A protection signal semi-supervised classification method, system, device and storage medium based on multi-dimensional cascade matching

Abstract

The application discloses a kind of based on multi-dimension cascade matching protection signal semi-supervised classification method, system, equipment and storage medium, the method is based on topological interval classification result constructs multi-dimensional feature space;With feature vector similarity as the leading, longest common subsequence similarity as auxiliary multi-dimension cascade matching method is used to divide candidate standard type into multistage confidence cluster, after electrical logic post-check and back to recalculation, select optimal matching as initial classification result;Record artificial check correction as clustering center and initialize similarity level threshold;Based on multi-dimension cascade matching, the similarity level between the signal to be classified and clustering center is calculated, and when the current threshold is reached, it is classified into the corresponding standard type;With clustering success accumulation gradually lower threshold to improve efficiency, immediately back up threshold to correct error when excessive clustering, by the combination of multi-dimension cascade matching and adaptive threshold semi-supervised clustering, the efficient and accurate classification of protection signal is realized.

Description

Technical Field

[0001] This invention relates to the field of power automation, and in particular to a semi-supervised classification method, system, device and storage medium for protection signals based on multi-dimensional cascade matching. Background Technology

[0002] In existing power monitoring systems (hereinafter referred to as "EMS systems"), standardizing protection signals and automatically identifying the fault types corresponding to common protection signal action combinations can significantly reduce the workload of monitoring personnel while improving judgment accuracy and subsequent processing efficiency. Accurately mapping these signals to a unified standard template is the foundation for achieving the above objectives.

[0003] Patent CN202510964741.X discloses an automatic mapping method and system for power bay classification based on power topology and graph algorithms. This method achieves standard classification of power plant bays through undirected graph construction and a breadth-first search algorithm. However, bay classification is only the first step in standardizing protection signals; accurately mapping protection signals to their corresponding standard protection types remains a problem to be solved.

[0004] Engineering practice shows that relying entirely on manual labor to complete this mapping task is extremely difficult. Taking the maintenance of a 110kV substation as an example, not only does it require maintenance personnel to have a thorough understanding of various common protection signals, but the entire process is also extremely time-consuming and has a high probability of misclassification. Most existing automated classification methods adopt a one-time matching strategy, which is difficult to adapt to the differences in specific signals on site and lacks the ability to adaptively optimize the classification results; at the same time, these methods cannot accumulate experience during manual verification and do not have the ability to automatically adjust the classification strategy. Summary of the Invention

[0005] Purpose of the invention: The present invention aims to provide a semi-supervised classification method for protection signals that achieves efficient and accurate classification, has memory function, and can be extended across fields within a region. Another purpose of the present invention is to provide a semi-supervised classification system, device, and storage medium for protection signals based on multi-dimensional cascaded matching.

[0006] Technical solution: The semi-supervised classification method for protection signals based on multi-dimensional cascade matching in the field of power automation described in this invention includes:

[0007] (1) Based on the bay classification results based on power topology and graph algorithm, a multi-dimensional feature space is constructed that includes information on substation type, bay type, wiring method within the bay, corresponding primary equipment type and protection type;

[0008] (2) A multi-dimensional cascaded matching method is adopted to calculate the matching degree between the protection signal to be classified and each candidate standard type. The candidate set is divided into multiple confidence level clusters. After electrical logic verification, the best match is selected as the initial classification result. The cluster with higher feature matching degree adopts a more relaxed matching threshold. Within each cluster, the candidates that meet the threshold conditions are sorted by edit distance and the best is selected. If all candidates cannot meet the threshold requirements of the cluster, the classification is judged to be unsuccessful and the manual verification is initiated.

[0009] (3) Record the protection signal information that has been manually checked and corrected and its correct classification results as cluster centers, and initialize the similarity level threshold;

[0010] (4) Using a multi-dimensional cascaded matching method, the similarity level between the signal to be classified and protected and the cluster center is calculated. When the similarity level reaches or exceeds the current threshold level, it is classified into the standard protection type corresponding to the cluster center.

[0011] (5) Gradually lower the similarity level threshold as the number of successful clusterings accumulates to improve clustering efficiency; when over-clustering leads to classification errors and is manually corrected, immediately raise the similarity level threshold.

