An automatic matching method and system for power system secondary equipment setting check

Abstract

The application discloses a kind of power system secondary equipment setting value checking automatic matching method and system, method includes: obtaining the setting value list and the list of keeping signal of same equipment;Two lists in setting value item are traversed, if the matching relationship of setting value item is in artificial intervention table, corresponding matching pair is generated, if the matching relationship of setting value item is not in artificial intervention table, setting value item is input to the model that is trained to obtain corresponding similarity;Similarity is greater to generate corresponding matching pair;If similarity is smaller or generated matching pair is wrong, correct the matching relationship of corresponding setting value item and record in artificial intervention table;Matching relationship data in artificial intervention table is added to model data set to retrain model.The application can adapt to field environment in running process, constantly learn matching relationship, maintain higher matching precision in longer time, compared with prior art matching accuracy is high, adaptation range is wide, it is helpful to establish perfect power system secondary equipment management system.

Description

Technical Field

[0001] This invention relates to the management of secondary equipment settings, and more particularly to an automatic matching method and system for verifying secondary equipment settings in a power system. Background Technology

[0002] Since primary equipment in power systems is usually high-voltage, high-current equipment, such as 500kV and 1000kV, it is not convenient to directly observe its operating status. Therefore, it is necessary to use secondary equipment to observe, control and protect the power system. The reasonable setting of secondary equipment setting parameters has naturally become the top priority to ensure the stable operation of the power system.

[0003] Generally, the process for changing the settings of secondary equipment is as follows: First, the setting system calculates and generates a setting sheet, then the setting sheet is submitted to the management department for approval. After approval, the maintenance personnel adjust the equipment settings according to the setting sheet. This manual method of modifying settings is not only inefficient but also prone to errors. Based on this, an automatic setting change system with string matching as its core has been developed. Theoretically, this system is sufficient to meet the requirements, but in actual operation of secondary equipment in the field, the following three problems exist:

[0004] 1. Equipment produced by different manufacturers, and even different models of equipment from the same manufacturer, may have the same set value function, but different set value names;

[0005] 2. For the same equipment in different plants, due to non-standard construction of substations, the set value items may have the same function but different set value names;

[0006] 3. Fixed values ​​with the same function may have different origins and names.

[0007] The aforementioned problems result in low matching accuracy of the automatic setpoint change system, requiring multiple manual reworks and leading to low overall work efficiency. Summary of the Invention

[0008] The technical problem to be solved by this invention is to provide an automatic matching method and system for setting value verification of secondary equipment in power systems, which utilizes a trained setting value name similarity model to complete setting value matching and effectively improves matching accuracy.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0010] An automatic matching method for setting verification of secondary equipment in a power system includes the following steps:

[0011] S1. Obtain the setting value list and the guarantee value list. The setting value list includes all setting values ​​of the target device in the setting value sheet of the setting system, and the guarantee value list includes all setting values ​​of the target device in the guarantee system database.

[0012] S2. Traverse the first fixed value item in the fixed value list and the second fixed value item in the credit guarantee fixed value list. If the matching relationship between the first fixed value item and the second fixed value item is in the manual intervention table, generate the corresponding matching pair. If the matching relationship between the first fixed value item and the second fixed value item is not in the manual intervention table, input the first fixed value item and the second fixed value item into the fixed value name similarity model to obtain the corresponding similarity. If the maximum similarity corresponding to the first fixed value item or the second fixed value item is greater than the first threshold, generate a matching pair between the first fixed value item or the second fixed value item and the second fixed value item or the first fixed value item with the maximum similarity.

[0013] S3. If the maximum similarity corresponding to the first or second fixed value item is less than the first threshold, or the generated matching pair is a wrong match, correct the matching relationship of the corresponding first and second fixed value items and record the correct matching relationship in the manual intervention table; use the matching relationship data in the manual intervention table to update the model dataset and retrain the fixed value name similarity model.

[0014] Furthermore, step S2 specifically includes:

[0015] S21. Iterate through and select the current first value item in the list of setpoints;

[0016] S22. Iterate through and select the current second value item in the list of guaranteed value settings;

[0017] S23. If the matching relationship between the current first fixed value item and the current second fixed value item already exists in the manual intervention table, then proceed to step S27; otherwise, proceed to step S24.

[0018] S24. Input the current first fixed value and the current second fixed value into the trained fixed value name similarity model to obtain the corresponding similarity;

[0019] S25. Return to step S22 until the list of guaranteed credit values ​​has been traversed.

[0020] S26. If the maximum similarity corresponding to the current first fixed value item is less than the first threshold, then jump to step S3; if the maximum similarity corresponding to the current first fixed value item is greater than the first threshold, then jump to step S27.

