Converter station valve area fault diagnosis method based on SER event and recording information

Abstract

The invention discloses a converter station valve area fault diagnosis method based on SER events and recording information, and the method comprises the steps: carrying out a fault simulation experiment based on an RTDS simulation platform, obtaining the fault SER data of a converter station valve area, and constructing a fault SER corpus; inputting SER data in the fault SER corpus into a bidirectional encoder representation model to extract semantic feature information; inputting the extracted semantic feature information into a hybrid neural network for analysis and diagnosis, and preliminarily judging the fault type of the converter station valve region; and carrying out a fault simulation experiment based on an RTDS simulation platform, obtaining an electrical wave recording analog quantity, carrying out mechanism analysis, extracting a key electrical wave recording analog quantity, and further accurately determining the fault type of the converter station valve region. According to the invention, the accuracy and positioning precision of converter station valve area fault identification can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system fault diagnosis, and particularly relates to a converter station valve area fault diagnosis method based on SER events and recorded wave information. BACKGROUND

[0002] High-voltage direct current (HVDC) transmission technology is widely used in long-distance power transmission, power grid interconnection and new energy grid connection due to its advantages of long-distance, large capacity and low loss. The HVDC system mainly consists of a converter station, a DC line and a control and protection system. The safety and stability of the converter station, as the energy conversion hub of alternating current and direct current, directly affect the reliability of the entire power transmission system. However, the internal equipment of the converter station is complex, involving key components such as converter valves, DC buses, smoothing reactors and DC circuit breakers, and the operating environment is harsh, which is easily affected by external faults, equipment aging and control system abnormalities, leading to faults. Therefore, ensuring the safe and stable operation of the converter station is crucial for the reliability of the HVDC system.

[0003] With the continuous development of high-voltage transmission technology, the scale of the direct current system is increasing, and its fault characteristics are quite different from those of alternating current systems, such as fast short-circuit current rise, strong overvoltage impact and complex transient process. Traditional alternating current protection schemes cannot be directly applied to HVDC systems, so it is necessary to build an efficient and accurate fault diagnosis method for the converter station based on the characteristics of the direct current system. In particular, when a valve area fault occurs, the damage to the converter valve may trigger a chain of faults, affecting the stability of the entire power transmission system. Therefore, how to achieve fast and accurate fault identification and diagnosis based on converter station fault data has become an important research direction in the operation and maintenance of HVDC systems.

[0004] Currently, the diagnosis methods for direct current faults mainly include analysis methods based on electrical quantity feature extraction, signal processing and pattern recognition methods, artificial intelligence and data-driven methods. However, these methods still have certain limitations: traditional rule-based fault diagnosis relies on expert experience and is difficult to adapt to complex and variable working conditions; signal processing-based methods are greatly affected by noise interference, limiting their accuracy; intelligent diagnosis methods based on machine learning still face challenges such as feature extraction difficulty, data imbalance and lack of efficient feature fusion mechanisms when dealing with converter station valve area faults. SUMMARY

[0005] To solve the problems in the prior art, the present application provides a converter station valve area fault diagnosis method based on SER events and recorded wave information, which combines SER events and recorded wave information and other control and protection signals to improve the accuracy of fault identification and positioning precision in the valve area of the converter station.

[0006] To achieve the above technical purposes, the application adopts the following technical solutions: A converter station valve area fault diagnosis method based on SER events and recording information, comprising the following steps:

[0007] Step S1: Based on the RTDS simulation platform, a fault simulation experiment is carried out to obtain fault SER data of the converter station valve area, and a fault SER corpus is constructed;

[0008] Step S2: Input the SER data in the fault SER corpus into the bidirectional encoder representation model to extract semantic feature information;

[0009] Step S3: Input the extracted semantic feature information into the mixed neural network for analysis and diagnosis, and preliminarily judge the fault type of the converter station valve area;

[0010] Step S4: Based on the RTDS simulation platform, a fault simulation experiment is carried out to obtain electrical recording simulation quantities, perform mechanism analysis, extract key electrical recording simulation quantities, and further accurately determine the fault type of the converter station valve area.

[0011] Further, step S1 includes the following sub-steps:

[0012] Step S1.1: Based on the RTDS simulation platform, a fault simulation experiment is carried out to obtain fault SER data of the converter station valve area, and the obtained SER data is processed by dimension reduction according to the typical fault mode and protection strategy of the converter station valve area;

[0013] Step S1.2: The SER data processed by dimension reduction is standardized, including removing duplicate records, filling missing values and unifying data formats;

[0014] Step S1.3: According to the alarm level of the fault SER data, a fault event grading model of the converter station valve area is constructed, and the standardized SER data is graded through the fault event grading model;

[0015] Step S1.4: The graded SER data is mined for association rules using the Apriori algorithm, and the minimum support, minimum confidence and minimum lift are set. SER data meeting the minimum support, minimum confidence and minimum lift requirements is saved to the fault SER corpus.

