing model to determine the circuit to which the outage event belongs and the event type, specifically uses a deep semantic understanding model to determine the circuit to which the outage event belongs and intelligently determines the event type. This includes employing an ALBERT-Seq2Seq dual encoder-decoder architecture: the Encoder layer captures the contextual semantics of the power text through a bidirectional Transformer; the Bi-LSTM encoder further extracts temporal features; and an attention mechanism is introduced to weight key entities, strengthening the device-circuit association modeling. The Decoder layer outputs multi-label classification results, including circuit type and outage nature.

[0077] A transfer learning strategy is adopted to freeze the underlying parameters of the ALBERT pre-trained model to retain its general semantic understanding capabilities. The upper-layer network is adjusted using labeled power data to adapt to entities and topological relationships in the power domain. Focal Loss is used to solve the problem of sample class imbalance, and the training set is expanded through data augmentation to improve the model's classification accuracy.

[0078] In one alternative implementation, using a natural language processing model to determine the circuit to which an outage event belongs and the event type can be based on a joint learning architecture of the pre-trained language model BERT and a graph neural network. Specifically, firstly, the BERT model is used to perform deep semantic encoding on the text describing the outage event to obtain text feature representations containing rich contextual information; simultaneously, a device relationship graph is constructed based on the power grid topology, and the embedding representation of device nodes in the topological space is learned through a graph neural network; finally, the text features and graph structure features are fused across modalities, and a multilayer perceptron is used to achieve joint classification of circuit attribution and event type.

[0079] In another alternative implementation, using a natural language processing model to determine the circuit attribution and event type of an outage event can be achieved through a few-shot learning framework that combines cue learning and domain adaptation. This framework first constructs a cue template incorporating knowledge of the power domain, transforming the outage event classification task into a masked language modeling problem, fully utilizing the internal knowledge of the pre-trained language model. Simultaneously, it employs domain adaptation techniques, enabling the model to quickly adapt to terminology differences and recording standards across different regions through comparative learning on source domain data and a small amount of labeled data in the target domain. Finally, it jointly optimizes the two related tasks of circuit attribution identification and event type classification through multi-task learning.

[0080] Furthermore, in S3, the use of a natural language processing model to determine the circuit to which the outage event belongs and the event type includes steps C1-C3: C1. The ALBERT-Seq2Seq model is adopted. The contextual semantics are encoded by bidirectional Transformer and the contextual features are extracted by Bi-LSTM encoder. An attention mechanism is introduced to weight and focus the equipment operation description, so as to classify the multi-label loops of the shutdown event and judge the event type.

[0081] Specifically, an ALBERT-Seq2Seq dual encoder-decoder architecture is adopted: the encoder layer captures the contextual semantics of the power text through a bidirectional Transformer; the Bi-LSTM encoder further extracts temporal features; and an attention mechanism is introduced to assign weights to key entities, strengthening the modeling of equipment-loop associations. The decoder layer outputs multi-label classification results, including loop type and outage nature.

[0082] C2. Event types include scheduled shutdown, affected shutdown, planned shutdown, and forced shutdown.

[0083] The C3 and ALBERT-Seq2Seq models employ a transfer learning strategy, freezing the underlying parameters of the pre-trained model to retain general semantic understanding capabilities, and adjusting the upper-layer network to adapt to entities and topological relationships in the power industry.

[0084] Specifically, a transfer learning strategy is adopted to freeze the underlying parameters of the ALBERT pre-trained model to retain its general semantic understanding capabilities, and to adjust the upper-layer network using labeled power data to adapt to entities and topological relationships in the power domain. Focal Loss is used to address the problem of sample class imbalance, and the training set is expanded through data augmentation to improve the model's classification accuracy.

[0085] Furthermore, step S4 involves comprehensive analysis of the operational data, completing loop status classification and outage statistics, and generating loop reliability operational data, including steps D1-D3: D1. Filter the operation data, remove data below 110kV and data related to non-substation circuits, split multiple operation contents according to punctuation marks or serial numbers, and extract key information.

