A hierarchical constraint fuzzy positioning power operation record association method and system
Patent Information
- Application Number
- CN202611432979.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-16
- Publication Date
- 2026-10-30
AI Technical Summary
[0004]为了解决现有技术中跨源数据格式异构且缺乏统一匹配框架、匹配精度与效率难以兼顾、设备编号标识作用未充分利用、以及实时一致性校验手段缺失的问题，本申请设计了一种分层约束模糊定位的电力操作记录关联方法及系统，通过构建由粗到精的分层约束模糊定位引擎，融合设备编号倒排索引、字符级特征匹配、位置感知加权和多维融合评分等技术手段，以求在消除数据格式壁垒的同时，大幅提升关联定位的效率和准确性，强化设备编号的标识作用，并支持事后批量关联与实时一致性校验的双重应用
[0045]1、采用三层分层约束递进架构，兼顾匹配效率与精度：通过候选域过滤层利用设备编号倒排索引与字符级相似度快速将候选集缩小至全量的12%左右，第二层在有限集内执行高精度模糊匹配和位置加权定位候选行，第三层精确定位单元格。该架构将全局扫描平均耗时由3200ms降至85ms，效率提升37.6倍，同时匹配F1值达0.89，解决了现有技术中精度与效率不可兼得的问题。
Smart Images
Figure CN123012079A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of power system dispatch automation and data governance technology, specifically relating to a hierarchical constraint fuzzy positioning method and system for associating power operation records. Background Technology
[0002] Power dispatching systems generate a large amount of heterogeneous data during daily operation and emergency repairs, mainly including repair records, switching operation logs, dispatching instructions, and graphical operation records. This data is scattered across different systems, with inconsistent field definitions and naming conventions, making automatic correlation difficult for post-event analysis and heavily reliant on manual verification.
[0003] In existing technologies, exact matching-based methods are sensitive to naming differences and have low recall rates; methods based on TF-IDF and cosine similarity have sparse features in short text scenarios and are insufficient in distinguishing high-frequency words in the power sector; semantic matching methods based on deep learning have high computational complexity, high domain adaptation costs, and poor interpretability, making it difficult to meet the real-time response requirements of scheduling systems. In addition, existing methods generally lack the utilization of equipment number sensitivity, cannot balance the accuracy and efficiency of cross-source data matching, and have not formed an effective automated means of consistency verification. Summary of the Invention
[0004] To address the problems in existing technologies, such as heterogeneous cross-source data formats, lack of a unified matching framework, difficulty in balancing matching accuracy and efficiency, underutilization of the identification function of equipment numbers, and lack of real-time consistency verification methods, this application designs a hierarchical constraint fuzzy positioning method and system for associating power operation records. By constructing a hierarchical constraint fuzzy positioning engine from coarse to fine, it integrates techniques such as inverted index of equipment numbers, character-level feature matching, location-aware weighting, and multi-dimensional fusion scoring. This aims to significantly improve the efficiency and accuracy of association positioning while eliminating data format barriers, strengthening the identification function of equipment numbers, and supporting the dual application of post-event batch association and real-time consistency verification.
[0005] To achieve the above objectives, the present invention adopts the following technical solution:
[0006] A hierarchical constrained fuzzy localization method for associating power operation records includes the following steps:
[0007] S1. Obtain emergency repair record text and scheduling operation log, generate description text based on the emergency repair record text, extract the device number from the description text and construct an inverted index of the device number, and perform character-level feature segmentation on the description text and the device name in the scheduling operation log.
[0008] Preferably, extracting the device number and constructing the inverted index specifically includes: based on a preset number extraction rule, extracting continuous numeric strings and / or alphanumeric combinations from the description text as device number tokens; when the candidate number's adjacent context contains any power equipment keyword such as switch, disconnector, line, or main transformer, it is determined to be a valid device number token, and a mapping is established between the token and the operation log record index.
[0009] S2. Based on the inverted index and character-level features, the operation log is filtered for candidate domains to obtain a candidate record set. The candidate domain filtering includes: obtaining a first candidate set using the inverted index of the device number, obtaining a second candidate set by coarsely screening the full operation log using the character-level similarity between the description text and the device name, and determining the search range according to the preset power system rule constraints. The intersection of the three filtering results is used to obtain the final candidate record set.
