A method, system, medium and device for automatically associating an eventized alarm object with a measurement point

Abstract

The application discloses an eventized alarm object and measuring point automatic association method, system, medium and equipment, belongs to the intelligent monitoring technical field of power systems, and aims to solve the problems of single semantic analysis, lack of topological constraints, fixed weight and non-support of adaptive optimization in the prior art. The method comprises the following steps: acquiring object description and measuring point description in monitoring information; inputting the object description and the measuring point description into a pre-processed semantic analysis model to obtain a comprehensive association confidence of the object and the measuring point; according to a confidence threshold and the comprehensive association confidence, the object and the measuring point are hierarchically associated; collecting result feedback samples after hierarchical association, and continuously optimizing the semantic analysis model through an online learning mechanism. The application can fully capture semantic information from multiple levels, and significantly improve the association accuracy of the object and the measuring point under the premise of ensuring the calculation efficiency.

Description

Technical Field

[0001] This invention relates to a method, system, medium, and equipment for automatically associating event-based alarm objects with measurement points, belonging to the field of intelligent monitoring technology for power systems. Background Technology

[0002] With the construction of a new generation of centralized monitoring systems for substations, event-based alarm functionality has become a core capability of power monitoring.

[0003] Event-based alarms require accurate association between monitored information objects and measurement points to aggregate scattered monitoring signals into meaningful alarm events. However, current technologies primarily rely on manual configuration for object-to-measure point association, which presents the following problems: measurement point naming conventions vary significantly across different manufacturers and projects, making simple string matching insufficient for handling semantic equivalence relationships; multi-level semantic fusion mechanisms are lacking; device topology constraints are not considered; and weight parameters are fixed, preventing dynamic optimization. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method, system, medium and device for automatically associating event-based alarm objects with measurement points, thereby overcoming the shortcomings of existing technologies such as single semantic parsing, lack of topological constraints, fixed weights and lack of support for adaptive optimization.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] In a first aspect, the present invention proposes a method for automatically associating event-based alarm objects with measurement points, comprising:

[0007] Obtain object descriptions and measurement point descriptions from monitoring information;

[0008] The object description and the measurement point description are input into the pre-processed semantic parsing model to obtain the comprehensive association confidence of the object and the measurement point.

[0009] Based on the confidence threshold and the comprehensive association confidence, the objects and measurement points are hierarchically associated.

[0010] Collect feedback samples of the results after hierarchical association, and continuously optimize the semantic parsing model through an online learning mechanism;

[0011] The processing steps of the semantic parsing model include:

[0012] Obtain a pre-built semantic knowledge base specific to the power field, perform multi-granular semantic parsing on the object description and measurement point description of the monitoring information, and obtain word-level feature vectors, key semantic units and sentence semantic embedding vectors of the object description and measurement point description;

[0013] Based on the word-level feature vectors, key semantic units, and sentence semantic embedding vectors of the object description and the test point description, the BM25 similarity and word-shift distance similarity between the object and the test point are dynamically calculated, and the BM25 similarity and word-shift distance similarity between the object and the test point are fused to obtain the pure text semantic similarity between the object and the test point.

[0014] Calculate structural similarity based on the path distance between the pre-acquired object descriptions and measurement point descriptions in the device topology diagram;

[0015] The semantic similarity of the plain text and the structural similarity are fused to obtain a comprehensive similarity.

[0016] Based on the comprehensive similarity, the expression for the comprehensive association confidence is calculated.

[0017] Furthermore, the dedicated semantic knowledge base for the power industry includes an equipment type ontology, an electrical function ontology, a naming convention ontology, a thesaurus mapping table, and a semantic association rule base;

[0018] A hierarchical classification system for substation equipment is determined based on the equipment type body, and the hierarchical classification system includes primary equipment, secondary equipment, and auxiliary equipment.

[0019] The semantic relationships between protection functions, measurement functions, and control functions are determined based on the electrical function body.

[0020] The naming rules mapping for measurement points of different manufacturers and different projects is determined based on the naming specification ontology;

[0021] Establish a synonym mapping relationship for device name, action type, and status description based on the synonym mapping table;

[0022] The semantic association rule base stores matching rules between device types and measurement point types, voltage level consistency rules, and interval assignment rules.

[0023] Furthermore, the multi-granularity semantic parsing includes word-level feature extraction, phrase-level feature extraction, and sentence-level feature extraction;

[0024] The word-level feature extraction adopts the maximum matching word segmentation algorithm based on the dictionary of the power field to identify professional terms and construct word-level feature vectors for object description and measurement point description.

[0025] The phrase-level feature extraction is based on dependency parsing to identify noun phrases and verb phrases, and extracts key semantic units for object description and measurement point description;

[0026] The sentence-level feature extraction employs a lightweight semantic coding model based on Sentence-BERT to extract sentence semantic embedding vectors for object description and measurement point description.