[0016] Further, step S2 includes the following sub-steps:

[0017] Step S2.1: The SER data in the fault SER corpus is segmented using the WordPiece segmenter to generate token sequences;

[0018] Step S2.2: adding [CLS] and [SEP] marks to the token sequence, and embedding the token sequence to input into a bidirectional encoder representation model to extract a context semantic vector of each token;

[0019] Step S2.3: converting the extracted context semantic vector of the token into a high-dimensional embedding vector as semantic feature information of the SER data.

[0020] Further, the specific process of step S3 is as follows: a hybrid neural network is constructed, including a convolutional neural network, a bidirectional long short-term memory network, and an attention module; wherein the convolutional neural network predicts a first fault type probability of the valve area of the converter station according to the extracted semantic feature information, the bidirectional long short-term memory network adopts a forward LSTM layer and a backward LSTM layer in parallel, and respectively predicts a second fault type probability of the valve area of the converter station according to the extracted semantic feature information; the attention module adopts a dot product attention mechanism to weight and fuse the predicted first fault type probability and the second fault type probability of the valve area of the converter station to obtain a preliminary fault type of the valve area of the converter station.

[0021] Further, the process of the attention module using the dot product attention mechanism to weight and fuse the local features and the global features is as follows:

[0022]

[0023] Wherein, Q is the fault diagnosis probability of the valve area of the converter station output by the attention module, K is a feature identifier of the fault diagnosis probability of the valve area of the converter station, V is the predicted first fault type probability and the second fault type probability of the valve area of the converter station, d k is the dimension of K.

[0024] Further, the preliminary fault type of the valve area of the converter station in step S3 includes: converter differential protection and pole differential protection, Y-bridge valve short-circuit protection, Y and D-bridge valve short-circuit protection, and D-bridge valve short-circuit protection.

[0025] The key electrical wave recording analog quantities in step S4 include: high-end converter high-voltage side current IDC1P, high-end converter low-voltage side current IDC1N, YY converter transformer valve side winding current maximum IVY_MAX, YD converter transformer valve side winding current maximum IVD_MAX, Y-bridge alternating current side phase current, and AB-phase current.

[0026] Further, when the preliminary fault type of the valve area of the converter station is the converter differential protection and the pole differential protection:

[0027] If the difference between the high-voltage side current IDC1P of the high-end converter and the maximum value IVY_MAX of the valve side winding current of the YY converter transformer shows an upward trend, it is determined that the valve high-end grounding fault occurs.

[0028] If the difference between the maximum value IVD_MAX of the valve side winding current of the YD converter transformer and the low-voltage side current IDC1N of the high-end converter shows an upward trend, it is determined that the valve neutral end grounding fault occurs.

[0029] If the difference between the maximum value IVY_MAX of the valve side winding current of the YY converter transformer and the maximum value IVD_MAX of the valve side winding current of the YD converter transformer shows an upward trend, and the three-phase current of the Y bridge AC side is not always 0, it is determined that the valve midpoint grounding fault occurs.

[0030] If the difference between the maximum value IVY_MAX of the valve side winding current of the YY converter transformer and the maximum value IVD_MAX of the valve side winding current of the YD converter transformer shows an upward trend, and the current of one phase of the Y bridge AC side is always 0, it is determined that the YY valve group AC side one-way grounding fault occurs.

[0031] Further, when the preliminary discrimination of the fault type of the valve area of the converter station is the Y bridge valve short circuit protection, and the difference between the maximum value IVY_MAX of the valve side winding current of the YY converter transformer and the maximum value IVD_MAX of the valve side winding current of the YD converter transformer shows an upward trend, it is determined that the valve high-end to midpoint short circuit fault occurs.

[0032] Further, when the preliminary discrimination of the fault type of the valve area of the converter station is the Y and D bridge valve short circuit protection, and the difference between the maximum value IVY_MAX of the valve side winding current of the YY converter transformer and the maximum value IVD_MAX of the valve side winding current of the YD converter transformer is 0, it is determined that the valve midpoint to neutral end short circuit fault occurs.

[0033] Further, when the preliminary discrimination of the fault type of the valve area of the converter station is the D bridge valve short circuit protection:

[0034] If the difference between the maximum value IVD_MAX of the valve side winding current of the YD converter transformer and the maximum value IVY_MAX of the valve side winding current of the YY converter transformer shows an upward trend, and the AB phase current periodically returns to 0, it is determined that the valve midpoint to neutral end short circuit fault occurs.

[0035] If the difference between the maximum value IVD_MAX of the valve side winding current of the YD converter transformer and the maximum value IVY_MAX of the valve side winding current of the YY converter transformer shows an upward trend, and the AB direction always flows through a short circuit current, it is determined that the YD valve group AC side inter-phase short circuit fault occurs.

