Transformer fault diagnosis method based on online gas monitoring
By monitoring and analyzing the changes in light gas volume value, capacitive sensors and neural network technology are used to identify internal faults of the transformer, solving the problems of false protection and data gaps in the existing technology, and improving the accuracy and safety of transformer fault diagnosis .
Patent Information
- Application Number
- CN202410321351.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-20
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-03-20
AI Technical Summary
The existing technology is difficult to effectively monitor and diagnose minor faults inside the transformer, resulting in false protection and frequent alarms, increasing the burden on on-site operation inspection personnel, and there are data vacancy problems, which may cause safety hazards.
By monitoring the change of light gas volume value, the data collected by the capacitive sensor is used to perform data regularization, feature extraction and semantic discrete degree quantification, combining the one-dimensional convolutional neural network model and population separation measurement module to determine whether to initiate light gas action, and Quantitative analysis of gas components is carried out to generate a transformer fault alarm.
Real-time monitoring and fault diagnosis of internal transformers is realized, reducing false protection and alarm frequency, reducing data vacancy and improving safety.
Smart Images

Figure CN118243747B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of transformer fault diagnosis, and more specifically, to a transformer fault diagnosis method based on online monitoring of gas. Background Art
[0002] With the increasingly complex power grid, higher requirements are placed on the adaptability of transformer relay protection. At present, current differential protection and gas protection are mostly used as the main protection of transformers. Among them, the main protection of transformer power - current differential protection, only introduces the current of the transformer to reflect the operating condition of the transformer, which is difficult to fully describe the operating condition of the transformer.
[0003] When low-energy faults such as minor turn-to-turn faults, partial discharge, and non-electrical faults such as core heating occur inside the transformer, the gas generated in the oil causes oil flow surges and pressure increases. The non-electrical quantity protection formed based on fault characteristics such as gas, oil flow, and pressure has higher sensitivity to internal weak faults than electrical quantity protection.
[0004] When the transformer is subjected to external short circuit shock, oil pump start and stop, asymmetric start and stop of the cooler, drastic temperature changes and other external influences, it may cause non-internal faults and false operation of non-electrical protection. Frequent alarm signals and false operation of protection devices will increase the burden on on-site operation and maintenance personnel and even endanger personal safety.
[0005] The traditional way to collect transformer non-electric data is to read meters and conduct inspections on site regularly. During the peak summer season or major events, more frequent inspections are required to check the gas volume and the traces of the pressure relief valve. However, the changes in the light gas volume and other parameter information of the transformer body during the interval between meter reading and inspection are unknown, which may easily lead to data gaps during this period. In addition, this method may cause the transformer to be in an abnormal operating state, causing more serious safety hazards.
[0006] Therefore, a transformer fault diagnosis method based on online monitoring of gas is expected. Summary of the invention
[0007] In order to solve the above technical problems, the present application is proposed. The present application provides a transformer fault diagnosis method based on online gas monitoring, which can identify possible abnormal conditions inside the transformer and realize monitoring and fault diagnosis of the internal conditions of the transformer.
[0008] According to one aspect of the present application, a transformer fault diagnosis method based on online gas monitoring is provided, which includes:
[0009] Monitor the changes in light gas volume value;
[0010] Determining whether to start a light gas action based on the change in the light gas volume value;
[0011] In response to starting the light gas action, performing a quantitative component analysis on the gas to obtain an analysis result; and
[0012] Based on the analysis result, it is determined whether to generate a transformer fault alarm.
[0013] In the above transformer fault diagnosis method based on online gas monitoring, determining whether to start the light gas action based on the change of the light gas volume value includes:
[0014] Acquire a time series of capacitance values collected by a capacitance sensor;
[0015] Performing data regularization on the time series of the capacitance values to obtain a sequence of local capacitance timing input vectors;
[0016] Performing capacitance timing feature extraction on the sequence of local capacitance timing input vectors to obtain a sequence of local capacitance timing feature vectors;
[0017] quantizing the degree of capacitance semantic discreteness of the sequence of the local capacitance time series feature vectors to obtain a capacitance semantic discrete feature vector; and
[0018] Based on the capacitance semantic discrete feature vector, it is determined whether to start the light gas action.
[0019] In the above transformer fault diagnosis method based on online gas monitoring, data regularization is performed on the time series of the capacitance value to obtain a sequence of local capacitance time series input vectors, including:
[0020] Arranging the time series of the capacitance values into a capacitance time series input vector according to the time dimension; and
[0021] The capacitance timing input vector is vector-segmented to obtain a sequence of the local capacitance timing input vectors.