[0027] Furthermore, based on the word-level feature vectors, the word frequency information of the query words in the object description and the word vector set of the object and the measurement point are obtained;

[0028] Based on the key semantic units, the weights of the word-level feature vectors are adjusted;

[0029] Based on the sentence semantic embedding vector, a global semantic background is provided for the word vector set of the object and the test point.

[0030] Furthermore, the dynamic calculation of the BM25 similarity and word-shift distance similarity between the object and the test point, and the fusion of the BM25 similarity and word-shift distance similarity between the object and the test point to obtain the pure text semantic similarity between the object and the test point, includes:

[0031] Based on the improved BM25 algorithm, the BM25 similarity between the object and the measurement point is calculated using the following formula:

[0032] ;

[0033] in, Documentation describing measurement points Query with object description BM25 similarity score; Represents object description query The i-th query term in; Indicates query terms Inverse document frequency, Indicates query terms In the measurement point description document Frequency of occurrence in; This represents the word frequency saturation parameter, used to control the degree of influence of word frequency on the score; This represents the length normalization parameter, used to control the degree to which document length affects the score; Documentation describing measurement points The total number of terms; This represents the average length of all documents in the corpus. This represents the semantic editing distance adjustment coefficient; Indicates query terms With measurement point description document Semantic edit distance similarity of words in the text;

[0034] Based on the word-shift distance algorithm, the word-shift distance similarity between the object and the measurement point is calculated using the following formula:

[0035] ;

[0036] in, Representation Object With measuring points Word-shift similarity; Representation Object Word vector set With measuring points word vector set Word shift distance between;

[0037] The formula for calculating the feedback entropy of the BM25 similarity dimension is as follows:

[0038] ;

[0039] in, This represents the feedback entropy of the BM25 similarity dimension. The lower the feedback entropy, the stronger the discriminative power of this dimension. Indicates the first The probability distribution values ​​of each feedback sample on the BM25 similarity dimension;

[0040] The formula for calculating the feedback entropy of the word-shift distance similarity dimension is:

[0041] ;

[0042] in, The feedback entropy represents the word-shift distance similarity dimension; Indicates the first The probability distribution values ​​of each feedback sample on the word-shift distance similarity dimension;

[0043] Based on the feedback entropy of the BM25 similarity dimension and the feedback entropy of the word shift distance similarity dimension, the weight coefficients of BM25 similarity and word shift distance similarity are calculated using the following formula:

[0044] ;

[0045] ;

[0046] in, This represents the BM25 similarity weighting coefficient. This represents the word-shift distance similarity weight coefficient;

[0047] Based on the weighting coefficients of BM25 similarity and word-shift distance similarity, the BM25 similarity and word-shift distance similarity between the object and the test point are fused to obtain the pure text semantic similarity between the object and the test point; the formula for calculating the pure text semantic similarity is as follows:

[0048] ;

[0049] in, Represents the plain text semantic similarity between object A and measurement point B; This refers to the description document corresponding to measurement point B; This indicates a description query corresponding to object A. This refers to the description document corresponding to measurement point B. Description query corresponding to object A The BM25 similarity score.

[0050] Furthermore, the formula for calculating the structural similarity is:

[0051] ;

[0052] in, Indicates the structural similarity between object A and measurement point B; This represents the path distance between object A and measurement point B in the device topology diagram.

[0053] The formula for calculating the overall similarity is:

[0054] ;

[0055] in, Representation Object With measuring points The overall similarity;

[0056] The formula for calculating the comprehensive association confidence level is as follows:

[0057] ;

[0058] in, This represents the overall confidence level of the association between object A and measurement point B; This represents the normalized feedback entropy, with a value range of [0,1].

[0059] Furthermore, the step of hierarchically associating the object with the measurement point based on the confidence threshold and the comprehensive association confidence includes:

[0060] When the overall association confidence between object A and measurement point B is greater than or equal to the high confidence threshold, the association between object A and measurement point B will be automatically confirmed.

[0061] When the overall association confidence of object A and measurement point B is less than the high confidence threshold but greater than or equal to the low confidence threshold, object A and measurement point B will be submitted for manual review.

[0062] When the overall association confidence between object A and measurement point B is less than the low confidence threshold, object A and measurement point B will be rejected from association.

[0063] The feedback samples from the collected association results are used to continuously optimize the semantic parsing model through an online learning mechanism, including:

[0064] The results of manual review are collected as feedback samples, and the weight parameters of the semantic parsing model are updated using an online gradient descent algorithm.