[0012] Furthermore, in step (1), the power topology relationship and the interval classification results of the graph algorithm include the interval type dimension, the wiring method dimension within the interval, the primary equipment type dimension, and the protection type dimension;

[0013] Spacing type dimension: including standard spacing types such as busbar spacing, winding side spacing, outgoing line spacing, bridge spacing, main transformer spacing, as well as common spacing;

[0014] Wiring methods within a bay: including single busbar wiring, double busbar wiring, 3 / 2 wiring, bridge wiring, etc.

[0015] Primary equipment type dimension: including main transformers, lines, busbars, capacitors, reactors, station service transformers, etc.;

[0016] Protection type dimension: includes protection signal types such as main protection, backup protection, abnormal alarm, and switch position change.

[0017] Among them, an undirected graph is constructed based on the power system's plant information, bay information, equipment information, and equipment association information. It is then divided into several undirected subgraphs based on connected components and voltage levels to determine the power topology.

[0018] A breadth-first search algorithm is used to perform the first standard bay classification based on the relationship between the bus and the main transformer; the results of the first classification and the attributes of the circuit components are combined to perform the second standard bay classification, thus completing the automatic classification of bay types and obtaining the bay classification results.

[0019] Furthermore, in step (2), the multi-dimensional cascaded matching method uses feature vector similarity as the main factor and the longest common subsequence similarity as the auxiliary factor for clustering; wherein, the calculation method of the longest common subsequence similarity is: after removing the English characters in the protection signal name, calculate the ratio of the length of the longest common subsequence of the protection signal name and the candidate standard type name to the length of the original string.

[0020] The clustering rules for multi-dimensional cascaded matching are as follows: candidates with completely matched feature vectors are assigned to cluster c1; those with feature vectors that are contained in each other and have a high similarity of their longest common subsequence are assigned to cluster c2; those with feature vectors that are intertwined and have a high similarity of their longest common subsequence, or those that are contained in each other and have a low similarity of their longest common subsequence, are assigned to cluster c3; cluster c1 uses the most lenient matching threshold, cluster c2 is the next most lenient, and cluster c3 is the most lenient.

[0021] After matching is completed, electrical logic post-verification is required. Verification items include voltage level consistency, equipment type matching degree, wiring method compatibility, protection stage rationality, and interval assignment correctness. If all pass, the matching is confirmed to be successful. If any fails, the current category is removed from the candidate set and cluster matching is re-executed until the matching is successful or the candidate set is exhausted.

[0022] Furthermore, in step (4), the multi-dimensional cascaded matching includes three dimensions: the first dimension is the text similarity based on the longest common subsequence, the second dimension is the feature vector similarity, and the third dimension is the matching degree of the plant and station and the interval attribute; when the text similarity of the first dimension is greater than 95%, the similarity level 6 is output; when it is in the range of 50% to 95%, the second and third dimensions are activated to participate in the comprehensive calculation and the similarity level 3 to 5 is output; when it is less than 50%, the weight of the first dimension is reduced and the second and third dimensions are used as the main output for similarity level 1 to 3; the output similarity level is compared with the current threshold level, and when it reaches or exceeds the threshold level, it is classified into the standard protection type corresponding to the cluster center.

[0023] Furthermore, in step (3), the format of the manually checked and corrected record includes the protection signal name, power topology information, automatic classification result, manual correction result and initial similarity threshold value; each record serves as an independent cluster center.

[0024] Furthermore, in step (5), when the cumulative number of successful corrections of a cluster center reaches a set threshold, the similarity level threshold of the cluster center is automatically lowered; when a protection signal that was originally correctly classified by automatic means is incorrectly clustered and is rejected by human qualification, the similarity level threshold of the cluster center is immediately raised.

[0025] Furthermore, all clustering correction records, similarity threshold change trajectories, and clustering results are saved as structured log files, allowing other sites in the same area to reuse the same clustering rules by importing these log files.

[0026] The semi-supervised classification system for protection signals based on multi-dimensional cascaded matching described in this invention includes:

[0027] The multidimensional feature space construction module is used to construct a multidimensional feature space containing interval type, wiring method, equipment type and protection type based on the topology interval classification results;

[0028] The automatic classification and matching module is used to map protection signals to a multi-dimensional feature space and calculate similarity to obtain initial classification results.

[0029] A semi-supervised clustering optimization module is used to record manually verified and corrected cluster centers and manage the adaptive adjustment of similarity thresholds;

[0030] The memory and promotion module is used to save cluster correction records as structured logs and supports other on-site import and reuse.