[0036] Compared with the prior art, the present application has the following beneficial effects:

[0037] The application is based on a converter station valve area fault diagnosis method based on SER events and recording wave information, which extracts semantic feature information based on a bidirectional encoder representation model, inputs the mixed neural network, respectively extracts local features and global features for fault matching diagnosis, realizes preliminary screening of fault range, and improves the intelligent level of fault positioning; combined with electrical recording wave analog quantity, mechanism analysis is carried out, key electrical recording wave analog quantity is extracted, further accurate converter station valve area fault type is extracted, fault diagnosis evidence information is obtained step by step according to SER events and recording wave information, and a progressive diagnosis strategy is adopted, compared with the existing converter station valve area fault which can only be positioned to a certain range, the application can realize step-by-step convergence from preliminary fault screening to accurate positioning, and finally realize accurate positioning of the converter station valve area fault point. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 The flowchart of the converter station valve area fault diagnosis method based on SER events and recording wave information of the application is shown in the figure.

[0039] Figure 2 The schematic diagram of fault SER data dimension reduction processing in the application is shown in the figure.

[0040] Figure 3 The converter station partial fault SER corpus of the converter station valve area fault diagnosis method based on SER events and recording wave information of the application is shown in the figure.

[0041] Figure 4 The equivalent circuit schematic diagram of YY valve group AC side A phase ground fault is shown in the figure.

[0042] Figure 5 The converter station valve area fault setting topology graph of the converter station valve area fault diagnosis method based on SER events and recording wave information of the application is shown in the figure. DETAILED DESCRIPTION

[0043] The technical solutions of the application will be further explained and described below in combination with the drawings.

[0044] The protection strategy of high voltage direct current transmission system is usually configured according to zones, including converter protection zone, pole protection zone, bipolar protection zone, converter transformer protection zone, AC filter protection zone and converter bus protection zone. The performance forms of fault characteristics are significantly different for different protection zones or fault types, and the corresponding electrical quantities, switch quantities and SER event information are different. Therefore, in order to realize accurate positioning of faults, the mechanism analysis of various faults needs to be comprehensively analyzed, combined with SER event and recording wave information, etc. to improve the accuracy of fault identification and positioning accuracy. Therefore, it is particularly important to study a valve area fault diagnosis method of converter station based on SER event and recording wave information. A hierarchical fault diagnosis system is constructed: first, the semantic feature information of SER data is extracted and input into a hybrid neural network for fault diagnosis to realize the first layer of valve area fault range screening, and then the key features of fault recording wave data are deeply mined to realize accurate identification of fault points. The invention adopts a progressive diagnosis strategy to gradually converge to the fault point, and finally realizes accurate positioning of the valve area fault of the converter station.

[0045] As Figure 1 The application provides a valve area fault diagnosis method of converter station based on SER event and recording wave information, which comprises the following steps:

[0046] Step S1: based on the RTDS simulation platform, a fault simulation experiment is carried out to obtain fault SER data of the valve area of the converter station, a fault SER corpus is constructed, efficient utilization of the fault SER data of the valve area of the converter station is realized, and effective and reliable SER data support is provided for subsequent fault diagnosis.

[0047] Step S1 comprises the following substeps:

[0048] Step S1.1: based on the RTDS simulation platform, a fault simulation experiment is carried out, high-precision time synchronization technology is used to obtain fault SER data of the valve area of the converter station in real time, and high time resolution and accuracy of fault SER data acquisition are ensured, wherein the fault SER data are as shown in Table 1 and include time, host, system alarm, event level, alarm group and event list. The event list includes valve area trigger event, protection action information, locking / unlocking signal, DC side overvoltage / overcurrent alarm, etc. In the process of fault SER data acquisition, unified data interface standards are adopted for control protection systems of different manufacturers to ensure consistency and compatibility of data sources. According to the typical fault mode and protection strategy of the valve area of the converter station, the obtained SER data are processed by dimension reduction to improve the effectiveness of the data and the calculation efficiency of subsequent analysis, including time level such as emergency, alarm and slight, alarm group such as converter valve state, DC bus voltage and current, and event list such as differential current protection, overcurrent protection and commutation failure detection. By Figure 2The dimensionality reduction preprocessing result shows that each original SER data sample set contains 100-300 events, and after event dimensionality reduction preprocessing, all sample sets retain about 50 key events that can be used for fault diagnosis and contain important semantic information, which also verifies the necessity of SER data dimensionality reduction.

[0049] Step S1.2: Standardizing the dimensionality reduction processed SER data, including: removing duplicate records, filling missing values, and unifying data formats; wherein removing duplicate records eliminates redundant or duplicate reported SER data through event timestamp comparison, improving data consistency; filling missing values uses a pattern matching method based on historical data to reasonably fill in missing alarm or event information, avoiding misdiagnosis caused by data missing; unifying data formats includes unifying SER data field naming methods, standardizing time format and event type coding, realizing data format standardization, and ensuring data traceability and compatibility.

[0050] Step S1.3: According to the alarm level of fault SER data, a fault event classification model of the valve area of the converter station is constructed, and the standardized SER data is classified through the fault event classification model.