[0065] Secondly, this invention proposes an automatic association system between event-based alarm objects and measurement points, comprising:

[0066] The information acquisition module is used to acquire object descriptions and measurement point descriptions from monitoring information;

[0067] The data processing module is used to input the object description and the measurement point description into the pre-processed semantic parsing model to obtain the comprehensive association confidence of the object and the measurement point;

[0068] The association decision module is used to classify and associate objects with measurement points according to the confidence threshold and the comprehensive association confidence.

[0069] The model optimization module is used to collect feedback samples of the associated results and continuously optimize the semantic parsing model through an online learning mechanism.

[0070] The processing steps of the semantic parsing model include:

[0071] Obtain a pre-built semantic knowledge base specific to the power field, perform multi-granular semantic parsing on the object description and measurement point description of the monitoring information, and obtain word-level feature vectors, key semantic units and sentence semantic embedding vectors of the object description and measurement point description;

[0072] Based on the word-level feature vectors, key semantic units, and sentence semantic embedding vectors of the object description and the test point description, the BM25 similarity and word-shift distance similarity between the object and the test point are dynamically calculated, and the BM25 similarity and word-shift distance similarity between the object and the test point are fused to obtain the pure text semantic similarity between the object and the test point.

[0073] Calculate structural similarity based on the path distance between the pre-acquired object descriptions and measurement point descriptions in the device topology diagram;

[0074] The semantic similarity of the plain text and the structural similarity are fused to obtain a comprehensive similarity.

[0075] Based on the comprehensive similarity, the expression for the comprehensive association confidence is calculated.

[0076] Thirdly, the present invention proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of an automatic association method between event-based alarm objects and measurement points.

[0077] Fourthly, the present invention provides a computing device, comprising,

[0078] Memory, used to store computer programs;

[0079] A processor is used to execute the computer program to implement the steps of the method for automatically associating event-based alarm objects with measurement points.

[0080] The beneficial effects achieved by this invention are as follows:

[0081] This invention captures semantic information at three levels—word, phrase, and sentence—through multi-granularity semantic parsing; it introduces semantic edit distance through an improved BM25 algorithm, solving the semantic equivalence problem under different naming conventions; it effectively avoids erroneous associations across intervals and voltage levels through device topology constraints; and it achieves continuous self-optimization of the model through adaptive weight optimization based on feedback entropy. Attached Figure Description

[0082] Figure 1 A flowchart of an event-based alarm object and measurement point automatic association method provided by the present invention;

[0083] Figure 2 The multi-granularity semantic parsing flowchart provided by this invention;

[0084] Figure 3 The flowchart for similarity calculation and adaptive weight optimization provided by this invention;

[0085] Figure 4 This is a system architecture diagram for the automatic association of object measurement points provided by the present invention. Detailed Implementation

[0086] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0087] Example 1, such as Figure 1 As shown, this embodiment proposes a method for automatically associating event-driven alarm objects with measurement points, including:

[0088] Obtain object descriptions and measurement point descriptions from monitoring information; the object descriptions are obtained from the alarm configuration library or monitoring information point table of the substation centralized monitoring system, and the measurement point descriptions are obtained from the SCD model file or SCADA database.

[0089] The object description and the test point description are input into the pre-processed semantic parsing model to obtain the comprehensive association confidence of the object and the test point; the semantic parsing model uses a pre-trained language model based on the Transformer structure as the core sentence semantic encoder, and the pre-trained language model includes BERT or Sentence-BERT model.

[0090] Based on the confidence threshold and the comprehensive association confidence, the objects and measurement points are hierarchically associated.

[0091] Collect feedback samples of the results after hierarchical association, and continuously optimize the semantic parsing model through an online learning mechanism;

[0092] The processing steps of the semantic parsing model include:

[0093] Obtain a pre-built semantic knowledge base specific to the power field, perform multi-granular semantic parsing on the object description and measurement point description of the monitoring information, and obtain word-level feature vectors, key semantic units and sentence semantic embedding vectors of the object description and measurement point description;

[0094] Based on the word-level feature vectors, key semantic units, and sentence semantic embedding vectors of the object description and the test point description, the BM25 similarity and word-shift distance similarity between the object and the test point are dynamically calculated, and the BM25 similarity and word-shift distance similarity between the object and the test point are fused to obtain the pure text semantic similarity between the object and the test point.

[0095] Calculate structural similarity based on the path distance between the pre-acquired object descriptions and measurement point descriptions in the device topology diagram;

[0096] The semantic similarity of the plain text and the structural similarity are fused to obtain a comprehensive similarity.

[0097] Based on the comprehensive similarity, the expression for the comprehensive association confidence is calculated.

[0098] The dedicated semantic knowledge base for the power industry includes an equipment type ontology, an electrical function ontology, a naming convention ontology, a thesaurus mapping table, and a semantic association rule base.

[0099] A hierarchical classification system for substation equipment is determined based on the equipment type body, and the hierarchical classification system includes primary equipment, secondary equipment, and auxiliary equipment.