[0031] The computer device of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above method.

[0032] The computer-readable storage medium of the present invention stores a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0033] Beneficial Effects: Compared with existing technologies, the significant advantages of this invention are: 1. This invention employs a multi-dimensional cascaded matching method for automatic classification of protection signals. It uses feature vector similarity as the primary factor and longest common subsequence similarity as an auxiliary factor to divide the candidate set into multi-level confidence clusters. Combined with electrical logic post-verification and a recalculation mechanism, it offers higher fault tolerance and field adaptability compared to single-dimensional text matching schemes. 2. This invention uses a semi-supervised clustering mechanism. Through manual verification, it automatically accumulates and corrects experience. The similarity level threshold can be adaptively adjusted according to the clustering effect, solving the problem of traditional methods lacking field adaptability. 3. The invention's bidirectional adjustment mechanism of lowering the threshold to improve efficiency and raising the threshold to correct errors effectively balances clustering coverage and accuracy, avoiding misclassification caused by over-clustering. 4. The clustering memory of this invention can be exported as a structured log, supporting rule reuse between different sites within the same region, significantly reducing the engineering maintenance costs of multi-site deployments. Attached Figure Description

[0034] Figure 1 This is a flowchart of the method of the present invention;

[0035] Figure 2 A flowchart for constructing a multidimensional feature space;

[0036] Figure 3 Flowchart for automatic classification of multi-dimensional cascading matching;

[0037] Figure 4 Flowchart for semi-supervised clustering optimization;

[0038] Figure 5 This is a schematic diagram illustrating the adaptive adjustment of the similarity level threshold. Detailed Implementation

[0039] The semi-supervised classification method for protection signals based on multi-dimensional cascaded matching in the field of power automation, as described in this invention, includes:

[0040] (1) Based on the interval classification results based on power topology and graph algorithm, a multi-dimensional feature space is constructed that includes information on substation type, interval type, wiring method within the interval, corresponding primary equipment type and protection type.

[0041] An undirected graph is constructed based on the power system's plant information, bay information, equipment information, and equipment relationship information. It is then divided into several undirected subgraphs based on connected components and voltage levels to determine the power topology. A breadth-first search algorithm is used to perform the first standard bay classification based on the relationship between the bus and the main transformer. A second standard bay classification is performed by combining the results of the first classification with the attributes of the circuit components to complete the automatic classification of bay types and obtain the bay classification results.

[0042] The bay classification method is described in patent 202510964741X. It constructs an undirected graph based on the power system's plant information, bay information, equipment information, and equipment relationship information, and divides it into several undirected subgraphs based on connected components and voltage levels. Then, a breadth-first search algorithm is used to perform the first standard bay classification based on the relationship between the bus and the main transformer, resulting in types such as bus bay, winding side bay, outgoing line bay, bridge bay, and main transformer bay. Finally, a second standard bay classification is performed by combining the results of the first classification with the attributes of the circuit elements to refine the bay types. A feature space is constructed, encompassing the following dimensions: Bay type dimension: bus bay, winding-side bay, outgoing line bay, bridge bay, main transformer bay, bus tie bay, segmented bay, etc.; Wiring method dimension: single busbar, double busbar, 3 / 2 connection, bridge connection (internal / external bridge), unit connection, line-transformer group connection, etc.; Primary equipment type dimension: main transformer, line, busbar, capacitor, reactor, station service transformer, grounding transformer, arc suppression coil, load, etc.; Protection type dimension: main protection action, backup protection action, reclosing action, abnormal alarm, switch change, device alarm, etc. Each standard protection type has a unique feature vector expression in this feature space. Simultaneously, by combining the substation's own voltage level, substation type, and overall wiring layout, a complete protection signal dimension space is obtained.

[0043] (2) Using a multi-dimensional cascaded matching method, the matching degree between the protection signal to be classified and each candidate standard type is calculated. The candidate set is divided into multiple confidence level clusters. After electrical logic verification, the best match is selected as the initial classification result. The cluster with higher feature matching degree adopts a more relaxed matching threshold. Within each cluster, the candidates that meet the threshold conditions are sorted by edit distance and the best is selected. If all candidates cannot meet the threshold requirements of the cluster, the classification is judged to fail and the manual verification is initiated.