[0021] S27. Form a matching pair between the current first fixed value item and the target second fixed value item, and delete the current first fixed value item and the target second fixed value item from the set fixed value list and the guarantee fixed value list respectively. The target second fixed value item is the second fixed value item that has a matching relationship with the current first fixed value item in the manual intervention table, or the second fixed value item corresponding to the maximum similarity of the current first fixed value item.

[0022] S28. Verify whether the matching pair formed by the current first fixed value item and the target second fixed value item is correct. If yes, proceed to step S29; otherwise, proceed to step S3.

[0023] S29. Return to step S21 until the setpoint list has been traversed.

[0024] Furthermore, in step S3, when correcting the matching relationship between the corresponding first and second fixed-value items and recording the correct matching relationship in the manual intervention table, the specific steps include:

[0025] Obtain the first and second fixed-value terms from the corrected matching relation;

[0026] Associate the first setpoint item with the second setpoint item and the same standard template setpoint item, and store the correspondence between the first setpoint item and the standard template setpoint item as the matching relationship data between the newly added setpoint item and the standard template setpoint item. Store the correspondence between the second setpoint item and the standard template setpoint item as the matching relationship data between the newly added guarantee setpoint item and the standard template setpoint item.

[0027] Furthermore, when updating the model dataset and retraining the fixed-value name similarity model using the matching relationship data from the manual intervention table, the following steps are included:

[0028] If the matching data in the current manual intervention table is less than the second threshold, return to step S2;

[0029] If the matching relationship data in the current manual intervention table is greater than the second threshold, add the matching relationship data in the current manual intervention table to the model dataset, and retrain the fixed-value name similarity model using the model dataset, then return to step S2.

[0030] Furthermore, after retraining the fixed-value name similarity model using the aforementioned model dataset, the process also includes:

[0031] Convert the current manual intervention table into a historical backup table, and create a new manual intervention table. Iterate through the first and second fixed value items in the historical backup table and input them into the trained fixed value name similarity model to obtain the corresponding similarity.

[0032] If the maximum similarity corresponding to the first or second fixed value item is greater than the first threshold, a matching pair between the first or second fixed value item and the second or first fixed value item with the maximum similarity is generated and the matching pair is verified to be correct; if the maximum similarity corresponding to the first or second fixed value item is less than the first threshold, or the matching pair is a mismatch, the matching relationship between the corresponding first and second fixed value items is corrected and the correct matching relationship is recorded in a new manual intervention table.

[0033] Furthermore, in step S2, when inputting the first and second fixed-value terms into the trained fixed-value name similarity model, the following steps are included:

[0034] S201. Preprocess the text of the first fixed value item and the second fixed value item to obtain the first input text and the second input text;

[0035] S202. Encode the first input text and the second input text to obtain the first word vector and the second word vector;

[0036] S203. Input the first word vector and the second word vector into the attention model respectively to calculate the corresponding weights, and obtain the weighted first word vector and the second word vector;

[0037] S204. Input the weighted first word vector and the second word vector into the Siamese bidirectional GUA neural network of the fixed-value name similarity model to obtain the feature vector;

[0038] S205. Input the feature vector into the similarity calculation neural network of the fixed-value name similarity model to obtain the similarity between the first fixed-value term and the second fixed-value term.

[0039] Furthermore, step S201, text preprocessing, includes: converting Roman numerals, pseudo-Roman numerals, and Chinese numerals to Arabic numerals; replacing non-standard keywords with standard keywords; converting all letters to uppercase; and performing word segmentation using a Chinese lexical analysis toolkit.

[0040] Furthermore, the similarity calculation neural network in step S204 includes two fully connected layers and one normalization layer. The similarity calculation formula is as follows:

[0041] S = sigmoid(F2(F1(X1;X2)))

[0042] Where sigmoid is the normalization function, F1 and F2 represent the output vectors of each fully connected layer, and X1 and X2 represent the input vectors of each fully connected layer. For a fully connected layer, we have:

[0043] F(X) = relu(wX + b)

[0044] Where ReLU is the activation function, w is the weight matrix, b is the bias, and X is the input vector.

[0045] This invention also proposes an automatic matching system for setting verification of secondary equipment in a power system, comprising:

[0046] The setting value list acquisition unit is used to acquire the setting value list and the maintenance value list of the same device.

[0047] The automatic matching unit iterates through the first fixed value item in the setting value list and the second fixed value item in the guarantee value list. If the matching relationship between the first and second fixed value items is in the manual intervention table, a corresponding matching pair is generated. If the matching relationship between the first and second fixed value items is not in the manual intervention table, the first and second fixed value items are input into the fixed value name similarity model to obtain the corresponding similarity. If the maximum similarity corresponding to the first or second fixed value item is greater than a first threshold, a matching pair between the first or second fixed value item and the second or first fixed value item with the maximum similarity is generated. If the maximum similarity corresponding to the first or second fixed value item is less than the first threshold, a setting value item matching failure log is generated.