[0100] The semantic relationships between protection functions, measurement functions, and control functions are determined based on the electrical function body.

[0101] The naming rules mapping for measurement points of different manufacturers and different projects is determined based on the naming specification ontology;

[0102] Establish a synonym mapping relationship for device name, action type, and status description based on the synonym mapping table;

[0103] The semantic association rule base stores matching rules between device types and measurement point types, voltage level consistency rules, and interval assignment rules.

[0104] The multi-granularity semantic parsing includes word-level feature extraction, phrase-level feature extraction, and sentence-level feature extraction;

[0105] The word-level feature extraction adopts the maximum matching word segmentation algorithm based on the dictionary of the power field to identify professional terms and construct word-level feature vectors for object description and measurement point description.

[0106] The phrase-level feature extraction is based on dependency parsing to identify noun phrases and verb phrases, and extracts key semantic units for object description and measurement point description;

[0107] The sentence-level feature extraction employs a lightweight semantic coding model based on Sentence-BERT to extract sentence semantic embedding vectors for object description and measurement point description.

[0108] Based on the word-level feature vectors, obtain the word frequency information of the query words in the object description and the word vector set of the object and the measurement point;

[0109] Based on the key semantic units, the weights of the word-level feature vectors are adjusted;

[0110] Based on the sentence semantic embedding vector, a global semantic background is provided for the word vector set of the object and the test point.

[0111] The dynamic calculation of BM25 similarity and word-shift distance similarity between the object and the test point, and the fusion of BM25 similarity and word-shift distance similarity between the object and the test point to obtain the pure text semantic similarity between the object and the test point, includes:

[0112] Based on the improved BM25 algorithm, the BM25 similarity between the object and the measurement point is calculated using the following formula:

[0113] ;

[0114] in, Documentation describing measurement points Query with object description BM25 similarity score; Represents object description query The i-th query term in; Indicates query terms Inverse document frequency, Indicates query terms In the measurement point description document Frequency of occurrence in; This represents the word frequency saturation parameter, used to control the degree of influence of word frequency on the score; This represents the length normalization parameter, used to control the degree to which document length affects the score; Documentation describing measurement points The total number of terms; This represents the average length of all documents in the corpus. This represents the semantic editing distance adjustment coefficient; Indicates query terms With measurement point description document Semantic edit distance similarity of words in the text;

[0115] Based on the word-shift distance algorithm, the word-shift distance similarity between the object and the measurement point is calculated using the following formula:

[0116] ;

[0117] in, Representation Object With measuring points Word-shift similarity; Representation Object word vector set With measuring points Word vector set Word shift distance between;

[0118] The formula for calculating the feedback entropy of the BM25 similarity dimension is as follows:

[0119] ;

[0120] in, This represents the feedback entropy of the BM25 similarity dimension. The lower the feedback entropy, the stronger the discriminative power of this dimension. Indicates the first The probability distribution values ​​of each feedback sample on the BM25 similarity dimension;

[0121] The formula for calculating the feedback entropy of the word-shift distance similarity dimension is:

[0122] ;

[0123] in, The feedback entropy represents the word-shift distance similarity dimension; Indicates the first The probability distribution values ​​of each feedback sample on the word-shift distance similarity dimension;

[0124] Based on the feedback entropy of the BM25 similarity dimension and the feedback entropy of the word shift distance similarity dimension, the weight coefficients of BM25 similarity and word shift distance similarity are calculated using the following formula:

[0125] ;

[0126] ;

[0127] in, This represents the BM25 similarity weighting coefficient. This represents the word-shift distance similarity weight coefficient;

[0128] Based on the weighting coefficients of BM25 similarity and word-shift distance similarity, the BM25 similarity and word-shift distance similarity between the object and the test point are fused to obtain the pure text semantic similarity between the object and the test point; the formula for calculating the pure text semantic similarity is as follows:

[0129] ;

[0130] in, Represents the plain text semantic similarity between object A and measurement point B; This refers to the description document corresponding to measurement point B; This indicates a description query corresponding to object A.

[0131] The formula for calculating the structural similarity is:

[0132] ;

[0133] in, Indicates the structural similarity between object A and measurement point B; This represents the path distance between object A and measurement point B in the device topology diagram.

[0134] The formula for calculating the overall similarity is:

[0135] ;

[0136] in, Representation Object With measuring points The overall similarity;

[0137] The formula for calculating the comprehensive association confidence level is as follows:

[0138] ;

[0139] in, This represents the overall confidence level of the association between object A and measurement point B; This represents the normalized feedback entropy, with a value range of [0,1].

[0140] The step of hierarchically associating objects with measurement points based on confidence thresholds and the comprehensive association confidence includes:

[0141] When the overall association confidence between object A and measurement point B is greater than or equal to the high confidence threshold, the association between object A and measurement point B will be automatically confirmed.