[0044] For protection signals to be classified, data cleaning, electrical attribute association, and regular expression matching are performed sequentially; for signals that do not match the regular expression, a multi-dimensional cascaded matching method is used for automatic classification.

[0045] Multi-dimensional cascaded matching, primarily based on feature vector similarity and secondarily on longest common subsequence (LCS) similarity, divides candidate criteria types into different confidence level clusters. The specific clustering rules are as follows:

[0046] c1 cluster: The feature vectors perfectly match the feature vectors of the candidate standard type;

[0047] c2 cluster: The feature vectors and candidate standard types are related by inclusion, and the longest common subsequences have high similarity;

[0048] c3 cluster: The feature vectors and candidate standard types are intertwined and have a high similarity of the longest common subsequence, or they are inclusive and have a low similarity of the longest common subsequence.

[0049] Clusters with higher feature matching scores use a more lenient edit distance matching threshold. During matching, clusters are processed sequentially in the priority order of c1→c2→c3. Within each cluster, candidates that meet the threshold conditions are sorted by edit distance, and the candidate with the smallest edit distance is selected as the optimal candidate for the current cluster.

[0050] After matching is completed, the matching results are subjected to electrical logic post-verification. Verification items include: voltage level consistency; equipment type matching degree; wiring method compatibility; protection phase rationality; bay assignment correctness; and plant category compatibility.

[0051] If all validation items pass, the match is confirmed as successful. If any validation item fails, the current failed category is removed from the candidate set, and the clustering step is returned to re-execute the matching process until a match is successful or the candidate set is exhausted.

[0052] If all candidate classifications fail to meet the threshold requirements of each cluster, or fail to pass the electrical logic verification after being returned for recalculation, the protection signal classification is deemed to have failed, and the process proceeds to step (3) for manual verification.

[0053] (3) Record the protection signal information that has been manually checked and corrected and its correct classification results as cluster centers, and initialize the similarity level threshold.

[0054] When manually verifying the automatic classification results, for protection signals of classification failure or misclassification, the system automatically records their complete information as cluster centers, and the recording format is as follows:

[0055] Protection signal name: [Signal name]

[0056] Power topology information: [Power plant / bay / equipment / wiring method]

[0057] Automatic classification result: [Initial category]

[0058] Manual correction result: [Correct Category]

[0059] Similarity threshold: [Initial threshold t]

[0060] A similarity threshold value t is initialized for each cluster center. This threshold value is automatically set according to the information completeness of the protection signal: a lower threshold is set for protection signals with known electrical properties and names containing clear feature points; a higher threshold is set for protection signals that are in common intervals or have unclear electrical properties.

[0061] (4) Using a multi-dimensional cascaded matching method, the similarity level between the signal to be classified and protected and the cluster center is calculated. When the similarity level reaches or exceeds the current threshold level, it is classified into the standard protection type corresponding to the cluster center.

[0062] In the subsequent automatic classification of protection signals, for protection signals that have the same or similar electrical attributes (including voltage level and bay type) as the cluster center, a multi-dimensional cascade matching method is used to calculate the similarity level between them and each cluster center.

[0063] Multi-dimensional cascading matching includes the following three dimensions:

[0064] D1 (Text Similarity): After removing English characters from the protected signal name, calculate its text similarity (LCS length / original string length) with the cluster center signal name based on the longest common subsequence (LCS), and use the edit distance as the lower limit for verification;

[0065] D2 (Feature Vector Similarity): Calculates the coding consistency between the protection signal and the cluster center in terms of bay type, wiring method, primary equipment type and protection type based on the multi-dimensional feature space;

[0066] D3 (Topology Attribute Matching Degree): The matching degree of the plant wiring method, bay composition, naming consistency of protection signals throughout the plant, and overall wiring method.

[0067] The cascading rules are as follows:

[0068] When D1 is greater than 95%, the similarity level is output as 6 based solely on D1.

[0069] When D1 is in the 50% to 95% range, D2 and D3 are activated to participate in the comprehensive calculation, and the similarity level is output as 3 to 5.

[0070] When D1 is less than 50%, the weight of D1 is reduced, and D2 and D3 are used as the main components for comprehensive calculation, outputting a similarity level of 1 to 3.