[0051] Step S1.4: The Apriori algorithm is used to mine the association rules of the classified SER data, and the minimum support is set to 0.05, the minimum confidence is set to 0.5, and the minimum lift is set to 1. The SER data that meets the minimum support, minimum confidence and minimum lift requirements is saved to the fault SER corpus, and part of the association rule mining results are shown in Table 2.

[0052] Table 1: Fault SER data

[0053] Number Data feature dimension Feature dimension content 1 Time Event record generated to the millisecond 2 Host Control protection device server generating event 3 System alarm Redundant system (A / B / C) employed by SER 4 Event level Operating state level at which event was generated 5 Alarm group Control device or software generating event 6 Event list Content description of event generated

[0054] Table 2: Part of the association rule mining

[0055] Rule antecedent Rule consequent Support Confidence Lift {Converter differential protection 2nd stage action} {Alarm, converter} 0.064 0.6 4.71 {Converter differential protection 2nd stage action} {Converter} 0.106 1 4.71 {Alarm, converter} {Phase failure detected} 0.064 0.5 7.83 {Emergency} {D-bridge valve short circuit protection action} 0.149 0.875 1.96 {Lockout sequence} {Alarm} 0.085 0.571 1.03 {D-bridge valve short circuit protection action} {Converter} 0.074 1 5.4 {Emergency} {CCP PAM pole isolation command present} 0.185 0.625 1.05

[0056] Combined with the information in the above association rule mining table, the SER text data is integrated, the non-fault related information is removed, the high-value fault feature description is extracted, and finally the fault SER corpus of the valve area of the converter station is constructed, as shown in Table 3, which provides data support for fault diagnosis of the valve area of the converter station and improves the intelligent level of fault diagnosis and alarm reasoning. Figure 3

[0057] Step S2: Inputting the SER data in the fault SER corpus into the bidirectional encoder representation model to extract semantic feature information; including the following sub-steps:

[0058] ​Step S2.1: Tokenize the SER data in the fault SER corpus using the WordPiece tokenizer to ensure that power industry terms such as "differential protection II section" and "valve short circuit protection" are correctly parsed. The SER domain dictionary and corresponding tokenization results are shown in Table 3. Generate a token sequence based on the text tokenization results;

[0059] Table 3: SER domain dictionary and corresponding tokenization results

[0060]

[0061] Step S2.2: Add [CLS] and [SEP] markers to the token sequence and perform embedding representation on the token sequence to facilitate the correct parsing of sentence boundaries by the Transformer structure. Input the token sequence into the bidirectional encoder representation model to extract the context semantic vector of each token.

[0062] Step S2.3: Convert the extracted context semantic vector of the token into a high-dimensional embedding vector as the semantic feature information of the SER data.

[0063] Step S3: Input the extracted semantic feature information into the hybrid neural network for analysis and diagnosis to preliminarily determine the fault type of the valve area of the converter station. Specifically, a hybrid neural network is constructed, including a convolutional neural network, a bidirectional long short-term memory network, and an attention module. The convolutional neural network uses a one-dimensional convolutional network structure to extract local features from the extracted semantic feature information using different size convolutional kernels, identify high correlation event patterns in a short time window, reduce the dimension through pooling operations to improve computational efficiency and enhance the robustness of local features, and thus predict the first fault type probability of the valve area of the converter station. The bidirectional long short-term memory network uses a forward LSTM layer and a backward LSTM layer in parallel to capture the forward and backward dependencies of the SER sequence based on the extracted semantic feature information, model the time series of the fault development process, and predict the second fault type probability of the valve area of the converter station. The attention module uses a dot product attention mechanism to weight and fuse the predicted first fault type probability and second fault type probability of the valve area of the converter station to obtain the preliminary fault type of the valve area of the converter station, so that key features occupy a higher proportion in the final decision and improve the accuracy of fault classification. The preliminary fault type of the valve area of the converter station includes: converter differential protection and pole differential protection, Y-bridge valve short circuit protection, Y and D-bridge valve short circuit protection, and D-bridge valve short circuit protection.

[0064] In the present application, the attention module uses a dot product attention mechanism to weight and fuse the local features and global features as follows:

[0065]

[0066] Wherein, Q is the fault diagnosis probability of the converter station valve area output by the attention module, K is a feature identifier of the fault diagnosis probability of the converter station valve area, V is the predicted first fault type probability and second fault type probability of the converter station valve area, d k is the dimension of K, and a scaling factor is introduced The gradient vanishing or explosion problem caused by the excessively large dot product result in the high-dimensional space is effectively avoided.

[0067] In one technical solution of the present application, the correspondence between the typical fault type and the fault SER data is established through the feature information expert library and the historical fault cases, and part of the mapping relationship is shown in Table 4, which is used to train the hybrid neural network, so as to improve the fault type prediction performance of the hybrid neural network.