[0022] In the above transformer fault diagnosis method based on gas online monitoring, the sequence of local capacitance time series input vectors is subjected to capacitance time series feature extraction to obtain a sequence of local capacitance time series feature vectors, including:
[0023] The sequence of the local capacitance timing input vectors is passed through a capacitance timing feature capture module based on a one-dimensional convolutional neural network model to obtain a sequence of the local capacitance timing feature vectors.
[0024] In the above transformer fault diagnosis method based on online gas monitoring, the sequence of the local capacitance time series feature vector is quantified according to the degree of capacitance semantic discreteness to obtain the capacitance semantic discrete feature vector, including:
[0025] Performing nonlinear response compensation on the sequence of local capacitance timing characteristic vectors to obtain a sequence of compensated local capacitance timing characteristic vectors;
[0026] Using the last compensated local capacitance time series feature vector in the sequence of compensated local capacitance time series feature vectors as a query vector; and
[0027] The query vector and other compensated local capacitance time series feature vectors in the sequence of compensated local capacitance time series feature vectors are respectively input into a population divergence measurement module to obtain the capacitance semantic discrete feature vector composed of multiple capacitance semantic discrete coefficients.
[0028] In the above transformer fault diagnosis method based on online gas monitoring, nonlinear response compensation is performed on the sequence of local capacitance time series characteristic vectors to obtain a sequence of compensated local capacitance time series characteristic vectors, including:
[0029] The local capacitance timing characteristic vector is processed by the following nonlinear response compensation formula to obtain the compensated local capacitance timing characteristic vector; wherein the nonlinear response compensation formula is:
[0030]
[0031] Among them, L 0i is the eigenvalue of the i-th position in the local capacitance time series eigenvector, L 1i is the characteristic value of the i-th position in the compensated local capacitance timing characteristic vector, and A, B, C and D are adjustment parameters with different values.
[0032] In the above-mentioned transformer fault diagnosis method based on gas online monitoring, the query vector and other compensated local capacitance time series feature vectors in the sequence of the compensated local capacitance time series feature vector are respectively input into the population divergence measurement module to obtain the capacitance semantic discrete feature vector composed of multiple capacitance semantic discrete coefficients, including:
[0033] The query vector and other compensated local capacitance time series feature vectors in the sequence of the compensated local capacitance time series feature vector are processed using the following population separation measurement formula to obtain the capacitance semantic discrete feature vector composed of the multiple capacitance semantic discrete coefficients; wherein the population separation measurement formula is:
[0034]
[0035] Among them, e i is the semantic discrete coefficient of the i-th capacitor, v q is the query vector, v i is the i-th compensated local capacitance time series feature vector in the sequence of compensated local capacitance time series feature vectors, W is the weight coefficient matrix, B is the weight coefficient vector, tanh(·) represents the hyperbolic tangent function, (·) T Represents the transpose of a vector.
[0036] In the above transformer fault diagnosis method based on gas online monitoring, determining whether to start the light gas action based on the capacitance semantic discrete feature vector includes:
[0037] The capacitance semantic discrete feature vector is passed through a control decision maker based on a classifier to obtain a decision result, and the decision result is used to indicate whether to start a light gas action.
[0038] In the above-mentioned transformer fault diagnosis method based on online gas monitoring, it also includes a training step: training the capacitor timing feature capture module based on the one-dimensional convolutional neural network model, the population separation measurement module and the classifier-based control decision maker.
[0039] In the above transformer fault diagnosis method based on online gas monitoring, the training step includes:
[0040] Acquire training data, where the training data includes a time series of training capacitance values collected by the capacitance sensor and a real value of whether to start a light gas action;
[0041] Arranging the time series of the training capacitance values into a training capacitance timing input vector according to the time dimension;
[0042] Performing vector segmentation on the training capacitance timing input vector to obtain a sequence of training local capacitance timing input vectors;
[0043] Passing the sequence of the training local capacitance timing input vectors through the capacitance timing feature capture module based on the one-dimensional convolutional neural network model to obtain a sequence of training local capacitance timing feature vectors;
[0044] Performing nonlinear response compensation on the sequence of training local capacitance time series feature vectors to obtain a sequence of training compensated local capacitance time series feature vectors;
[0045] The last trained compensated local capacitance time series feature vector in the sequence of the trained compensated local capacitance time series feature vectors is used as a query vector, and the query vector and other trained compensated local capacitance time series feature vectors in the sequence of the trained compensated local capacitance time series feature vectors are respectively input into the population divergence measurement module to obtain a trained capacitance semantic discrete feature vector composed of a plurality of capacitance semantic discrete coefficients;
[0046] Passing the training capacitor semantic discrete feature vector through the classifier-based control decision maker to obtain a classification loss function value; and
[0047] The capacitor timing feature capture module based on the one-dimensional convolutional neural network model, the population separation measurement module and the classifier-based control decision maker are trained using the classification loss function value, wherein in each round of iteration of the training, the training capacitor semantic discrete feature vector is iteratively optimized.