[0071] Let sig2 be a new protection signal, and sig1 be an existing cluster center (with a current threshold of t). Both sig1 and sig2 are structured data. Calculate the similarity level L between sig2 and sig1:

[0072] If L ≥ t (i.e., the similarity level reaches or exceeds the threshold level), then sig2 will be classified into the standard protection type corresponding to sig1.

[0073] If L < t (insufficient similarity), then maintain the original automatic classification result of sig2.

[0074] (5) Gradually lower the similarity level threshold as the number of successful clusterings accumulates to improve clustering efficiency; when over-clustering leads to classification errors and is manually corrected, immediately raise the similarity level threshold.

[0075] Lowering the threshold (expanding coverage): When a cluster center successfully completes several cluster corrections and passes all manual verifications, it indicates that the current threshold level setting is too strict. The similarity level threshold value of the cluster center is automatically lowered (i.e., the t value is reduced, for example, from t=5 to t=4), allowing signals with lower similarity levels to enter the cluster, thereby expanding the cluster coverage and reducing the number of signals that require manual intervention.

[0076] Raising the threshold (correcting errors): When over-clustering causes a correctly automatically classified protection signal to be incorrectly assigned to a cluster center, and manual verification negates this correction, it indicates that the current threshold level is too lenient. The similarity threshold value for that cluster center is immediately raised. The increase is proportional to the distance between the current threshold and the initial threshold; that is, the farther the current threshold is from the initial threshold, the greater the increase (e.g., from t=4 to t=5). If the current threshold has already returned to its initial value, it bounces back to the initial value +1 to ensure immediate interception of similar errors. Through this proportional increase mechanism, the threshold value of the cluster center can quickly converge to a suitable strict range during multiple error correction processes.

[0077] Through the aforementioned two-way adjustment mechanism, the similarity level threshold of each cluster center can adaptively converge to the optimal value, achieving a balance between cluster coverage and accuracy.

[0078] Memory and Promotion: All clustering correction records, similarity threshold change trajectories, and clustering results are saved as structured log files. Other regional sites within the same system can import these log files to reuse the same clustering rules, eliminating the need for manual verification and accumulation, and significantly reducing the cost of multi-site deployment.

[0079] The semi-supervised classification system for protection signals based on multi-dimensional cascaded matching described in this invention includes:

[0080] The multidimensional feature space construction module is used to construct a multidimensional feature space containing interval type, wiring method, equipment type and protection type based on the topology interval classification results;

[0081] The automatic classification and matching module is used to map protection signals to a multi-dimensional feature space and calculate similarity to obtain initial classification results.

[0082] A semi-supervised clustering optimization module is used to record manually verified and corrected cluster centers and manage the adaptive adjustment of similarity thresholds;

[0083] The memory and promotion module is used to save cluster correction records as structured logs and supports other on-site import and reuse.

[0084] The computer device of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above method.

[0085] The computer-readable storage medium of the present invention stores a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

Claims

1. A semi-supervised classification method for protection signals based on multi-dimensional cascaded matching in the field of power automation, characterized in that, include: (1) Based on the bay classification results based on power topology and graph algorithm, a multi-dimensional feature space is constructed that includes information on substation type, bay type, wiring method within the bay, corresponding primary equipment type and protection type; (2) Using a multi-dimensional cascaded matching method, the matching degree between the protection signal to be classified and each candidate standard type is calculated, the candidate set is divided into multiple confidence level clusters, and the optimal match is selected as the initial classification result after electrical logic verification. Clusters with higher feature matching degree adopt a more lenient matching threshold. Within each cluster, candidates that meet the threshold conditions are sorted by edit distance and the best is selected. If all candidates fail to meet the threshold requirements of the cluster, the classification is judged as a failure and the process is transferred to manual verification. (3) Record the protection signal information that has been manually checked and corrected and its correct classification results as cluster centers, and initialize the similarity level threshold; (4) Using a multi-dimensional cascaded matching method, the similarity level between the signal to be classified and protected and the cluster center is calculated. When the similarity level reaches or exceeds the current threshold level, it is classified into the standard protection type corresponding to the cluster center. (5) Gradually lower the similarity level threshold as the number of successful clusterings accumulates to improve clustering efficiency; when over-clustering leads to classification errors and is manually corrected, immediately raise the similarity level threshold.