[0010] Preferably, obtaining the first candidate set using the inverted index includes: querying the inverted index for each token to obtain its respective record set, prioritizing the intersection as the first candidate set, and taking the union when the intersection is empty. Character-level similarity uses Jaccard similarity after bi-gram segmentation; power system rule constraints are based on a rule base constructed according to naming rules such as substation names, equipment types, and voltage levels. The rule base includes regular expression rules such as "220kV line names containing voltage level prefixes" and "switching station equipment names containing station name fields," which are matched according to priority to limit the search range.
[0011] S3. Within the candidate record set, calculate the basic similarity between the device name and the description text in each candidate record using fuzzy matching, and calculate the position-weighted score based on the position distribution of the matching characters in the device name. Combine the basic similarity and position-weighted score using a weighted summation method to filter and locate candidate rows. Extract the title information of the title row from the candidate rows back up, and perform set similarity matching in combination with the target device type set corresponding to the description text to determine the column candidate region.
[0012] Preferably, the calculation of the position-weighted score specifically includes: recording the index position of the matching character in the candidate device name, calculating the average position of the matching character and normalizing it to obtain the basic position score;
[0013] If the average position falls within the first or last third of the text, the base position score is increased by a first preset weight; if the number of consecutive unmatched characters exceeds a preset threshold, the base position score is reduced by a second preset weight; after adjustment, the score is truncated to the [0,1] interval and used as the final position-weighted score.
[0014] Furthermore, if the average position falls within the first or second third of the text, the base position score is increased by 0.1; if the number of consecutive unmatched characters exceeds 3, the base position score is decreased by 0.05.
[0015] The basic similarity and position-weighted scores are fused using a weighted summation method.
[0016] S4. Within the candidate column area, perform column name similarity matching to determine the target column, and within the data range corresponding to the candidate column area, perform fuzzy matching and position weighting again based on the description text and device name to determine the target row, thereby achieving two-dimensional coordinate positioning of the target cell.
[0017] S5. Calculate the comprehensive score based on the multidimensional fusion model, which integrates the following factors: character-level similarity factor between description text and device name, column name similarity factor, inverted index hit weighting factor, location weighting factor, site relevance factor, time factor, and key device weighting factor.
[0018] Preferably, the multidimensional fusion model is a product model:
[0019] Score = Base × Context;
[0020] in:
[0021] Base=Jaccard bg a1 ×Jaccard col a2 ×IDF boost a3 ,
[0022] Context=station factor ×time factor ×position factor a4 ×device hit_weight ;
[0023] In the formula: Score is the overall matching score; Base is the basic matching score; Context is the context correction term; Jaccard bg Character-level similarity factor; Jaccard col IDF is the column name similarity factor. boost To hit the weighting factor for the inverted index; station factor For site relevance factors; time factor For time factor; position factor Location-weighted factor; devicehit_weight a1, a2, a3, and a4 are the weighting factors for key equipment; a1, a2, a3, and a4 are preset weighting indices that can be determined through experimental optimization.
[0024] Preferably, the time factor is determined using a piecewise linear function based on the time difference Δt between the emergency repair record time and the operation log time:
[0025]
[0026] In the formula: Δt is the time difference between the emergency repair record and the operation log; min is the time unit in minutes; v1, v2, and v3 are the baseline coefficients of the three-segment function, with typical values of v1=1.0, v2=0.7, and v3=0.3.
[0027] When the time difference is no greater than 10 minutes, the first value v1 is taken;
[0028] When the time interval is between 10 minutes and 1 hour, the value decreases linearly from the first value v1 to the second value v2.
[0029] When the time interval is between 1 hour and 24 hours, the value decreases linearly from the second value v2 to the third value v3.
[0030] If more than 24 hours have passed, the third value v3 will be used.
[0031] S6. Based on whether a device number is matched during the candidate domain filtering stage, dynamically adjust the judgment threshold and output the association result based on the comparison result between the comprehensive score and the judgment threshold.