[0068] Table 4: Correspondence between typical fault type and fault SER data

[0069]

[0070] Step S4: Based on the RTDS simulation platform, a fault simulation experiment is carried out to obtain electrical wave simulation quantities, mechanism analysis is carried out, key electrical wave simulation quantities are extracted, and the fault type of the converter station valve area is further accurately determined.

[0071] Mechanism analysis of valve high-end grounding fault: IDC1P measurement point is in the short-circuit current loop, so the IDC1P measurement value increases greatly, the fault point forms a loop with the grounding pole, the double-bridge loses alternating voltage, the corresponding alternating current is 0, and the IVY_MAX, IVD_MAX and IDC1N measurement points are not in the loop, and the corresponding measurement values are all 0. IDC1P = IVY_MAX when the system is normally running, IDC1P increases greatly after the fault occurs, and IVY_MAX is 0, so the great increase of IDC1P-IVY_MAX value can reflect the occurrence of valve high-end grounding.

[0072] Mechanism analysis of valve midpoint grounding fault: IVY_MAX = IVD_MAX when the system is normally running, IVY_MAX increases greatly after the valve midpoint grounding fault occurs, and IVD_MAX is 0, so the great increase of IVY_MAX-IVD_MAX value can reflect the occurrence of valve midpoint grounding.

[0073] Mechanism analysis of neutral grounding: when normally running, IVD_MAX = IDC1N, IVD_MAX increases greatly after the neutral grounding fault occurs, and IDC1N is 0, so the great increase of IVD_MAX-IDC1N value can reflect the occurrence of valve neutral grounding.

[0074] Mechanism analysis of YY valve group alternating current side single-phase grounding fault: as shown in Figure 4Taking the YY valve group AC side A phase ground fault as an example, after the fault, the thyristor V4 in the three current channels of the YY valve group AC high voltage side is triggered, and the thyristor V4 loses the commutation ability due to the fault current and is continuously turned on, at this time, the three-phase AC measured currents are all 0. The thyristor V6 in the three current channels of the YY valve group AC high voltage side is triggered, and the thyristor V4 is continuously turned on, and since the YY valve group AC side A phase voltage Uya < YY valve group AC side B phase voltage Uyb, the thyristor V6 is successfully turned on, at this time, the AB phases have opposite currents, that is, the sum of the YY valve group AC side A phase current and the B phase current is 0, and the YY valve group AC side C phase current is 0. The thyristor V2 in the three current channels of the YY valve group AC high voltage side is triggered, and the thyristor V4 is continuously turned on, and then Uya < Uyc causes the thyristor V2 to fail to be turned on, at this time, the three-phase currents are all 0. It can be seen that when the A phase ground fault occurs at the YY valve group AC side, the IVY three-phase AC current corresponding to the measuring point will flow through the fault current, and IVY_MAX will increase significantly after the fault, and this fault will have the feature that IVY_MAX-IVD_MAX increases significantly as in the case of the converter neutral point ground. The biggest difference between the YY valve group AC side A phase ground and the valve neutral point ground is that no matter how the thyristors V2, V4 and V6 of the Y bridge common anode group are turned on, the value of the C phase current IVY_L3 is always 0. Therefore, if IVY_MAX-IVD_MAX increases significantly after the fault, and IVY_L3 is always 0, it can be judged that the YY converter A phase ground fault occurs. It can be further known that if IVY_MAX-IVD_MAX increases significantly after the fault, and one of the Y bridge AC side currents is always 0, it can be judged that the YY valve group AC single-phase ground fault occurs.

[0075] The valve high end to neutral point short circuit and the neutral point to neutral end short circuit fault mechanism are analyzed: when the high end to neutral point short circuit occurs, the Y bridge being turned on valve and the short circuit point form a loop, resulting in the phase-to-phase short circuit of the converter transformer, and since the Y bridge loop current is composed of the normal operating current and the fault current, the Y bridge converter transformer valve side current increases sharply, and the D bridge converter transformer valve side current is less affected. Therefore, the significant increase of IVY_MAX-IVD_MAX can reflect the occurrence of the high end to neutral point short circuit, which is similar to the valve neutral point ground, but the protection actions are different, and the two can be accurately distinguished, and the significant increase of IVD_MAX-IVY_MAX can reflect the occurrence of the valve neutral point to neutral end short circuit.