[0048] The manual intervention table update unit is used to obtain the matching failure log of the setpoint item, or when the matching is incorrect, to correct the matching relationship between the corresponding first and second setpoint items and record the correct matching relationship in the manual intervention table. The manual intervention table consists of two parts: a standard template setpoint index table and a non-standard setpoint association data table. The standard template setpoint index table pre-stores standardized template setpoint items and their unique corresponding indices, while the non-standard setpoint association data table stores the correspondence between the indices of non-standard items and standard template setpoint items. When processing setpoint items that fail to match or are incorrectly matched, the process essentially involves associating the setpoint item and the guarantee setpoint item with the same standard template setpoint item, thereby achieving a match between them. The advantage of introducing standard template values ​​is that if the fixed value item is directly associated with the credit guarantee value item, only one matching relationship data will be generated by one manual intervention. However, after the introduction, two matching relationship data will be generated: the fixed value item and the standard template value item, and the credit guarantee value item and the standard template value item. Sufficient data helps to obtain a more complete value name similarity model in the model training unit.

[0049] Furthermore, it also includes a model training unit for updating the model dataset and retraining the fixed-value name similarity model using matching relationship data from the manual intervention table.

[0050] Furthermore, the manual intervention table update unit is also used to convert the current manual intervention table into a historical backup table and create a new manual intervention table when the matching relationship data in the current manual intervention table is greater than the second threshold. The automatic matching unit is also used to input the first fixed value item and the second fixed value item in the historical backup table into the trained fixed value name similarity model to obtain the corresponding similarity.

[0051] Compared with the prior art, the present invention has the following advantages:

[0052] This invention first reads the fixed values ​​and the guaranteed value; then, it preprocesses the names of the fixed value items; next, it converts them into word vectors; finally, it inputs them into a trained fixed value name similarity model to obtain the fixed value name similarity and complete the fixed value matching. During the fixed value matching process, it checks for matching failures, incorrect matches, and missing matches, corrects errors manually, and updates the manual intervention records in the manual intervention table. When the number of manual intervention records reaches a threshold, the fixed value name similarity model is retrained. This invention can continuously learn during operation, ensuring high fixed value matching accuracy. Attached Figure Description

[0053] Figure 1 This is a flowchart of an embodiment of the present invention.

[0054] Figure 2 This is a flowchart of the fixed-value name similarity algorithm according to an embodiment of the present invention.

[0055] Figure 3 This is a flowchart of the input data preprocessing process according to an embodiment of the present invention.

[0056] Figure 4 This is a structural diagram of the fixed-value name similarity model in an embodiment of the present invention.

[0057] Figure 5 This is a diagram of the bidirectional GUA neural network structure of the present invention.

[0058] Figure 6 This is a topological diagram of the GRU neuron structure of the present invention.

[0059] Figure 7 This is a flowchart of retraining the fixed-value name similarity model in an embodiment of the present invention.

[0060] Figure 8 This is a comparison of the error metrics between the algorithm of this invention and conventional algorithm implementations. Detailed Implementation

[0061] The present invention will be further described below with reference to the accompanying drawings and specific preferred embodiments, but this does not limit the scope of protection of the present invention.

[0062] Example 1

[0063] This embodiment proposes an automatic matching method for setting verification of secondary equipment in power systems. Compared with traditional methods, it has higher matching accuracy and a wider range of applications, which helps to establish a more complete management system for secondary equipment in power systems. Figure 1 As shown, it includes the following steps:

[0064] S1. Obtain the setting value list and the guarantee value list of the same device. In this embodiment, all setting value items of the target device are read from the setting value sheet of the setting system and recorded as the setting value list. All setting value items of the target device are searched from the guarantee value system database and recorded as the guarantee value list.

[0065] S2. Traverse the first fixed value item in the fixed value list and the second fixed value item in the credit guarantee fixed value list. If the matching relationship between the first fixed value item and the second fixed value item is in the manual intervention table, generate the corresponding matching pair. If the matching relationship between the first fixed value item and the second fixed value item is not in the manual intervention table, input the first fixed value item and the second fixed value item into the fixed value name similarity model to obtain the corresponding similarity. If the maximum similarity corresponding to the first fixed value item or the second fixed value item is greater than the first threshold, generate a matching pair between the first fixed value item or the second fixed value item and the second fixed value item or the first fixed value item with the maximum similarity.

[0066] S3. If the maximum similarity corresponding to the first or second fixed value item is less than the first threshold, or the generated matching pair is a wrong match, correct the matching relationship of the corresponding first and second fixed value items and record the correct matching relationship in the manual intervention table; use the matching relationship data in the manual intervention table to update the model dataset and retrain the fixed value name similarity model.

[0067] The following provides a detailed explanation of each step.

[0068] like Figure 1 As shown, step S2 in this embodiment specifically includes:

[0069] S21. Select the current first fixed value item in the fixed value list and use the name of the current first fixed value item as the input of the fixed value name similarity algorithm.