[0086] Specifically, multiple work items are split using punctuation marks or serial numbers as delimiters to ensure that each record corresponds to a single circuit operation; key information such as substation name, equipment set, and operation time are extracted from the parsed text and linked to the equipment topology relationship in the ledger data through 16-bit circuit codes.

[0087] D2. Divide the circuit state into operating state and shutdown state. The shutdown state includes scheduled shutdown, affected shutdown, planned shutdown and forced shutdown. The state is determined based on whether the circuit can complete the power transmission function within a specific capacity range.

[0088] D3. Based on the loss of circuit function, statistically analyze outage events. For different wiring methods of main transformer circuits, outgoing circuits, and bus circuits, determine the corresponding conversion factor according to the magnitude of the impact of outage on circuit function, and generate circuit reliability operation data that includes circuit code, outage nature, conversion factor, and cause of responsibility.

[0089] Specifically, the outage count and start / end time are recorded according to circuit type, and only outage events of related circuits caused by the bus circuit itself are counted. For different wiring scenarios, a conversion factor is calculated according to the rule of "number of out-of-service circuit breakers / total number of circuit breakers" to quantify the impact of partial outages on circuit functions.

[0090] For partial outage events, a coefficient, denoted by λ, is used to determine the impact of the outage on the circuit's function. This conversion factor only affects the calculated results and should not be used to convert time during event statistics; 0 < λ ≤ 1. The conversion formula is as follows: Main transformer circuit conversion factor: Note: The power outage conversion factor for the high-voltage side of the 110kV main transformer circuit is 1, and the power outage conversion factors for the high-medium-voltage side of the 220kV main transformer circuit are 1 / 2 respectively.

[0091] Outgoing circuit conversion factor: λ=1 Bus circuit conversion factor: Example 3, referring to Figure 2 This embodiment of the present invention provides a multi-source data fusion system for substation circuit reliability based on natural language processing technology, including: a data governance and processing module, a semantic analysis and feature engineering module, an intelligent judgment and classification module, and an operation data analysis and statistics module.

[0092] The data governance and processing module intelligently manages the ledger data and operational data of substation circuits, builds a standardized data foundation, and outputs datasets.

[0093] The semantic analysis and feature engineering module, based on the dataset, achieves semantic association and integration of cross-system data through multi-dimensional feature similarity evaluation, and outputs a fused dataset.

[0094] The intelligent analysis and classification module, based on the fused dataset, uses a deep semantic understanding model to determine the circuit to which the outage event belongs and intelligently determines the event type.

[0095] The operation data analysis and statistics module performs a full-dimensional analysis of the operation data based on the event type judgment results, completes the circuit status classification and outage statistics, and generates circuit reliability operation data.

[0096] This embodiment also provides an electronic device applicable to a method for fusing multi-source data on the reliability of substation circuits based on natural language processing technology, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method for fusing multi-source data on the reliability of substation circuits based on natural language processing technology as proposed in the above embodiment.

[0097] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a multi-source data fusion method for substation circuit reliability based on natural language processing technology as proposed in the above embodiment.

[0098] The storage medium proposed in this embodiment belongs to the same inventive concept as the multi-source data fusion method for substation circuit reliability based on natural language processing technology proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0099] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0100] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for fusing multi-source data on substation circuit reliability based on natural language processing technology, characterized in that: include, Intelligent management of ledger data and operational data of substation circuits; construction of a standardized data foundation; output of datasets. Based on the dataset, cross-system data is associated and integrated through multi-dimensional feature matching, and a fused dataset is output. Based on the fused dataset, a natural language processing model is used to determine the circuit to which the outage event belongs and the event type; Based on the event type analysis results, the operational data is analyzed in all dimensions to complete the circuit status classification and outage statistics, and generate circuit reliability operational data.

2. The method for fusing multi-source data on substation circuit reliability based on natural language processing technology as described in claim 1, characterized in that: The intelligent management of substation circuit ledger data and operational data, and the construction of a standardized data foundation, includes: Regular expressions and ETL tools are used to standardize the ledger data, graph computing is used to achieve intelligent error correction of topology, and a rule engine is used to complete the ledger association verification. Semantic analysis is performed on the task descriptions in the operational data to identify and correct errors in the data entry, and the operation time is verified through an intelligent verification algorithm.