[0142] When the overall association confidence of object A and measurement point B is less than the high confidence threshold but greater than or equal to the low confidence threshold, object A and measurement point B will be submitted for manual review.

[0143] When the overall association confidence between object A and measurement point B is less than the low confidence threshold, object A and measurement point B will be rejected from association.

[0144] The feedback samples from the collected association results are used to continuously optimize the semantic parsing model through an online learning mechanism, including:

[0145] The results of manual review are collected as feedback samples, and the weight parameters of the semantic parsing model are updated using an online gradient descent algorithm.

[0146] Example 2, based on the same inventive concept as Example 1, proposes a method for automatically associating event-driven alarm objects with measurement points, including the following steps:

[0147] Step S1: Semantic Knowledge Base Construction. A dedicated semantic knowledge base for the power industry is constructed, including an equipment type ontology, an electrical function ontology, a naming convention ontology, a thesaurus mapping table, and a semantic association rule base.

[0148] The object description is obtained from the alarm configuration library or monitoring information point table of the substation centralized monitoring system. Its text content typically includes equipment name, protection action type, alarm status, etc., such as 220kV I-section bus differential protection action. The measurement point description is obtained from the substation's SCD model file, remote control point table, or SCADA database. Its text content typically includes information such as substation, bay, equipment, and attributes. For example, the measurement point description corresponding to the above objects may be: XX substation / 220kV / M1 bus / protection set A / differential action signal.

[0149] Step S2, multi-granular semantic parsing. For example... Figure 2 As shown, three levels of semantic parsing are performed on the description of monitoring information objects and measurement points, respectively extracting word-level feature vectors, key semantic units, and sentence semantic embedding vectors;

[0150] Based on the word-level feature vectors, obtain the word frequency information of the query words in the object description and the word vector set of the object and the measurement point;

[0151] Based on the key semantic units, the weights of the word-level feature vectors are adjusted;

[0152] Based on the sentence semantic embedding vector, a global semantic background is provided for the word vector set of the object and the test point.

[0153] Step S3: Calculate the semantic similarity of the pure text.

[0154] like Figure 3 As shown, this invention introduces a semantic edit distance term based on the traditional BM25 algorithm. This allows a score to be obtained even when a query term is semantically similar to but not exactly a match for a term in a document, thus providing the BM25 similarity between the object and the measurement point. This effectively solves the problem of literal mismatch caused by differences in naming conventions among different manufacturers, such as situations where tripping and action, closing and shutting down are semantically equivalent but literally different.

[0155] The formula for calculating the BM25 similarity between the object and the measuring point is:

[0156] ;

[0157] in, Documentation describing measurement points Query with object description BM25 similarity score; Represents object description query The i-th query term in; Indicates query terms Inverse document frequency, Indicates query terms In the measurement point description document Frequency of occurrence in; This represents the word frequency saturation parameter, used to control the degree to which word frequency affects the score; This represents the length normalization parameter, used to control the degree to which document length affects the score; Documentation describing measurement points The total number of terms; This represents the average length of all documents in the corpus. This represents the semantic editing distance adjustment coefficient; Indicates query terms With measurement point description document Semantic edit distance similarity of each word in the text.

[0158] Simultaneously, word-shift distance similarity is calculated based on the Word Shift Distance (WMD) algorithm. The WMD algorithm treats object descriptions and measurement point descriptions as sets of word vectors respectively, and calculates the minimum cumulative distance between the two sets by solving the optimal transportation problem. This can capture the deeper meaning of the text from a semantic level, supplementing the BM25 algorithm's reliance solely on word frequency statistics.

[0159] The formula for calculating the word-shift distance similarity between an object and a measurement point is:

[0160] ;

[0161] in, Representation Object With measuring points Word-shift similarity; Representation Object Word vector set With measuring points Word vector set Word shift distance between;

[0162] The BM25 similarity and word-shift distance similarity are linearly weighted and fused, with an initial weight of 0.5 for each, to obtain the pure text semantic similarity.

[0163] The formula for the semantic similarity of the plain text is:

[0164] ;

[0165] in, Represents the plain text semantic similarity between object A and measurement point B; This represents the BM25 similarity weighting coefficient. This represents the word-shift distance similarity weight coefficient; This refers to the description document corresponding to measurement point B; This indicates a description query corresponding to object A.

[0166] Step S4: Introduce equipment topology constraints and construct structural similarity. Based on the substation equipment topology map, calculate the path distance between the corresponding equipment nodes of object A and measuring point B. If the two are not in the same interval or voltage level, the path distance is large and the structural similarity is low, thus playing a constraining role;

[0167] The formula for calculating the structural similarity is:

[0168] ;

[0169] in, Indicates the structural similarity between object A and measurement point B; This represents the path distance between object A and measurement point B in the device topology diagram.