[0070] S22. Select the current second value item in the credit rating value list, and use the name of the current second value item as the input of the value name similarity algorithm.

[0071] S23. If the matching relationship between the current first fixed value item and the current second fixed value item already exists in the manual intervention table, then proceed to step S27; otherwise, proceed to step S24.

[0072] S24. Input the current first fixed value and the current second fixed value into the trained fixed value name similarity model. The fixed value name similarity model executes the fixed value name similarity algorithm to obtain the corresponding similarity.

[0073] S25. Return to step S22 until the list of guaranteed credit values ​​has been traversed.

[0074] S26. If the maximum similarity of the current first fixed value item is less than the first threshold, it is considered that the matching of the current first fixed value item has failed. The matching failure log of the current first fixed value item is output and the process jumps to step S3. If the maximum similarity of the current first fixed value item is greater than the first threshold, the matching is successful and the process jumps to step S27. In this embodiment, the first threshold is 0.5.

[0075] S27. Form a matching pair between the current first fixed value item and the target second fixed value item, and delete them from the set value list and the guarantee value list. The target second fixed value item is the second fixed value item corresponding to the current first fixed value item in the manual intervention table, or the second fixed value item corresponding to the maximum similarity of the current first fixed value item.

[0076] S28. Verify whether the matching pair formed by the current first fixed value item and the target second fixed value item is correct. If yes, proceed to step S29; otherwise, proceed to step S3. In this embodiment, manual verification is used when verifying whether the matching pair is correct.

[0077] S29. Return to step S21 until the setpoint list has been traversed.

[0078] In this embodiment, while executing step S2, step S3 is also executed to correct errors manually in a timely manner. Whenever a mismatch or omission occurs, the correct value item name matching pair is manually specified and saved in the database to update the manual intervention table. Therefore, in step S3 of this embodiment, correcting the matching relationship between the corresponding first and second value items and recording the correct matching relationship in the manual intervention table specifically includes:

[0079] A1. If a matching failure log for the set value item is obtained, or if the matching is incorrect, then proceed to step A2.

[0080] A2. Obtain the corrected matching pairs and update the corrected matching pairs in the current manual intervention table, specifically including:

[0081] Obtain the first and second fixed-value terms from the corrected matching relation;

[0082] Associate the first setpoint item with the second setpoint item and the same standard template setpoint item, and store the correspondence between the first setpoint item and the standard template setpoint item as the matching relationship data between the newly added setpoint item and the standard template setpoint item. Store the correspondence between the second setpoint item and the standard template setpoint item as the matching relationship data between the newly added guarantee setpoint item and the standard template setpoint item.

[0083] In this embodiment, the manual intervention table can be divided into two parts: a standard template value index table and a non-standard value association data table. The standard template value index table pre-stores standardized template value items and their unique corresponding indices. The non-standard value association data table stores the correspondence between indices of non-standard items and standard template value items. When processing value items that fail to match or mismatch, the process essentially involves associating the fixed value item and the guaranteed value item with the same standard template value item, thereby achieving a match. The advantage of introducing standard template values ​​is that directly associating the fixed value item with the guaranteed value item would only generate one matching relationship data per manual intervention. However, after introducing standard template values, two matching relationship data points are generated: one between the fixed value item and the standard template value item, and another between the guaranteed value item and the standard template value item. Sufficient data helps to obtain a more complete value name similarity model during model training.

[0084] like Figure 7 As shown, whenever the number of matching pairs of fixed-value item names in the manual intervention table reaches a certain amount, these matching pairs are added to the model dataset as new data. The model dataset with the added new data is then used to retrain and update the fixed-value name similarity model to improve the matching success rate. Therefore, in step S3 of this embodiment, updating the model dataset and retraining the fixed-value name similarity model using the matching relationship data in the manual intervention table includes:

[0085] B1. If the matching relationship data in the current manual intervention table is less than the second threshold, return to step S2; if the matching relationship data in the current manual intervention table is greater than the second threshold, add the matching relationship data in the current manual intervention table to the model dataset, and retrain the fixed value name similarity model using the model dataset. In this embodiment, the fixed value name similarity model belongs to the neural network model. The specific training process of using the dataset to train the neural network model is well known to those skilled in the art and will not be described in detail here.

[0086] B2. Convert the current manual intervention table into a historical backup table and create a new manual intervention table. The historical backup table is obtained by adding the model update time to the name of the current manual intervention table.

[0087] B3. Input the fixed value items in the historical backup table as the data to be matched into the model to obtain the matching results. Specifically, traverse the first and second fixed value items in the historical backup table and input them into the trained fixed value name similarity model to obtain the corresponding similarity. If the maximum similarity corresponding to the first or second fixed value item is greater than the first threshold, generate a matching pair between the first or second fixed value item and the second or first fixed value item with the maximum similarity. The specific process is basically the same as steps S21 to S29, except that it is not necessary to determine whether the matching relationship between the first and second fixed value items already exists in the manual intervention table.