[0048] Compared with the prior art, the transformer fault diagnosis method based on online monitoring of gas provided by the present application first monitors the change of light gas volume value, then determines whether to start the light gas action based on the change of the light gas volume value, then, in response to starting the light gas action, performs quantitative component analysis on the gas to obtain the analysis result, and finally, based on the analysis result, determines whether to generate a transformer fault alarm. In this way, by collecting and analyzing the capacitance value, the dynamic characteristics of the capacitance value changing over time can be mined, thereby inferring the dynamic change characteristics of the light gas volume value, so as to identify the abnormal conditions that may exist inside the transformer, and realize the monitoring and fault diagnosis of the internal conditions of the transformer. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. The following drawings are not intentionally scaled to the actual size, and the focus is on illustrating the main purpose of the present application.
[0050] Figure 1 The present invention is a flowchart of a transformer fault diagnosis method based on online gas monitoring according to an embodiment of the present application.
[0051] Figure 2 This is a flowchart of sub-step S120 of the transformer fault diagnosis method based on online gas monitoring according to an embodiment of the present application.
[0052] Figure 3 This is a flowchart of sub-step S124 of the transformer fault diagnosis method based on online gas monitoring according to an embodiment of the present application.
[0053] Figure 4It is a block diagram of a transformer fault diagnosis system based on online monitoring of gas according to an embodiment of the present application.
[0054] Figure 5 This is an application scenario diagram of a transformer fault diagnosis method based on online gas monitoring according to an embodiment of the present application. DETAILED DESCRIPTION
[0055] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.
[0056] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.
[0057] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.
[0058] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.
[0059] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.
[0060] The present application provides a transformer fault diagnosis method based on online monitoring of gas. Figure 1 FIG. 1 is a flow chart of a transformer fault diagnosis method based on online gas monitoring according to an embodiment of the present application. Figure 1As shown, according to the transformer fault diagnosis method based on online monitoring of gas according to the embodiment of the present application, the specific steps include: S110, monitoring the change of the light gas volume value; S120, determining whether to start the light gas action based on the change of the light gas volume value; S130, in response to starting the light gas action, performing quantitative analysis of the gas composition to obtain an analysis result; and, S140, determining whether to generate a transformer fault alarm based on the analysis result.
[0061] Among them, in the process of determining whether to start the light gas action based on the change of the light gas volume value, it is considered that it is very difficult to accurately measure the light gas volume value directly. The internal environment of the transformer is affected by many factors, such as oil flow, temperature, pressure, etc. These factors will affect the distribution and concentration of the light gas volume value, increasing the difficulty of direct measurement. However, inside the transformer, the generation and accumulation of light gas will cause changes in the capacitance value, and the change in the capacitance value can indirectly reflect the change in the light gas volume value. The technical concept of the present application is to utilize the above-mentioned characteristics, collect and analyze the capacitance value, and explore the dynamic characteristics of the capacitance value changing over time, so as to infer the dynamic change characteristics of the light gas volume value, so as to identify possible abnormal conditions inside the transformer and realize the monitoring and fault diagnosis of the internal conditions of the transformer.
[0062] Based on this, Figure 2 As shown, based on the change of the light gas volume value, determining whether to start the light gas action includes: S121, acquiring the time series of capacitance values collected by the capacitance sensor; S122, performing data regularization on the time series of capacitance values to obtain a sequence of local capacitance timing input vectors; S123, performing capacitance timing feature extraction on the sequence of local capacitance timing input vectors to obtain a sequence of local capacitance timing feature vectors; S124, performing capacitance semantic discrete degree quantization on the sequence of local capacitance timing feature vectors to obtain a capacitance semantic discrete feature vector; and, S125, determining whether to start the light gas action based on the capacitance semantic discrete feature vector.
[0063] It should be understood that this process involves processing the data collected by the capacitance sensor to determine whether to start the light gas action. Step S121 is the starting point for obtaining the original data, and the time series data of the capacitance value is collected from the capacitance sensor to provide basic data for subsequent processing. In step S122, the original time series data is regularized, which may include data cleaning, interpolation, smoothing and other operations to obtain a local capacitance timing input vector sequence with a certain regularity. Step S123 is to extract features from the local capacitance timing input vector sequence, which will involve extracting statistical features, frequency domain features, time domain features, etc., to obtain a more representative and informative local capacitance timing feature vector sequence. In step S124, the local capacitance timing feature vector sequence is further processed, such as clustering, classification or other forms of semantic analysis of the feature vector to quantify the degree of semantic discreteness of the capacitance, thereby obtaining a capacitance semantic discrete feature vector. Step S125 is to make a decision based on the obtained discrete feature vector of capacitance semantics to determine whether to initiate the light gas action, which involves setting a threshold, classifier or other decision logic to determine whether to take corresponding actions based on the processed feature vector. The purpose of this process is to extract useful feature information from the data collected by the capacitance sensor through a series of processing and analysis to determine whether the light gas action needs to be initiated.