[0170] Step S5: Adaptive weight optimization. In actual operation, the distinguishing ability of the BM25 similarity dimension and the word-shift distance similarity dimension may vary in different scenarios. For example, in projects with relatively uniform naming conventions, the BM25 similarity dimension has a stronger distinguishing ability; while in projects with mixed equipment from multiple manufacturers, the semantic understanding advantage of the word-shift distance similarity dimension is more prominent.

[0171] To address this, this invention employs an adaptive weight optimization algorithm based on feedback entropy to dynamically adjust the weight coefficients α and β of the BM25 similarity dimension and the word-shift distance similarity dimension in pure text semantic similarity. A lower feedback entropy indicates a stronger discriminative ability for that dimension, and therefore it should be assigned a higher weight.

[0172] Specifically, the association results reviewed manually are collected as feedback samples. The probability distributions of BM25 similarity dimension values ​​and word shift distance similarity dimension values ​​on correct and incorrect association samples are statistically analyzed, and their respective feedback entropies are calculated. and Then update the weighting coefficients according to the inverse relationship:

[0173] ;

[0174] ;

[0175] The adjusted semantic similarity of the plain text is calculated using the updated weights, and then multiplied by the structural similarity to obtain the comprehensive similarity. The formula for calculating the comprehensive similarity is as follows:

[0176] ;

[0177] in, Representation Object With measuring points The overall similarity.

[0178] Step S6: Confidence assessment and hierarchical decision-making.

[0179] Taking into account both the overall similarity and the overall feedback uncertainty, the overall association confidence is calculated as follows:

[0180] ;

[0181] in, This represents the overall confidence level of the association between object A and measurement point B; The normalized feedback entropy has a value range of [0,1].

[0182] Set a high confidence threshold and low confidence threshold To make decisions in a tiered manner:

[0183] when ≥ The association is automatically confirmed without manual intervention.

[0184] when ≤ < When necessary, submit for manual review, and have it confirmed or corrected by the operations and maintenance personnel;

[0185] when < At that time, the association is rejected.

[0186] Step S7: Online Learning and Optimization. Collect manual review results as feedback samples, update the semantic encoding model parameters using an online gradient descent algorithm, and update the weight parameters using an exponential moving average method. As the system runs longer and feedback samples accumulate, the model's adaptability to specific engineering environments gradually improves, and the association accuracy continues to increase.

[0187] In summary, as in this embodiment, Figure 4 As shown, the event-based alarm object and measurement point automatic association method proposed in this embodiment is divided into a data input layer, a core processing layer and a decision output layer.

[0188] At the data input layer, object descriptions are obtained from the alarm configuration library or monitoring information point table, measurement point descriptions are obtained from the SCD model file or SCADA database, and equipment topology data are obtained from the substation topology file.

[0189] In the core processing layer, features are extracted from the input data based on a semantic knowledge base and a multi-granularity parsing engine; the semantic similarity of the plain text is calculated based on the improved BM25 algorithm and WMD algorithm; the structural similarity is calculated based on the device topology graph; and the internal weights of the plain text are dynamically adjusted through an adaptive weight optimization algorithm and fused with the structural similarity to obtain a comprehensive similarity.

[0190] At the decision output layer, the comprehensive correlation confidence level is calculated and hierarchical decision-making is carried out, outputting the correlation results between the object and the measurement point; at the same time, the manual review results are fed back to the core processing layer, forming a closed-loop optimization mechanism for online learning.

[0191] Example 3, based on Example 1, proposes an automatic association system between event-based alarm objects and measurement points, including:

[0192] The information acquisition module is used to acquire object descriptions and measurement point descriptions from monitoring information;

[0193] The data processing module is used to input the object description and the measurement point description into the pre-processed semantic parsing model to obtain the comprehensive association confidence of the object and the measurement point;

[0194] The association decision module is used to classify and associate objects with measurement points according to the confidence threshold and the comprehensive association confidence.

[0195] The model optimization module is used to collect feedback samples of the associated results and continuously optimize the semantic parsing model through an online learning mechanism.

[0196] The processing steps of the semantic parsing model include:

[0197] Obtain a pre-built semantic knowledge base specific to the power field, perform multi-granular semantic parsing on the object description and measurement point description of the monitoring information, and obtain word-level feature vectors, key semantic units and sentence semantic embedding vectors of the object description and measurement point description;

[0198] Based on the word-level feature vectors, key semantic units, and sentence semantic embedding vectors of the object description and the test point description, the BM25 similarity and word-shift distance similarity between the object and the test point are dynamically calculated, and the BM25 similarity and word-shift distance similarity between the object and the test point are fused to obtain the pure text semantic similarity between the object and the test point.