[0088] B4. For the fixed-value items that still match incorrectly in step B3, record the correct matching pairs associated with them in the manual intervention table. Specifically, verify whether the matching pairs are correct. If the maximum similarity corresponding to the first or second fixed-value item is less than the first threshold, or if the matching pair is a wrong match, correct the matching relationship between the corresponding first and second fixed-value items and record the correct matching relationship in a new manual intervention table.

[0089] like Figure 2 As shown, in step S2, when the first and second fixed-value terms are input into the trained fixed-value name similarity model, the fixed-value name similarity algorithm executed by the fixed-value name similarity model includes the following steps:

[0090] S201. Preprocess the text of the first fixed value item and the second fixed value item to obtain the first input text and the second input text;

[0091] S202. Encode the first input text and the second input text to obtain the first word vector and the second word vector;

[0092] S203. In this embodiment, after obtaining the first word vector and the second word vector, the corresponding weight vector is calculated through the attention model to obtain the weighted first word vector and the second word vector.

[0093] S204. Input the weighted first word vector and the second word vector into the Siamese bidirectional GUA neural network of the fixed-value name similarity model to obtain the feature vector;

[0094] S205. Input the feature vector into the similarity calculation neural network of the fixed-value name similarity model to obtain the similarity between the first fixed-value term and the second fixed-value term.

[0095] like Figure 3As shown, in step S201, text preprocessing includes: uniformly converting Roman numerals, pseudo-Roman numerals and Chinese numerals into Arabic numerals; replacing non-standard keywords with standard keywords; converting all letters to uppercase; performing word segmentation using a lexical analysis toolkit. In this embodiment, the lexical analysis toolkit adopts the THULAC word segmentation toolkit, which is a Chinese lexical analysis toolkit constructed by Tsinghua University, has the largest existing Chinese manual word segmentation and part-of-speech tagging database, and has outstanding word segmentation performance.

[0096] It should be noted that, in general, the Chinese phrase matching task also requires a stop word removal operation in the preprocessing stage, that is, removing auxiliary words such as "de" and "le" in the text. However, in this embodiment, such auxiliary words hardly appear, so this operation can be omitted to speed up the matching speed.

[0097] In step S202, the RoBERTa-wwm-ext model is used to encode the text. The RoBERTa-wwm-ext model is released by the HIT iFLYTEK Joint Laboratory, adopts the whole word Mask technology and is trained on a large amount of word data such as Wikipedia, and is currently an excellent phrase encoding method with outstanding performance.

[0098] In step S203, different from the conventional phrase matching task, in the application scenario of this embodiment, there are widely existing phrases that differ by one or two characters but have completely different actual meanings, for example: "high-voltage side PT disconnection" and "high-voltage side CT disconnection", "Section I is valid" and "Section III is valid", and so on. The existence of these phrases greatly affects the subsequent calculation of the similarity of the setting value names. In response to this situation, weights are calculated through the attention model to assign higher weights to the key characters in the phrase, so as to obtain a weighted word vector.

[0099] In step S204, a Siamese bidirectional GUA neural network is used to better learn the correlation information between data, so that the matching speed and accuracy are better. As Figure 4 shown, the bidirectional GUA neural network has two inputs, and the weighted word vectors are respectively input into the left and right bidirectional GUA neural networks. The parameters of these two neural networks are shared, so the two word vectors can be mapped to the same dimensional space to form two new feature vectors.

[0100] As Figure 5 shown, the bidirectional GUA neural network is specifically formed by superimposing two GUA neural networks with opposite directions, which are respectively a forward GUA network and a backward GUA network, and its output y is the sum of the outputs h of the two networks (1) and h (2) , and this structure enables it to better learn the correlation information between data.

[0101] The inside of the GUA neural network is as Figure 6As shown, it is composed of GRU neurons stacked in a certain direction. GRU neurons are composed of reset gates and update gates. The specific principle is as follows:

[0102] (1) Reset and update doors:

[0103] r t =σ(W xr x t +W hr h t-1 +b r )

[0104] z t =σ(W xz x t +W hz h t-1 +b z )

[0105] Where σ is the activation function, W is the weight matrix, b is the bias, and x t h is the input to the t-th neuron. t-1 Let z be the output of neuron t-1, r be the reset gate used to determine which information needs to be retained, and z be the output of neuron t-1. t The gate is updated to determine which information needs to be transmitted.

[0106] (2) Hidden state update:

[0107]

[0108] in, h is a combination of new information at the current moment and hidden information from the previous moment. t The current hidden state is also the output of this neuron.