[0032] Preferably, based on a baseline threshold, if a device number matches during the candidate domain filtering stage, the decision threshold is set to the baseline threshold multiplied by a first coefficient; if no device number matches, the decision threshold is set to the baseline threshold multiplied by a second coefficient; the first coefficient is greater than 1, and the second coefficient is less than 1. A typical value for the baseline threshold is 0.25, a typical value for the first coefficient is 1.5, and a typical value for the second coefficient is 0.8.
[0033] Furthermore, this is based on a baseline threshold of 0.25.
[0034] Preferably, the output association result includes three levels of judgment: when the comprehensive score is greater than or equal to the first association threshold, it is judged as consistent association; when the comprehensive score is between the judgment threshold and the first association threshold, it is judged as an association warning; when the comprehensive score is less than the judgment threshold, it is judged as inconsistent association. The typical value of the first association threshold is 0.6.
[0035] Preferably, the method is further used for real-time consistency verification of scheduling instructions and graphic operation records. During verification, the scheduling instruction text is used as the input source of the description text, and the graphic operation records are used as the data source of the operation log to be matched. Steps S1 to S6 are executed to obtain the consistency judgment result.
[0036] Based on the above design, the present invention also designs a hierarchical constrained fuzzy positioning power operation record association system, including a data preprocessing module, a candidate domain filtering module, a multi-dimensional positioning module, a coordinate analysis module, a scoring module, and a judgment module;
[0037] The data preprocessing module is used to obtain emergency repair record text and scheduling operation log, generate descriptive text based on emergency repair record text, extract the device number from the descriptive text and build an inverted index, and perform character-level feature segmentation on the device name in the descriptive text and operation log.
[0038] The candidate field filtering module is used to filter the operation log based on the inverted index and character-level features, and to obtain a candidate record set by taking the intersection of the three filtering results.
[0039] The multi-dimensional positioning module is used to locate candidate rows within the candidate record set by filtering and locating them through fuzzy matching and position weighting, and to determine the candidate column region by backtracking title information and set similarity matching.
[0040] The coordinate parsing module is used to perform column name similarity matching within the column candidate area to determine the target column, and then perform fuzzy matching and position weighting within the corresponding data range to determine the target row, thereby achieving two-dimensional coordinate positioning of the target cell;
[0041] The scoring module is used to calculate a comprehensive score based on a multidimensional fusion model, which integrates character-level similarity factors, column name similarity factors, inverted index hit weighting factors, location weighting factors, site relevance factors, time factors, and key equipment weighting factors.
[0042] The judgment module is used to dynamically adjust the judgment threshold based on whether the device number is matched, and output the correlation result based on the comprehensive score and the judgment threshold.
[0043] Based on the above design, the present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned association method.
[0044] The advantages and effects of this application are as follows:
[0045] 1. A three-layer hierarchical constraint-progressive architecture is adopted to balance matching efficiency and accuracy: The candidate field filtering layer uses the device number inverted index and character-level similarity to quickly narrow down the candidate set to about 12% of the total. The second layer performs high-precision fuzzy matching and position-weighted location of candidate rows within the finite set. The third layer precisely locates cells. This architecture reduces the average global scan time from 3200ms to 85ms, improving efficiency by 37.6 times, while achieving a matching F1 score of 0.89, thus solving the problem of the trade-off between accuracy and efficiency in existing technologies.
[0046] 2. Constructing an inverted index of equipment numbers and a multi-token combination strategy to enhance the role of equipment number identification: Leveraging the unique identification characteristic of power equipment numbers, an inverted index is constructed by extracting pure numeric and alphanumeric mixed numbers, and a multi-token combination strategy with intersection priority and union fallback is adopted. This solution fully utilizes the unique identification characteristic of equipment numbers, increasing the accuracy of key equipment matching from 78% to 94%, significantly improving the accuracy of associated location.
[0047] 3. Introducing a location-aware weighted mechanism to improve the discriminative power of short text matching: By recording the distribution of matching characters in device names, a normalized position score is calculated and combined with first-and-last clustering weighting and mismatch deduction rules to correct similarity. This mechanism effectively overcomes interference from common words in short power-related texts and improves matching accuracy in short text scenarios.