[0199] Calculate structural similarity based on the path distance between the pre-acquired object descriptions and measurement point descriptions in the device topology diagram;

[0200] The semantic similarity of the plain text and the structural similarity are fused to obtain a comprehensive similarity.

[0201] Based on the comprehensive similarity, the expression for the comprehensive association confidence is calculated.

[0202] Example 4: Based on Example 1, this example proposes a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the method for automatically associating event-based alarm objects with measurement points in Example 1.

[0203] Example 5, based on Example 1, proposes a computing device, including,

[0204] Memory, used to store computer programs;

[0205] A processor is used to execute the computer program to implement the steps of the method for automatically associating event-based alarm objects with measurement points in Embodiment 2.

[0206] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0207] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0208] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0209] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0210] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for automatically associating event-driven alarm objects with measurement points, characterized in that, include: Obtain object descriptions and measurement point descriptions from monitoring information; The object description and the measurement point description are input into the pre-processed semantic parsing model to obtain the comprehensive association confidence of the object and the measurement point. Based on the confidence threshold and the comprehensive association confidence, the objects and measurement points are hierarchically associated. Collect feedback samples of the hierarchical association results, and optimize the semantic parsing model through an online learning mechanism; The processing steps of the semantic parsing model include: Obtain a pre-built semantic knowledge base specific to the power field, perform multi-granular semantic parsing on the object description and measurement point description of the monitoring information, and obtain word-level feature vectors, key semantic units and sentence semantic embedding vectors of the object description and measurement point description; Based on the word-level feature vectors, key semantic units, and sentence semantic embedding vectors of the object description and the test point description, the BM25 similarity and word-shift distance similarity between the object and the test point are dynamically calculated, and the BM25 similarity and word-shift distance similarity between the object and the test point are fused to obtain the pure text semantic similarity between the object and the test point. Calculate structural similarity based on the path distance between the pre-acquired object descriptions and measurement point descriptions in the device topology diagram; The semantic similarity of the plain text and the structural similarity are fused to obtain a comprehensive similarity. Based on the comprehensive similarity, the expression for the comprehensive association confidence is calculated.

2. The method for automatically associating event-based alarm objects with measurement points according to claim 1, characterized in that, The dedicated semantic knowledge base for the power industry includes an equipment type ontology, an electrical function ontology, a naming convention ontology, a thesaurus mapping table, and a semantic association rule base. A hierarchical classification system for substation equipment is determined based on the equipment type body, and the hierarchical classification system includes primary equipment, secondary equipment, and auxiliary equipment. The semantic relationships between protection functions, measurement functions, and control functions are determined based on the electrical function body. The naming rules mapping for measurement points of different manufacturers and different projects is determined based on the naming specification ontology; Establish a synonym mapping relationship for device name, action type, and status description based on the synonym mapping table; The semantic association rule base stores matching rules between device types and measurement point types, voltage level consistency rules, and interval assignment rules.

3. The method for automatically associating event-based alarm objects with measurement points according to claim 1, characterized in that, The multi-granularity semantic parsing includes word-level feature extraction, phrase-level feature extraction, and sentence-level feature extraction; The word-level feature extraction adopts the maximum matching word segmentation algorithm based on the dictionary of the power field to identify professional terms and construct word-level feature vectors for object description and measurement point description. The phrase-level feature extraction is based on dependency parsing to identify noun phrases and verb phrases, and extracts key semantic units for object description and measurement point description; The sentence-level feature extraction employs a lightweight semantic coding model based on Sentence-BERT to extract sentence semantic embedding vectors for object description and measurement point description.

4. The method for automatically associating event-based alarm objects with measurement points according to claim 3, characterized in that, Based on the word-level feature vectors, obtain the word frequency information of the query words in the object description and the word vector set of the object and the measurement point; Based on the key semantic units, the weights of the word-level feature vectors are adjusted; Based on the sentence semantic embedding vector, a global semantic background is provided for the word vector set of the object and the test point.