[0076] Theoretical analysis on YD valve group AC side phase-to-phase short circuit fault: Taking YD valve group AC side AB phase-to-phase short circuit as an example for analysis, the fault characteristics are very similar to the midpoint-to-neutral short circuit. Both are D-bridge valve short circuit protection actions, both have the characteristics of a large increase in IVD_MAX-IVY_MAX after the fault, and are easy to misjudge. After analysis, it is found that the biggest difference between the two is that the fault current loop of YD converter transformer interphase short circuit does not contain the thyristor group of D-bridge, while the fault current loop of converter midpoint-to-neutral short circuit contains the thyristor group of D-bridge. For YD valve group AC side phase-to-phase short circuit, the measurement values of YD valve group AC side A phase current IVD_L1 and YD valve group AC side B phase current IVD_L2 measurement points are less affected by the thyristor conduction, that is, the thyristor on the current channel of the low-voltage side of the YD valve group does not conduct or fails to conduct, which will not cause the measurement value of the IVD_L1 measurement point to be 0. For valve midpoint-to-neutral short circuit, the thyristor on the current channel of the low-voltage side of the YD valve group will be turned off or fail to conduct, which will cause the measurement value of the IVY_L1 measurement point to be 0 within a certain period. Therefore, the core feature to distinguish the two is that when YD valve group AC side phase-to-phase short circuit occurs, the current measurement points corresponding to AB phase can always measure the short circuit current and will not periodically become 0.

[0077] According to the topological characteristics and operating principles of the converter and the converter transformer, the following main fault types are sorted out: valve high-end grounding fault, valve midpoint grounding fault, valve neutral grounding fault, valve high-end-to midpoint short circuit fault, valve midpoint-to-neutral short circuit fault, valve side short circuit fault, YY valve group AC side single-phase grounding, and YD valve group AC side phase-to-phase short circuit. The occurrence of these faults will directly affect the current and voltage distribution of the converter station, producing significant recording wave characteristics, as shown in Table 1. Figure 5

[0078] Combined with mechanism analysis, the following key electrical quantities are selected for feature extraction for different fault types. Specifically, high-end converter high-voltage side current IDC1P, high-end converter low-voltage side current IDC1N, YY converter transformer valve side winding current maximum IVY_MAX, YD converter transformer valve side winding current maximum IVD_MAX, Y-bridge AC side phase current, and AB phase-to-phase current. The above electrical quantities constitute the main recording wave characteristics of the valve area fault diagnosis of the converter station, providing data support for subsequent layered diagnosis. As shown in Table 5.

[0079] When the preliminary judgment of the fault type of the valve area of the converter station is the converter differential protection and pole differential protection:

[0080] If the difference between the high-end converter high-voltage side current IDC1P and the YY converter transformer valve side winding current maximum IVY_MAX shows an upward trend, it is determined to be a valve high-end grounding fault;

[0081] ​If the difference between the maximum valve-side winding current of the YD converter transformer IVD_MAX and the low-voltage side current of the high-end converter IDC1N shows an upward trend, it is determined that the valve neutral end is grounded;

[0082] If the difference between the maximum valve-side winding current of the YY converter transformer IVY_MAX and the maximum valve-side winding current of the YD converter transformer IVD_MAX shows an upward trend, and the three-phase current of the Y bridge AC side is not always 0, it is determined that the valve midpoint is grounded;

[0083] If the difference between the maximum valve-side winding current of the YY converter transformer IVY_MAX and the maximum valve-side winding current of the YD converter transformer IVD_MAX shows an upward trend, and one phase current of the Y bridge AC side is always 0, it is determined that the YY valve group AC side is unidirectionally grounded.

[0084] When the preliminary determination of the fault type of the valve area of the converter station is the Y bridge valve short circuit protection, and the difference between the maximum valve-side winding current of the YY converter transformer IVY_MAX and the maximum valve-side winding current of the YD converter transformer IVD_MAX shows an upward trend, it is determined that the valve high end is short-circuited to the midpoint.

[0085] When the preliminary determination of the fault type of the valve area of the converter station is the Y and D bridge valve short circuit protection, and the difference between the maximum valve-side winding current of the YY converter transformer IVY_MAX and the maximum valve-side winding current of the YD converter transformer IVD_MAX is 0, it is determined that the valve midpoint is short-circuited to the neutral end.

[0086] When the preliminary determination of the fault type of the valve area of the converter station is the D bridge valve short circuit protection:

[0087] If the difference between the maximum valve-side winding current of the YD converter transformer IVD_MAX and the maximum valve-side winding current of the YY converter transformer IVY_MAX shows an upward trend, and the AB phase current periodically returns to 0, it is determined that the valve midpoint is short-circuited to the neutral end.

[0088] If the difference between the maximum valve-side winding current of the YD converter transformer IVD_MAX and the maximum valve-side winding current of the YY converter transformer IVY_MAX shows an upward trend, and the AB direction always flows a short-circuit current, it is determined that the YD valve group AC side is phase-to-phase short-circuited.

[0089] Table 5: Correspondence between fault type of valve area of converter station, fault SER data and key electrical wave simulation quantity

[0090]

[0091]

[0092] The application is based on a converter station valve area fault diagnosis method based on SER events and recording wave information, which extracts semantic feature information based on a bidirectional encoder representation model, inputs the mixed neural network, respectively extracts local features and global features for fault matching diagnosis, realizes preliminary screening of the fault range, and improves the intelligent level of fault positioning; combined with electrical recording wave analog quantity, mechanism analysis is carried out, key electrical recording wave analog quantity is extracted, further accurate converter station valve area fault type is extracted, fault diagnosis evidence information is obtained step by step according to SER events and recording wave information, and a progressive diagnosis strategy is adopted, compared with the existing converter station valve area fault which can only be positioned in a certain range, the application can realize step-by-step convergence from preliminary fault screening to accurate positioning, and finally realize accurate positioning of the fault point of the converter station valve area.