[0048] 4. Construct a multi-dimensional fusion product scoring model adapted to the characteristics of power business: A product model is constructed by integrating three basic features—character similarity, column name similarity, and IDF weighting—with four contextual factors: site relevance, temporal proximity, location factor, and equipment weight. Ablation experiments show that removing any factor leads to a significant decrease in the F1 score, and each factor contributes positively to the score. The model is highly adapted to the patterns of power dispatching business.
[0049] 5. An adaptive threshold and three-level verification mechanism are designed to support multiple application scenarios: the threshold is dynamically switched based on the presence or absence of a device number. Accuracy is strictly guaranteed when a device number is present, while recall is relaxed under boundary conditions without a device number, preventing both false negatives and false negatives. The same engine can be used for batch post-event correlation and real-time consistency verification of scheduling commands and graphical operations, with a real-time response time of less than 100ms and strong scalability.
[0050] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings.
[0051] The above and other objects, advantages and features of this application will become more apparent to those skilled in the art from the following detailed description of specific embodiments in conjunction with the accompanying drawings. Attached Figure Description
[0052] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.
[0053] Figure 1 This is a flowchart illustrating the steps of the hierarchical constraint fuzzy positioning method for associating power operation records according to the present invention. Detailed Implementation
[0054] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. In the following description, specific details such as specific configurations and components are provided merely to help fully understand the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. In addition, for clarity and brevity, descriptions of known functions and structures are omitted in the embodiments.
[0055] It should be understood that the phrase "an embodiment" or "this embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "an embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.
[0056] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.
[0057] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another type of relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the related objects before and after it are in an "or" relationship.
[0058] In this article, the term "at least one" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, "at least one of A and B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.
[0059] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion.
[0060] The core of this invention is the HCFL (Hierarchical Constraint Fuzzy Localization) engine, which achieves intelligent alignment and localization of power operation data from different sources through a technical chain of "data input - feature construction - hierarchical constraint localization - multi-dimensional fusion evaluation - verification output".
[0061] Figure 1 The overall execution flow of this method is shown, corresponding one-to-one with steps S1-S6. The whole process is divided into six stages: data preprocessing, candidate domain filtering, multidimensional fuzzy localization, two-dimensional coordinate analysis, multidimensional fusion scoring, and adaptive threshold determination. The connection between each stage realizes hierarchical constraint localization from coarse to fine.
[0062] (a) Data preprocessing stage
[0063] The system receives emergency repair records and switching operation logs. The preprocessing module first standardizes the text, including removing spaces, unifying full-width and half-width characters, and standardizing capitalization. It then concatenates the station name, equipment name, operation description, and other fields of the emergency repair record into a complete descriptive text.
[0064] The device names in the description text and operation logs are bigrammed to generate a character set for subsequent similarity calculation. Simultaneously, device ID tokens are extracted: consecutive numeric strings of 2-8 characters and consecutive combinations of letters and numbers are extracted as candidate IDs; if the adjacent context of a candidate ID contains any of the following keywords related to power equipment: switch, disconnector, line, main transformer, busbar, pole transformer, or station service transformer, it is considered a valid device ID token.
[0065] An inverted index from device number tokens to record indexes is built based on the full operation logs. At the same time, the inverse document frequency (IDF) of each token is calculated to build a set of station name tokens, laying the foundation for subsequent hierarchical positioning.
[0066] (II) First layer: Candidate domain filtering
[0067] The first layer performs three-way filtering in parallel to quickly narrow down the data range:
[0068] ID Inverted Index Fast Filtering: Query the inverted index using the extracted device ID token to obtain the corresponding set of records. When multiple tokens exist, prioritize the intersection of these sets as the first candidate set to ensure high confidence; if the intersection is empty, use the union as a fallback to avoid missed matches.