[0109] The similarity calculation neural network in step S204 of this embodiment includes two fully connected layers and one normalization layer. The formula for calculating similarity is as follows:

[0110] S = sigmoid(F2(F1(X1;X2)))

[0111] Where sigmoid is the normalization function, F1 and F2 represent the output vectors of each fully connected layer, and X1 and X2 represent the input vectors of each fully connected layer. For a fully connected layer, we have:

[0112] F(X) = relu(wX + b)

[0113] Where ReLU is the activation function, w is the weight matrix, b is the bias, and X is the input vector. In this embodiment, X is the original name of the matching two terms, namely the original name of the first and second terms.

[0114] Through the above steps, this embodiment first reads the setpoint and the guarantee value; then preprocesses the name of the setpoint item; then converts it into a word vector; finally, it is input into the setpoint name similarity algorithm to obtain the setpoint name similarity and complete the setpoint matching. In addition, during the operation of this method, whenever the number of manually intervened records reaches a threshold, the model is retrained to improve the accuracy of setpoint matching, effectively solving the problems of low matching accuracy and narrow applicability in the existing technology.

[0115] Example 2

[0116] The setting value list acquisition unit is used to acquire the setting value list and the maintenance value list of the same device.

[0117] The automatic matching unit iterates through the first fixed value item in the setting value list and the second fixed value item in the guarantee value list. If the matching relationship between the first and second fixed value items is in the manual intervention table, a corresponding matching pair is generated. If the matching relationship between the first and second fixed value items is not in the manual intervention table, the first and second fixed value items are input into the fixed value name similarity model to obtain the corresponding similarity. If the maximum similarity corresponding to the first or second fixed value item is greater than a first threshold, a matching pair between the first or second fixed value item and the second or first fixed value item with the maximum similarity is generated. If the maximum similarity corresponding to the first or second fixed value item is less than the first threshold, a setting value item matching failure log is generated.

[0118] The manual intervention table update unit is used to obtain the matching failure log of the setpoint item, or when the matching is a mismatch, to correct the matching relationship of the corresponding first and second setpoint items and record the correct matching relationship in the manual intervention table. The manual intervention table consists of two parts: a standard template setpoint index table and a non-standard setpoint association data table. The standard template setpoint index table pre-stores the standardized template setpoint items and their unique corresponding indices. The non-standard setpoint association data table stores the correspondence between the indices of non-standard items and standard template setpoint items. When maintenance personnel handle value items that fail to match or mismatch, they essentially associate the fixed value item and the guarantee value item with the same standard template value item to achieve matching between the two. The advantage of introducing the standard template value is that if the fixed value item and the guarantee value item are directly associated, only one matching relationship data will be generated by one manual intervention. However, after introducing the standard template value, two matching relationship data will be generated: one between the fixed value item and the standard template value item, and the other between the guarantee value item and the standard template value item. Sufficient data helps to obtain a more complete value name similarity model in the model training unit.

[0119] In addition, the manual intervention table update unit is also used to convert the current manual intervention table into a historical backup table and create a new manual intervention table when the number of matching pairs in the current manual intervention table exceeds the second threshold. The fixed value items in the historical backup table are used as matching data again to obtain matching results. The automatic matching unit is also used to input the first and second fixed value items in the historical backup table into the trained fixed value name similarity model to obtain the corresponding similarity. Based on the relationship between the maximum similarity and the first threshold, it generates matching pairs or sets the fixed value item matching failure log. After waiting for manual verification and correction, the manual intervention table update unit records the correct matching pair associated with the incorrectly matched fixed value item in the new manual intervention table.

[0120] The model training unit is used when the number of matching pairs in the current manual intervention table exceeds the second threshold. In this case, the matching data in the current manual intervention table is added to the model dataset to update the model dataset, and the fixed-value name similarity model is retrained using the updated model dataset.

[0121] In summary, this invention discloses an automatic matching method and system for verifying the setting values ​​of secondary equipment in a power system. The method and system first obtain a list of setting values ​​for the same equipment and a list of guarantee setting values. Then, they iterate through the setting items in both lists. If the matching relationship of a setting item is in a manual intervention table, a corresponding matching pair is generated. If the matching relationship is not in the manual intervention table, the setting item is input into a trained model to obtain the corresponding similarity. If the similarity is high, a corresponding matching pair is generated. If the similarity is low or the generated matching pair is incorrect, the matching relationship of the corresponding setting item is corrected and recorded in the manual intervention table. Finally, the matching relationship data from the manual intervention table is added to the model dataset to retrain the model. This invention can adapt to the field environment during operation, continuously learn matching relationships, and maintain high matching accuracy over a long period. Compared with existing technologies, it has higher matching accuracy, a wider range of applications, and contributes to the establishment of a comprehensive power system secondary equipment management system.