[0093] In one technical scheme of the application, a converter station valve area fault diagnosis device based on SER events and recording wave information is also provided, comprising:

[0094] The step-by-step hierarchical diagnosis module adopts data acquisition, signal processing, feature extraction, rule setting and matching analysis methods to build a hierarchical progressive fault diagnosis mechanism based on the decision tree theory. By establishing a fault feature model, the real-time monitoring data are matched step by step, and the hierarchical diagnosis is realized, so that the fault range is gradually narrowed until the fault point is accurately positioned;

[0095] The real-time data acquisition module is used for real-time detection, acquisition and synchronization of converter station sequential event record (SER) data and recording wave information, and converts these fault information into digital signals for subsequent processing and analysis;

[0096] The data preprocessing module screens the SER information of the core dimension, simultaneously performs data cleaning and format standardization, integrates evidence information, and builds a fault SER corpus;

[0097] The fault diagnosis module extracts SER semantic features and multi-dimensional recording wave trend features based on SER events and recording wave information, combined with fault mechanism and expert experience, and performs step-by-step hierarchical identification from fault partition to fault point, accurately judges the fault type, reason and severity, and generates a fault diagnosis report;

[0098] The decision and alarm module triggers the alarm mechanism based on the decision result of the fault diagnosis module, and timely pushes the alarm information to the operation and maintenance personnel. The alarm content includes fault type, reason and severity information, and provides corresponding processing suggestions, or automatically executes the preset emergency response strategy in necessary cases, so as to reduce the accident influence and improve the system recovery efficiency;

[0099] The human-computer interaction interface is equipped with a visual operation interface and an information display module, so that an operation and maintenance personnel can view monitoring data, fault diagnosis results and historical operation records in real time. The interface supports manual debugging and key parameter setting, so as to facilitate optimization of protection strategies, and improve the convenience of system maintenance and the efficiency of fault processing.

[0100] In the present application, a computer device is provided, comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above-mentioned converter station valve area fault diagnosis method based on SER events and recorded wave information when executing the computer program.

[0101] In the present application, a computer readable storage medium is provided, which stores a computer program, and the computer program implements the steps of the above-mentioned converter station valve area fault diagnosis method based on SER events and recorded wave information when executed by a processor.

[0102] In the present application, a computer program product is provided, comprising a computer program, and the computer program implements the steps of the above-mentioned converter station valve area fault diagnosis method based on SER events and recorded wave information when executed by a processor.

[0103] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties.

[0104] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0105] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solution falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the present application shall be considered as falling within the protection scope of the present application.

Claims

1. A converter station valve area fault diagnosis method based on SER events and recorded information, characterized in that: The steps include: Step S1: Conduct a fault simulation experiment based on the RTDS simulation platform to obtain fault SER data of the converter station valve area and construct a fault SER corpus; Step S2: Input the SER data in the fault SER corpus into the bidirectional encoder representation model to extract semantic feature information; Step S3: Inputting the extracted semantic feature information into the hybrid neural network for analysis and diagnosis, and preliminarily determining the fault type of the converter station valve area; Step S4: Conduct a fault simulation experiment based on the RTDS simulation platform to obtain electrical waveform simulation quantities, conduct mechanism analysis, extract key electrical waveform simulation quantities, and further accurately determine the fault type in the converter station valve area.

2. The converter station valve area fault diagnosis method based on SER events and recorded information according to claim 1 is characterized in that: Step S1 includes the following sub-steps: Step S1.1: Conduct a fault simulation experiment based on the RTDS simulation platform to obtain fault SER data of the converter station valve area. Perform dimensionality reduction processing on the obtained SER data based on the typical fault modes and protection strategies of the converter station valve area. Step S1.2: Standardize the SER data after dimension reduction, including removing duplicate records, filling missing values, and unifying the data format; Step S1.3: Based on the alarm level of the fault SER data, a fault event classification model for the converter station valve area is constructed, and the standardized SER data is classified using the fault event classification model; Step S1.4: Use the Apriori algorithm to mine association rules on the hierarchical SER data, set the minimum support, minimum confidence, and minimum lift, and save the SER data that meets the minimum support, minimum confidence, and minimum lift requirements into the fault SER corpus.

3. The converter station valve area fault diagnosis method based on SER events and recorded information according to claim 1 is characterized in that: Step S2 includes the following sub-steps: Step S2.1: Use the WordPiece word segmenter to segment the SER data in the fault SER corpus and generate a token sequence; Step S2.2: Add [CLS] and [SEP] tags to the token sequence, embed the token sequence, input it into the bidirectional encoder representation model, and extract the contextual semantic vector of each token; Step S2.3: Convert the contextual semantic vector of the extracted token into a high-dimensional embedding vector as the semantic feature information of the SER data.