5. The method for automatically associating event-based alarm objects with measurement points according to claim 1, characterized in that, The dynamic calculation of BM25 similarity and word-shift distance similarity between the object and the test point, and the fusion of BM25 similarity and word-shift distance similarity between the object and the test point to obtain the pure text semantic similarity between the object and the test point, includes: Based on the improved BM25 algorithm, the BM25 similarity between the object and the measurement point is calculated using the following formula: ； in, Documentation describing measurement points Query with object description BM25 similarity score; Represents object description query The i-th query term in; Indicates query terms Inverse document frequency, Indicates query terms In the measurement point description document Frequency of occurrence in; Indicates the word frequency saturation parameter; This represents the length normalization parameter; Documentation describing measurement points The total number of terms; This represents the average length of all documents in the corpus. This represents the semantic editing distance adjustment coefficient; Indicates query terms With measurement point description document Semantic edit distance similarity of words in the text; Based on the word-shift distance algorithm, the word-shift distance similarity between the object and the measurement point is calculated using the following formula: ； in, Representation Object With measuring points Word-shift similarity; Representation Object Word vector set With measuring points Word vector set Word shift distance between; The formula for calculating the feedback entropy of the BM25 similarity dimension is as follows: ； in, The feedback entropy represents the BM25 similarity dimension; Indicates the first The probability distribution values ​​of each feedback sample on the BM25 similarity dimension; The formula for calculating the feedback entropy of the word-shift distance similarity dimension is: ； in, The feedback entropy represents the word-shift distance similarity dimension; Indicates the first The probability distribution values ​​of each feedback sample on the word-shift distance similarity dimension; Based on the feedback entropy of the BM25 similarity dimension and the feedback entropy of the word shift distance similarity dimension, the weight coefficients of BM25 similarity and word shift distance similarity are calculated using the following formula: ； ； in, This represents the BM25 similarity weighting coefficient. This represents the word-shift distance similarity weight coefficient; Based on the weighting coefficients of BM25 similarity and word-shift distance similarity, the BM25 similarity and word-shift distance similarity between the object and the test point are fused to obtain the pure text semantic similarity between the object and the test point; the formula for calculating the pure text semantic similarity is as follows: ； in, Represents the plain text semantic similarity between object A and measurement point B; This refers to the description document corresponding to measurement point B; This indicates a description query corresponding to object A. This refers to the description document corresponding to measurement point B. Description query corresponding to object A The BM25 similarity score.

6. The method for automatically associating event-based alarm objects with measurement points according to claim 5, characterized in that, The formula for calculating the structural similarity is: ； in, Indicates the structural similarity between object A and measurement point B; This represents the path distance between object A and measuring point B in the device topology diagram; The formula for calculating the overall similarity is: ； in, Representation Object With measuring points The overall similarity; The formula for calculating the comprehensive association confidence level is as follows: ； in, This represents the overall confidence level of the association between object A and measurement point B; This represents the normalized feedback entropy, with a value range of [0,1].

7. The method for automatically associating event-based alarm objects with measurement points according to claim 6, characterized in that, The step of hierarchically associating objects with measurement points based on confidence thresholds and the comprehensive association confidence includes: When the overall association confidence between object A and measurement point B is greater than or equal to the high confidence threshold, the association between object A and measurement point B will be automatically confirmed. When the overall association confidence of object A and measurement point B is less than the high confidence threshold but greater than or equal to the low confidence threshold, object A and measurement point B will be submitted for manual review. When the overall association confidence between object A and measurement point B is less than the low confidence threshold, object A and measurement point B will be rejected from association. The process of collecting hierarchical association results and feedback samples, and optimizing the semantic parsing model through an online learning mechanism, includes: The results of manual review are collected as feedback samples, and the weight parameters of the semantic parsing model are updated using an online gradient descent algorithm.

8. An event-based alarm object and measurement point automatic association system, characterized in that, include: The information acquisition module is used to acquire object descriptions and measurement point descriptions from monitoring information; The data processing module is used to input the object description and the measurement point description into the pre-processed semantic parsing model to obtain the comprehensive association confidence of the object and the measurement point; The association decision module is used to classify and associate objects with measurement points according to the confidence threshold and the comprehensive association confidence. The model optimization module is used to collect feedback samples of the associated results and continuously optimize the semantic parsing model through an online learning mechanism. The processing steps of the semantic parsing model include: Obtain a pre-built semantic knowledge base specific to the power field, perform multi-granular semantic parsing on the object description and measurement point description of the monitoring information, and obtain word-level feature vectors, key semantic units and sentence semantic embedding vectors of the object description and measurement point description; Based on the word-level feature vectors, key semantic units, and sentence semantic embedding vectors of the object description and the test point description, the BM25 similarity and word-shift distance similarity between the object and the test point are dynamically calculated, and the BM25 similarity and word-shift distance similarity between the object and the test point are fused to obtain the pure text semantic similarity between the object and the test point. Calculate structural similarity based on the path distance between the pre-acquired object descriptions and measurement point descriptions in the device topology diagram; The semantic similarity of the plain text and the structural similarity are fused to obtain a comprehensive similarity. Based on the comprehensive similarity, the expression for the comprehensive association confidence is calculated.

9. 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 method for automatically associating event-based alarm objects with measurement points as described in any one of claims 1 to 7.

10. A computing device, characterized in that, include, Memory, used to store computer programs; A processor is configured to execute the computer program to implement the steps of the method for automatically associating event-based alarm objects with measurement points as described in any one of claims 1 to 7.

Images