[0122] like Figure 8 The figure shows a comparison of error metrics between the algorithm of this invention and conventional algorithm embodiments. Other algorithms compared are the LSTM and GRU algorithms. The evaluation metrics are the mean absolute error ratio (MAPE) and the root mean square error (RMSE). It is evident that the algorithm of this invention has higher matching accuracy.

[0123] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should fall within the protection scope of the present invention.

Claims

1. An automatic matching method for setting verification of secondary equipment in a power system, characterized in that, Includes the following steps: S1. Obtain the setting value list and the guarantee value list. The setting value list includes all setting values ​​of the target device in the setting value sheet of the setting system, and the guarantee value list includes all setting values ​​of the target device in the guarantee system database. S2. Traverse the first fixed value item in the fixed value list and the second fixed value item in the credit guarantee fixed value list. If the matching relationship between the first fixed value item and the second fixed value item is in the manual intervention table, generate the corresponding matching pair. If the matching relationship between the first fixed value item and the second fixed value item is not in the manual intervention table, input the first fixed value item and the second fixed value item into the fixed value name similarity model to obtain the corresponding similarity. If the maximum similarity corresponding to the first fixed value item or the second fixed value item is greater than the first threshold, generate a matching pair between the first fixed value item or the second fixed value item and the second fixed value item or the first fixed value item with the maximum similarity. Specifically, this includes: S21. Iterate through and select the current first value item in the list of setpoints; S22. Iterate through and select the current second value item in the list of guaranteed value settings; S23. If the matching relationship between the current first fixed value item and the current second fixed value item already exists in the manual intervention table, then proceed to step S27; otherwise, proceed to step S24. S24. Input the current first fixed value and the current second fixed value into the trained fixed value name similarity model to obtain the corresponding similarity; S25. Return to step S22 until the list of guaranteed credit values ​​has been traversed. S26. If the maximum similarity corresponding to the current first fixed value item is less than the first threshold, then jump to step S3; if the maximum similarity corresponding to the current first fixed value item is greater than the first threshold, then jump to step S27. S27. Form a matching pair between the current first fixed value item and the target second fixed value item, and delete the current first fixed value item and the target second fixed value item from the set fixed value list and the guarantee fixed value list respectively. The target second fixed value item is the second fixed value item that has a matching relationship with the current first fixed value item in the manual intervention table, or the second fixed value item corresponding to the maximum similarity of the current first fixed value item. S28. Verify whether the matching pair formed by the current first fixed value item and the target second fixed value item is correct. If yes, proceed to step S29; otherwise, proceed to step S3. S29. Return to step S21 until the setpoint list has been traversed. S3. If the maximum similarity corresponding to the first or second fixed value item is less than the first threshold, or the generated matching pair is a wrong match, correct the matching relationship of the corresponding first and second fixed value items and record the correct matching relationship in the manual intervention table; use the matching relationship data in the manual intervention table to update the model dataset and retrain the fixed value name similarity model.

2. The automatic matching method for setting value verification of secondary equipment in a power system according to claim 1, characterized in that, In step S3, correcting the matching relationship between the corresponding first and second fixed-value items and recording the correct matching relationship in the manual intervention table specifically includes: Obtain the first and second fixed-value terms from the corrected matching relation; Associate the first setpoint item with the second setpoint item and the same standard template setpoint item, and store the correspondence between the first setpoint item and the standard template setpoint item as the matching relationship data between the newly added setpoint item and the standard template setpoint item. Store the correspondence between the second setpoint item and the standard template setpoint item as the matching relationship data between the newly added guarantee setpoint item and the standard template setpoint item.

3. The automatic matching method for setting value verification of secondary equipment in a power system according to claim 1, characterized in that, In step S3, when updating the model dataset and retraining the fixed-value name similarity model using the matching relationship data from the manual intervention table, the following steps are included: If the matching data in the current manual intervention table is less than the second threshold, return to step S2; If the matching relationship data in the current manual intervention table is greater than the second threshold, add the matching relationship data in the current manual intervention table to the model dataset, and retrain the fixed-value name similarity model using the model dataset, then return to step S2.

4. The automatic matching method for setting value verification of secondary equipment in a power system according to claim 3, characterized in that, After retraining the fixed-value name similarity model using the aforementioned model dataset, the following steps are also included: Convert the current manual intervention table into a historical backup table, and create a new manual intervention table. Iterate through the first and second fixed value items in the historical backup table and input them into the trained fixed value name similarity model to obtain the corresponding similarity. If the maximum similarity corresponding to the first or second fixed value item is greater than the first threshold, a matching pair between the first or second fixed value item and the second or first fixed value item with the maximum similarity is generated and the matching pair is verified to be correct; if the maximum similarity corresponding to the first or second fixed value item is less than the first threshold, or the matching pair is a mismatch, the matching relationship between the corresponding first and second fixed value items is corrected and the correct matching relationship is recorded in a new manual intervention table.