[0069] Bi-gram Jaccard coarse filtering: Calculates the Jaccard similarity between the descriptive text and the bi-gram sets of each device name, retaining records with similarities above a preset threshold as a second candidate set, achieving fast coarse filtering. The bi-gram Jaccard similarity is the ratio of the intersection to the union of two sets of bi-gram characters.
[0070] Rule-based filtering: Matching is based on a pre-built power system rule base, which is constructed hierarchically: the first layer is voltage level rules, matching the voltage level field in the description text to limit equipment records at the corresponding voltage level; the second layer is site name rules, matching the site name field in the description text to limit equipment records at the corresponding site; the third layer is equipment type rules, matching the equipment type field in the description text to limit equipment records of the corresponding type. Rule priority is sorted from high to low as voltage level > site name > equipment type. When a high-priority rule is matched, a low-priority rule is not matched. The rule base includes regular expression rules such as "220kV line name containing voltage level prefix" and "switching station equipment name containing site name field". After a rule is matched, the search scope is limited to a specific line number range; if no rule is matched, the scope is expanded level by level until a full scan mode is triggered.
[0071] The intersection of the three results forms a significantly reduced set of candidate records, and at the same time, it records whether the device number was matched in this match.
[0072] (III) Second layer: Multidimensional fuzzy positioning
[0073] Perform fine sorting on each candidate record within the candidate record set:
[0074] The RapidFuzz algorithm is used to calculate the basic similarity FuzzyScore between candidate device names and description text, with a value range of [0, 100].
[0075] Calculate the position-weighted score PositionScore: Record the index position of all matching characters in the candidate device name, calculate the average position avg_pos, divide by the text length to obtain the normalized basic position score; if avg_pos is located in the first or second third of the text, the score increases by 0.1; if the number of consecutive unmatched characters exceeds 3, the score decreases by 0.05; finally, the score is truncated to the [0,1] interval.
[0076] The final matching score is obtained by weighted fusion: FinalScore = 0.8 × FuzzyScore + 0.2 × PositionScore × 100, where: FinalScore is the final fusion score for row matching; FuzzyScore is the basic similarity score of RapidFuzz; and PositionScore is the position-weighted score.
[0077] Candidate rows are obtained by sorting and filtering according to scores.
[0078] Then, title backtracking and set matching are performed: Backtracking upwards line by line from each candidate row, the row following the first empty row is determined as the title row; when encountering multi-level titles, the level closest to the candidate row is selected, cells are merged and the displayed value is taken, and if more than three consecutive empty rows are encountered, a full column scan is triggered. All non-empty column names of the title row are extracted to form a title set. The target device type set corresponding to the description text is then matched with the title set using Jaccard similarity to lock in the column candidate region.
[0079] (iv) Third layer: Two-dimensional coordinate analysis
[0080] Complete precise cell location within the candidate list area:
[0081] Column name matching to determine Y: After removing the fixed suffix from each column name in the candidate column region, calculate the Jaccard similarity with the target type, and determine the column with the highest score as the target column Y.
[0082] Precise positioning of row X: Within the data rows below the header row, the RapidFuzz algorithm is used again to calculate the fusion score using position weighting, and the smallest row with the highest score and a row number greater than the header row is taken as the target row X.
[0083] Finally, the target cell coordinates (X,Y) are located, and data such as device status within the cell are extracted.
[0084] (V) Multidimensional Fusion Scoring and Adaptive Threshold Determination
[0085] The overall score for the positioning results is calculated using a product model:
[0086] Base=Jaccard bg 0.5 ×Jaccard col 0.3 ×IDF boost 0.5 ;
[0087] Context=station factor ×time factor ×position factor 0.2 ×device hit_weight ;
[0088] Score = Base × Context;
[0089] Meaning and value rules of each factor:
[0090] Jaccard bg : Describes the bi-gram Jaccard similarity between the text and the device name, with a value range of [0,1];
[0091] Jaccard col : Jaccard similarity during the column name matching phase, with a value range of [0,1];
[0092] IDF boost The inverse document frequency weighted average of the numbered tokens is the ratio of the sum of the IDF of the intersection tokens to the sum of the IDF of all tokens, with a value range of [0,1].