3. The method for fusing multi-source data on substation circuit reliability based on natural language processing technology as described in claim 2, characterized in that: The intelligent error correction of the topology includes, Identifying device element information in a primary wiring diagram based on XML algorithm; A topology graph is generated by using BFS traversal, and the ledger data is compared with the topology graph using graph embedding technology to identify contradictions and trigger a closed-loop correction mechanism.

4. The method for fusing multi-source data on substation circuit reliability based on natural language processing technology as described in claim 3, characterized in that: The semantic analysis includes, Natural language processing technology is used to segment and semantically analyze the description of work tasks in order to identify and correct errors in the data entry. Clustering algorithms and Bayesian network regression were used to complete missing values.

5. The method for fusing multi-source data on substation circuit reliability based on natural language processing technology as described in claim 4, characterized in that: The method of achieving cross-system data association and integration through multi-dimensional feature matching, and the output fused dataset includes, The device text is divided into power-related professional terminology units using word segmentation technology. The matching degree between the terminology units is calculated, with the smallest matching unit being a power-related professional term. The semantic feature vectors of the text are extracted using the ALBERT model, and the semantic similarity of the text is evaluated by calculating the cosine similarity between the vectors.

6. The method for fusing multi-source data on substation circuit reliability based on natural language processing technology as described in claim 4, characterized in that: The method of using natural language processing models to determine the circuit to which the outage event belongs and the event type includes... The ALBERT-Seq2Seq model is adopted, which encodes the contextual semantics through bidirectional Transformer and extracts contextual features by combining Bi-LSTM encoder. An attention mechanism is introduced to weight and focus the equipment operation description, so as to classify the multi-label loops and judge the event type of shutdown events. The event types include scheduled shutdown, affected shutdown, planned shutdown, and forced shutdown; The ALBERT-Seq2Seq model employs a transfer learning strategy, freezing the low-level parameters of the pre-trained model to retain general semantic understanding capabilities, and adjusting the upper-level network to adapt to entities and topological relationships in the power industry.

7. The method for fusing multi-source data on substation circuit reliability based on natural language processing technology as described in claim 4, characterized in that: The process involves comprehensive analysis of operational data to classify circuit states and compile outage statistics, generating circuit reliability operational data, including... The operational data was filtered out, data below 110kV and data related to non-substation circuits were removed, and multiple work contents were split according to punctuation marks or serial numbers to extract key information. The circuit status is divided into operating status and shutdown status. The shutdown status includes scheduled shutdown, affected shutdown, planned shutdown and forced shutdown. The status is determined based on whether the circuit can complete the power transmission function within a specific capacity range. Based on the loss of circuit function, outage events are statistically analyzed. For different wiring methods of main transformer circuits, outgoing circuits, and bus circuits, corresponding conversion factors are determined according to the magnitude of the impact of outage on circuit function, generating circuit reliability operation data that includes circuit code, outage nature, conversion factor, and cause of responsibility.

8. A substation circuit reliability multi-source data fusion system based on natural language processing technology, employing the substation circuit reliability multi-source data fusion method based on natural language processing technology as described in any one of claims 1 to 7, characterized in that, include: The module includes data governance and processing, semantic analysis and feature engineering, intelligent judgment and classification, and runtime data analysis and statistics. The data governance and processing module intelligently governs the ledger data and operation data of the substation circuit, constructs a standardized data foundation, and outputs datasets. The semantic analysis and feature engineering module, based on the dataset, achieves semantic association and integration of cross-system data through multi-dimensional feature similarity evaluation, and outputs a fused dataset; The intelligent analysis and classification module, based on the fused dataset, uses a deep semantic understanding model to analyze the circuit to which the outage event belongs and intelligently analyzes the event type; The operational data analysis and statistics module performs a full-dimensional analysis of the operational data based on the event type judgment results, completes the circuit status classification and outage statistics, and generates circuit reliability operational data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the multi-source data fusion method for substation circuit reliability based on natural language processing technology as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-source data fusion method for substation circuit reliability based on natural language processing technology as described in any one of claims 1 to 7.

Images