5. The automatic matching method for setting value verification of secondary equipment in a power system according to claim 1, characterized in that, When inputting the first and second fixed-value terms into the trained fixed-value name similarity model in step S2, the following steps are included: S201. Preprocess the text of the first fixed value item and the second fixed value item to obtain the first input text and the second input text; S202. Encode the first input text and the second input text to obtain the first word vector and the second word vector; S203. Input the first word vector and the second word vector into the attention model respectively to calculate the corresponding weights, and obtain the weighted first word vector and the second word vector; S204. Input the weighted first word vector and the second word vector into the Siamese bidirectional GUA neural network of the fixed-value name similarity model to obtain feature vectors. The Siamese bidirectional GUA neural network includes two bidirectional GUA neural networks with shared parameters. The two word vectors are mapped to the same dimensional space to form two new feature vectors. The bidirectional GUA neural network is composed of two GUA neural networks stacked in opposite directions. Each GUA neural network is composed of stacked GRU neurons. The output of the bidirectional GUA neural network is the sum of the outputs of the two GUA neural networks. S205. Input the feature vector into the similarity calculation neural network of the fixed-value name similarity model to obtain the similarity between the first fixed-value term and the second fixed-value term.

6. The automatic matching method for setting value verification of secondary equipment in a power system according to claim 5, characterized in that, Step S201, text preprocessing, includes: converting Roman numerals, pseudo-Roman numerals, and Chinese numerals to Arabic numerals; replacing non-standard keywords with standard keywords; converting all letters to uppercase; and performing word segmentation using a Chinese lexical analysis toolkit.

7. The automatic matching method for setting value verification of secondary equipment in a power system according to claim 5, characterized in that, The similarity calculation neural network in step S204 consists of two fully connected layers and one normalization layer. The formula for calculating similarity is as follows: in, For normalization function, F 1. F 2 represents the output vector of each fully connected layer. X 1. X 2 represents the input vector of each fully connected layer. For a fully connected layer, we have: in, For activation function, This is the weight matrix. This is the bias value. The input vector.

8. An automatic matching system for setting verification of secondary equipment in a power system, characterized in that, include: The setting value list acquisition unit is used to acquire the setting value list and the maintenance value list of the same device. The automatic matching unit iterates through the first fixed value item in the setpoint list and the second fixed value item in the credit guarantee setpoint list. If the matching relationship between the first and second fixed value items is in the manual intervention table, a corresponding matching pair is generated. If the matching relationship between the first and second fixed value items is not in the manual intervention table, the first and second fixed value items are input into the fixed value name similarity model to obtain the corresponding similarity. If the maximum similarity corresponding to the first or second fixed value item is greater than a first threshold, a matching pair between the first or second fixed value item and the second or first fixed value item with the maximum similarity is generated. Specifically, this includes: S21. Iterate through and select the current first value item in the list of setpoints; S22. Iterate through and select the current second value item in the list of guaranteed value settings; S23. If the matching relationship between the current first fixed value item and the current second fixed value item already exists in the manual intervention table, then proceed to step S27; otherwise, proceed to step S24. S24. Input the current first fixed value and the current second fixed value into the trained fixed value name similarity model to obtain the corresponding similarity; S25. Return to step S22 until the list of guaranteed credit values ​​has been traversed. S26. If the maximum similarity corresponding to the current first fixed value item is less than the first threshold, then jump to step S3; if the maximum similarity corresponding to the current first fixed value item is greater than the first threshold, then jump to step S27. S27. Form a matching pair between the current first fixed value item and the target second fixed value item, and delete the current first fixed value item and the target second fixed value item from the set fixed value list and the guarantee fixed value list respectively. The target second fixed value item is the second fixed value item that has a matching relationship with the current first fixed value item in the manual intervention table, or the second fixed value item corresponding to the maximum similarity of the current first fixed value item. S28. Verify whether the matching pair formed by the current first fixed value item and the target second fixed value item is correct. If yes, proceed to step S29; otherwise, proceed to step S3. S29. Return to step S21 until the setpoint list has been traversed. If the maximum similarity corresponding to the first or second fixed value item is less than the first threshold, the automatic matching unit generates a fixed value item matching failure log. The manual intervention table update unit is used to obtain the matching failure log of the set value item, or when the matching is a wrong match, to correct the matching relationship of the corresponding first set value item and second set value item and record the correct matching relationship in the manual intervention table; The model training unit is used to update the model dataset and retrain the fixed-value name similarity model using the matching relationship data in the manual intervention table.

9. The automatic matching system for setting value verification of secondary equipment in a power system according to claim 8, characterized in that, The manual intervention table update unit is also used to convert the current manual intervention table into a historical backup table and create a new manual intervention table when the matching relationship data in the current manual intervention table is greater than the second threshold. The automatic matching unit is also used to input the first fixed value item and the second fixed value item in the historical backup table into the trained fixed value name similarity model to obtain the corresponding similarity.

Images