2. The semi-supervised classification method for protection signals based on multi-dimensional cascade matching in the field of power automation according to claim 1, characterized in that, In step (1), the power topology relationship and the interval classification results of the graph algorithm include the interval type dimension, the wiring method dimension within the interval, the primary equipment type dimension, and the protection type dimension; Among them, an undirected graph is constructed based on the power system's plant information, bay information, equipment information, and equipment association information. It is then divided into several undirected subgraphs based on connected components and voltage levels to determine the power topology. A breadth-first search algorithm is used to perform the first standard bay classification based on the relationship between the bus and the main transformer; the results of the first classification and the attributes of the circuit components are combined to perform the second standard bay classification, thus completing the automatic classification of bay types and obtaining the bay classification results.

3. The semi-supervised classification method for protection signals based on multi-dimensional cascade matching in the field of power automation according to claim 1, characterized in that, In step (2), the multi-dimensional cascaded matching method uses feature vector similarity as the main factor and the longest common subsequence similarity as the auxiliary factor for clustering. The calculation method of the longest common subsequence similarity is as follows: after removing the English characters in the protection signal name, calculate the ratio of the length of the longest common subsequence of the protection signal name and the candidate standard type name to the length of the original string. The clustering rules for multi-dimensional cascaded matching are as follows: candidates with completely matched feature vectors are assigned to cluster c1; those with feature vectors that are contained in each other and have a high similarity of their longest common subsequence are assigned to cluster c2; those with feature vectors that are intertwined and have a high similarity of their longest common subsequence, or those that are contained in each other and have a low similarity of their longest common subsequence, are assigned to cluster c3; cluster c1 uses the most lenient matching threshold, cluster c2 is the next most lenient, and cluster c3 is the most lenient. After matching is completed, electrical logic post-verification is required. Verification items include voltage level consistency, equipment type matching degree, wiring method compatibility, protection stage rationality, and interval assignment correctness. If all pass, the matching is confirmed to be successful. If any fails, the current category is removed from the candidate set and cluster matching is re-executed until the matching is successful or the candidate set is exhausted.

4. The semi-supervised classification method for protection signals based on multi-dimensional cascaded matching in the field of power automation according to claim 1, characterized in that, In step (4), the multi-dimensional cascaded matching includes three dimensions: the first dimension is the text similarity based on the longest common subsequence, the second dimension is the feature vector similarity, and the third dimension is the matching degree of the plant and station and the interval attribute. When the text similarity of the first dimension is greater than 95%, the similarity level 6 is output. When it is in the range of 50% to 95%, the second and third dimensions are activated to participate in the comprehensive calculation and the similarity level 3 to 5 is output. When it is less than 50%, the weight of the first dimension is reduced and the second and third dimensions are used as the main factors to output the similarity level 1 to 3. The output similarity level is compared with the current threshold level. When it reaches or exceeds the threshold level, it is classified into the standard protection type corresponding to the cluster center.

5. The semi-supervised classification method for protection signals based on multi-dimensional cascade matching in the field of power automation according to claim 1, characterized in that, In step (3), the format of the manual verification and correction record includes the protection signal name, power topology information, automatic classification result, manual correction result and initial similarity threshold value; each record serves as an independent cluster center.

6. The semi-supervised classification method for protection signals based on multi-dimensional cascade matching in the field of power automation according to claim 1, characterized in that, In step (5), when the number of successful corrections of a cluster center reaches a set threshold, the similarity level threshold of the cluster center is automatically lowered; when a protection signal that was originally correctly classified by automatic classification is incorrectly clustered and is rejected by manual qualification, the similarity level threshold of the cluster center is immediately raised.

7. The semi-supervised classification method for protection signals based on multi-dimensional cascade matching in the field of power automation according to claim 1, characterized in that, All clustering correction records, similarity threshold change trajectories, and clustering results are saved as structured log files. Other sites in the same area can reuse the same clustering rules by importing these log files.

8. A semi-supervised classification system for protection signals based on multi-dimensional cascaded matching, characterized in that, include: The multidimensional feature space construction module is used to construct a multidimensional feature space containing bay type, wiring method, equipment type and protection type based on the topology bay classification results; The automatic classification and matching module is used to map protection signals to a multi-dimensional feature space and calculate similarity to obtain initial classification results; A semi-supervised clustering optimization module is used to record manually verified and corrected cluster centers and manage the adaptive adjustment of similarity thresholds; The memory and promotion module is used to save clustering correction records as structured logs and supports other on-site import and reuse.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images