[0093] station factor The site relevance factor is calculated by pre-extracting all substation and switch station names from the full operation log to construct a set of station name tokens; calculating the Jaccard similarity between the station name field in the description text and the station name field of the candidate records, and linearly mapping the similarity value to the interval [0.5, 1.1]. A similarity of 1 corresponds to a factor value of 1.1, and a similarity of 0 corresponds to a factor value of 0.5, with a value range of [0.5, 1.1].
[0094] time factor Time proximity factor, using a piecewise linear function, with typical baseline coefficients v1=1.0, v2=0.7, v3=0.3;
[0095] When the time difference Δt is no greater than 10 minutes, v1 is used; when it is between 10 minutes and 1 hour, it linearly decreases from v1 to v2; when it is between 1 hour and 24 hours, it linearly decreases from v2 to v3; when it exceeds 24 hours, v3 is used.
[0096]
[0097] Where: Δt is the time difference between the operation time and the reference time, in minutes;
[0098] position factor : Position weighting factor, which is the weighted average of the position scores of column name matching and row number matching, with a value range of [0,1];
[0099] Key equipment is classified according to voltage level and equipment type: transmission lines, main transformers, and busbar equipment with voltage levels of 220kV and above are classified as key equipment; switches, knife switches, pole transformers, and station service transformers with voltage levels of 110kV and below are classified as ordinary equipment; records for which no specific equipment type is matched are classified as mismatched equipment.
[0100] device hit_weight The weighting factor for critical equipment is 1.2 for critical equipment, 1.0 for ordinary equipment, and 0.8 for equipment that is not hit.
[0101] The adaptive threshold rules are as follows: the baseline threshold is 0.25 (the threshold corresponding to the maximum value of F1 in historical data); the threshold for determining if a device number is matched is 0.25 × 1.5 = 0.375; the threshold for determining if a device number is not matched is 0.25 × 0.8 = 0.20. Three levels of determination are also set: a comprehensive score ≥ 0.6 indicates consistent association (green light); a determination threshold ≤ score < 0.6 indicates an association warning (yellow light, requiring manual confirmation); and a score < determination threshold indicates inconsistent association (red light).
[0102] Example 1: Normal Matching Scenario
[0103] Enter the emergency repair record: "Hongxing Home Furnishings Switch Station | 76 Shanhai Furniture B | Switch | Converted from Operation to Cooling Standby | 2026-04-06 07:05:43".
[0104] Preprocessing: Concatenate and generate the description text "Hongxing Home Furnishings Switch Station 76 Shanhai Home Furnishings Switch B from Operation to Cooling Standby", and extract the valid equipment number token {76}.
[0105] Candidate domain filtering: Record number "76" is matched in the inverted index; bigram Jaccard similarity is 0.467; the rule matches the switch station equipment rule, limiting the search range; the intersection of the three paths is used to obtain the candidate subset.
[0106] Multidimensional fuzzy localization: Calculate RapidFuzz similarity and position-weighted score within the candidate subset, sort to obtain candidate row set; backtrack upward to obtain header row, and lock column candidate region after matching column name set with device type set.
[0107] Two-dimensional coordinate analysis: column name matching locks the target column, row precise positioning locks the target row 128, and finally locks the target cell.
[0108] Scoring and Judgment: The calculated overall score is 1.127. Since the device number was matched, the judgment threshold is 0.375; the score 1.127 ≥ 0.6, so it is judged as a consistent association.
[0109] Example 2: Unnumbered Boundary Scenario
[0110] The emergency repair record only contains the operation action "close" and has no valid equipment number token.
[0111] If no valid ID is extracted during the preprocessing stage, a full scan mode is triggered, using all log records as the candidate set; the adaptive threshold is adjusted to 0.20. A comprehensive score is calculated for all logs through full matching. If the highest score is below 0.20, an inconsistency in association is output, and a message indicating a missing device ID is provided, suggesting that additional information be added.
[0112] Example 3: Real-time Consistency Verification Extension
[0113] This embodiment demonstrates real-time consistency verification between dispatch instructions and graphical operations. The input dispatch instruction is "220kV Songsi Line to switch from operation to maintenance," and the graphical operation record is "220kV Songsi Line Primary Wiring Diagram."