4. The converter station valve area fault diagnosis method based on SER events and recorded information according to claim 1 is characterized in that: The specific process of step S3 is as follows: constructing a hybrid neural network, including: a convolutional neural network, a bidirectional long short-term memory network and an attention module; wherein the convolutional neural network predicts the probability of the first fault type of the converter station valve area based on the extracted semantic feature information, and the bidirectional long short-term memory network adopts a forward LSTM layer and a backward LSTM layer in parallel to predict the probability of the second fault type of the converter station valve area based on the extracted semantic feature information; the attention module adopts a dot product attention mechanism to perform weighted fusion on the predicted probability of the first fault type and the second fault type of the converter station valve area to obtain a preliminary fault type of the converter station valve area.

5. The converter station valve area fault diagnosis method based on SER events and recorded information according to claim 4 is characterized in that: The process of weighted fusion of local features and global features by the attention module using the dot product attention mechanism is as follows: Where Q is the fault diagnosis probability of the converter station valve area output by the attention module, K is the characteristic identifier of the fault diagnosis probability of the converter station valve area, V is the predicted probability of the first fault type and the second fault type of the converter station valve area, d k is the dimension of K.

6. The converter station valve area fault diagnosis method based on SER events and recorded information according to claim 1, characterized in that: In step S3, the fault types of the converter station valve area are preliminarily determined to include: converter differential protection and pole differential protection, Y bridge valve short circuit protection, Y and D bridge valve short circuit protection, and D bridge valve short circuit protection; The key electrical waveform recording analog quantities in step S4 include: the high-side current IDC1P of the high-end converter, the low-side current IDC1N of the high-end converter, the maximum current IVY_MAX of the valve-side winding of the YY converter transformer, the maximum current IVD_MAX of the valve-side winding of the YD converter transformer, the AC side phase current of the Y bridge, and the AB phase current.

7. The converter station valve area fault diagnosis method based on SER events and recorded information according to claim 6, characterized in that: When the fault type in the converter station valve area is preliminarily determined to be converter differential protection and pole differential protection: If the difference between the high-side current IDC1P of the high-side converter and the maximum current IVY_MAX of the valve-side winding of the YY converter transformer shows an upward trend, it is determined that there is a valve high-side grounding fault; If the difference between the maximum current IVD_MAX of the YD converter transformer valve-side winding and the current IDC1N of the high-side converter low-voltage side shows an upward trend, it is determined that the valve neutral terminal is grounded; If the difference between the maximum current IVY_MAX of the valve-side winding of the YY converter transformer and the maximum current IVD_MAX of the valve-side winding of the YD converter transformer shows an upward trend, and the three-phase currents on the AC side of the Y bridge are not always 0, it is determined that there is a valve midpoint grounding fault; If the difference between the maximum current IVY_MAX of the valve-side winding of the YY converter transformer and the maximum current IVD_MAX of the valve-side winding of the YD converter transformer shows an upward trend, and one phase current on the AC side of the Y bridge is always 0, it is determined to be a unidirectional grounding fault on the AC side of the YY valve group.

8. The converter station valve area fault diagnosis method based on SER events and recorded information according to claim 6 is characterized in that: When the fault type in the converter station valve area is preliminarily determined to be Y-bridge valve short-circuit protection, and the difference between the maximum current IVY_MAX of the valve-side winding of the YY converter transformer and the maximum current IVD_MAX of the valve-side winding of the YD converter transformer shows an upward trend, it is determined to be a valve high-end to midpoint short-circuit fault.

9. The converter station valve area fault diagnosis method based on SER events and recorded information according to claim 6, characterized in that: When the fault type of the converter station valve area is preliminarily determined to be the Y and D bridge valve short-circuit protection, and the difference between the maximum current IVY_MAX of the valve-side winding of the YY converter transformer and the maximum current IVD_MAX of the valve-side winding of the YD converter transformer is 0, it is determined to be a valve midpoint to neutral short-circuit fault.

10. The converter station valve area fault diagnosis method based on SER events and recorded information according to claim 6, characterized in that: When the fault type of the converter station valve area is initially determined to be D bridge valve short-circuit protection: If the difference between the maximum current IVD_MAX of the valve-side winding of the YD converter transformer and the maximum current IVY_MAX of the valve-side winding of the YY converter transformer shows an upward trend, and the current between the AB phases returns to 0 periodically, it is determined that a valve midpoint to neutral short circuit fault has occurred; If the difference between the maximum current IVD_MAX of the valve-side winding of the YD converter transformer and the maximum current IVY_MAX of the valve-side winding of the YY converter transformer shows an upward trend, and a short-circuit current always flows in the AB direction, it is determined that there is a phase-to-phase short-circuit fault on the AC side of the YD valve group.

Images