[0114] Using the scheduling instruction as the descriptive text and the map record as the data to be matched, the above process is executed: extract the token number {220}, locate the corresponding map record through three layers of positioning, calculate the comprehensive score (0.909 ≥ 0.6), and determine that it is a match. The system prompts that the operation can continue. The real-time verification response time is less than 100ms, which meets the real-time requirements of the scheduling system.
[0115] After testing and verification with 5,000 real-world scenario data points, the method of this invention achieved an accuracy of 93.7%, a recall rate of 92.3%, and an average end-to-end response time of 721ms. Its overall performance is significantly better than traditional solutions and approaches the level of human experts.
[0116] The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made in accordance with the claims of the present invention should be included within the scope of the present invention.
[0117] The above description is merely a preferred embodiment of the present invention and does not limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any changes, modifications, substitutions, integrations, and parameter alterations to these embodiments within the spirit and principles of the present invention, achieved through conventional substitutions or by achieving the same function without departing from the principles and spirit of the present invention, fall within the scope of protection of the present invention.
Claims
1. A hierarchical constraint fuzzy localization method for associating power operation records, characterized in that, Includes the following steps: Step S1: Obtain emergency repair record text and scheduling operation log; generate description text based on the emergency repair record text; extract the device number from the description text and construct an inverted index of the device number; and perform character-level feature segmentation on the description text and the device name in the scheduling operation log. Step S2: Based on the inverted index and character-level features, perform candidate domain filtering on the operation log to obtain a candidate record set; the candidate domain filtering includes: obtaining a first candidate set using the inverted index of the device number, obtaining a second candidate set by coarsely screening the full operation log using the character-level similarity between the description text and the device name, and determining the search range according to preset power system rule constraints, and taking the intersection of the three filtering results to obtain the final candidate record set; Step S3: Within the candidate record set, calculate the basic similarity between the device name and the description text in each candidate record using fuzzy matching, and calculate the position-weighted score based on the position distribution of the matching characters in the device name. Combine the basic similarity and position-weighted score using a weighted summation method to filter and locate candidate rows. Extract the title information of the title row from the candidate rows back up, and perform set similarity matching in combination with the target device type set corresponding to the description text to determine the column candidate region. Step S4: Within the column candidate area, perform column name similarity matching to determine the target column, and within the data range corresponding to the column candidate area, perform fuzzy matching and position weighting again based on the description text and device name to determine the target row, thereby achieving two-dimensional coordinate positioning of the target cell; Step S5: Calculate the comprehensive score based on the multidimensional fusion model, which integrates the following factors: character-level similarity factor between description text and device name, column name similarity factor, inverted index hit weighting factor, location weighting factor, site relevance factor, time factor, and key device weight factor. Step S6: Based on whether there is a device number match in the candidate domain filtering stage, dynamically adjust the judgment threshold, and output the association result based on the comparison result between the comprehensive score and the judgment threshold.
2. The method for associating power operation records using hierarchical constraint fuzzy positioning according to claim 1, characterized in that, In step S1, extracting the device number and constructing the inverted index specifically includes: based on the preset number extraction rules, extracting continuous numeric strings and / or alphanumeric combinations from the description text as device number tokens; when the candidate number's adjacent context contains any power equipment keyword such as switch, knife switch, line, or main transformer, it is determined to be a valid device number token, and a mapping is established between the token and the operation log record index. In step S2, obtaining the first candidate set using the inverted index includes: querying the inverted index for each of the multiple tokens to obtain their respective record sets, prioritizing the intersection as the first candidate set, and taking the union when the intersection is empty.
3. The method for associating power operation records using hierarchical constraint fuzzy positioning according to claim 1, characterized in that, In step S3, the position-weighted score is calculated based on the positional distribution of the matching characters in the candidate device names, specifically including: Record the index position of the matching character in the candidate device name, calculate the average position of the matching character and normalize it to obtain the basic position score; If the average position falls within the first or second third of the text, the base position score is increased by a first preset weight; if the number of consecutive unmatched characters exceeds a preset threshold, the base position score is reduced by a second preset weight. After adjustment, the interval [0,1] is truncated and used as the final position-weighted score.
4. The method for associating power operation records using hierarchical constraint fuzzy positioning according to claim 1, characterized in that, In step S5, the multidimensional fusion model is a product model: Score = Base × Context; in: Base=Jaccard bg a1 ×Jaccard col a2 ×IDF boost a3 , Context=station factor ×time factor ×position factor a4 ×device hit_weight ; In the formula: Score is the overall matching score; Base is the basic matching score; Context is the context correction term; Jaccard bg Character-level similarity factor; Jaccard col IDF is the column name similarity factor. boost To hit the weighting factor for the inverted index; station factor For site relevance factors; time factor For time factor; position factor Location-weighted factor; device hit_weight a1, a2, a3, and a4 are the weighting factors for key equipment; a1, a2, a3, and a4 are the preset weighting indices.
5. The power operation record association method with hierarchical constraint fuzzy positioning according to claim 4, characterized in that, The time factor is determined using a piecewise linear function based on the time difference Δt between the emergency repair record time and the operation log time: In the formula: Δt is the time difference between the emergency repair record and the operation log; min is the unit of time, minutes; v1, v2, and v3 are the base coefficients of the three-segment function, respectively; When the time difference is no greater than 10 minutes, the first value v1 is taken; When the time interval is between 10 minutes and 1 hour, the value decreases linearly from the first value v1 to the second value v2. When the time interval is between 1 hour and 24 hours, the value decreases linearly from the second value v2 to the third value v3. If more than 24 hours have passed, the third value v3 will be used.
6. The method for associating power operation records using hierarchical constraint fuzzy positioning according to claim 1, characterized in that, In step S6, based on the baseline threshold, if a device number is matched in the candidate domain filtering stage, the judgment threshold is set to the baseline threshold multiplied by the first coefficient. If no device number is matched, the determination threshold is set to the baseline threshold multiplied by the second coefficient; the first coefficient is greater than 1, and the second coefficient is less than 1.
7. The method for associating power operation records using hierarchical constraint fuzzy positioning according to claim 1, characterized in that, In step S6, the output association result includes three levels of judgment: When the overall score is greater than or equal to the first correlation threshold, it is determined to be a consistent correlation. When the overall score is between the judgment threshold and the first association threshold, it is judged as an association warning; When the overall score is less than the judgment threshold, it is judged as inconsistent association.
8. The method for associating power operation records using hierarchical constraint fuzzy positioning according to claim 1, characterized in that, The method is also applicable to real-time consistency verification of scheduling instructions and graphic operation records. During verification, the scheduling instruction text is used as the input source of the description text, and the graphic operation record is used as the data source of the operation log to be matched. Steps S1 to S6 are executed to obtain the consistency judgment result.
9. A hierarchical constraint fuzzy positioning power operation record association system, characterized in that, include: The data preprocessing module is used to obtain emergency repair record text and scheduling operation log, generate descriptive text based on emergency repair record text, extract the device number from the descriptive text and build an inverted index, and perform character-level feature segmentation on the device name in the descriptive text and operation log. The candidate field filtering module is used to filter the operation log based on the inverted index and character-level features, and to obtain a candidate record set by taking the intersection of the three filtering results. The multi-dimensional positioning module is used to locate candidate rows within the candidate record set by filtering and locating them through fuzzy matching and position weighting, and to determine the candidate column region by backtracking title information and set similarity matching. The coordinate parsing module is used to perform column name similarity matching within the column candidate area to determine the target column, and then perform fuzzy matching and position weighting within the corresponding data range to determine the target row, thereby achieving two-dimensional coordinate positioning of the target cell; The scoring module is used to calculate a comprehensive score based on a multidimensional fusion model, which integrates character-level similarity factors, column name similarity factors, inverted index hit weighting factors, location weighting factors, site relevance factors, time factors, and key equipment weighting factors. The judgment module is used to dynamically adjust the judgment threshold based on whether the device number is matched, and output the correlation result based on the comprehensive score and the judgment threshold.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 8.