[0064] Specifically, in the technical solution of the present application, first, the time series of capacitance values collected by the capacitance sensor is obtained for monitoring and analyzing the change of capacitance values over time, so as to understand the dynamic change of the system. Then, the time series of the capacitance values is arranged as a capacitance time series input vector according to the time dimension; and the capacitance time series input vector is vector-segmented to obtain a sequence of local capacitance time series input vectors. Here, arranging the time series of capacitance values collected by the capacitance sensor as a capacitance time series input vector according to the time dimension can intuitively show the dynamic change characteristics of the capacitance value in the time dimension. This processing method helps to analyze and understand the change trend of the capacitance value over time and reveal the dynamic characteristics of the system. Then, the capacitance time series input vector is divided into smaller time windows, namely local capacitance time series input vectors, so as to facilitate subsequent sequence modeling. At the same time, the capacitance time series input vector containing a longer time period is divided into shorter sequences, which helps to extract more local and fine-grained features. This can better capture local patterns and changes in the data and improve the expressiveness of the model.
[0065] Correspondingly, in step S122, the time series of the capacitance values is data regularized to obtain a sequence of local capacitance timing input vectors, including: arranging the time series of the capacitance values into capacitance timing input vectors according to the time dimension; and vector segmenting the capacitance timing input vectors to obtain a sequence of local capacitance timing input vectors.
[0066] It should be understood that in step S122, the time series of capacitance values is vectorized to obtain a sequence of local capacitance timing input vectors in order to better process and analyze data. Specifically, the effects of vector segmentation include the following points: 1. Reduce data dimension: Dividing a long sequence into multiple local vectors can reduce the dimension of each vector, simplify the data structure, and make subsequent processing more efficient. 2. Improve computational efficiency: If the entire time series is very long, directly processing the entire sequence may result in excessive computation. By dividing it into local vectors, the data can be processed in stages and blocks to improve computational efficiency. 3. Increase time context information: Dividing it into local vectors can retain the time context information of each local vector, which is helpful for subsequent feature extraction and analysis, and better captures the time correlation in the data. 4. Improve the generalization ability of the model: Dividing it into local vectors helps the model to better learn the local features of the data, avoid overfitting the model to the entire long sequence, and improve the generalization ability of the model. 5. Processing variable-length sequences: For time series data of different lengths, dividing it into local vectors can unify the format of the data, which is convenient for subsequent processing and analysis. In general, vector segmentation helps to simplify data structures, improve computational efficiency, preserve temporal context information, improve model generalization capabilities, and process variable-length sequences, making data processing more efficient and effective.
[0067] Next, the sequence of the local capacitance time series input vector is passed through a capacitance time series feature capture module based on a one-dimensional convolutional neural network model to obtain a sequence of local capacitance time series feature vectors. Among them, the one-dimensional convolutional neural network model is a variant of the convolutional neural network (CNN). Unlike the traditional two-dimensional CNN, the one-dimensional convolutional neural network model is used to process one-dimensional data, such as time series data, univariate parameter data, etc. The one-dimensional convolutional neural network model generally includes components such as a convolution layer, a pooling layer, and an activation function, and extracts features from the data by sliding the convolution kernel on the input one-dimensional data and applying a nonlinear activation function. Here, the capacitance time series feature capture module based on the one-dimensional convolutional neural network model can capture the local patterns and time series features in each of the local capacitance time series input vectors, which is conducive to identifying the short-term time-dimensional dependencies in each of the local capacitance time series input vectors and understanding the variation characteristics of the capacitance value within each local time span.
[0068] Correspondingly, in step S123, capacitance timing feature extraction is performed on the sequence of local capacitance timing input vectors to obtain a sequence of local capacitance timing feature vectors, including: passing the sequence of local capacitance timing input vectors through a capacitance timing feature capture module based on a one-dimensional convolutional neural network model to obtain a sequence of local capacitance timing feature vectors.
[0069] Considering that the distribution of the local time-domain capacitance change pattern expressed in each of the local capacitance timing characteristic vectors is not linearly related to the change in the light gas volume value. Specifically, during the operation of the transformer, the relationship between the capacitance value and the light gas volume value may be affected by many factors, including temperature, pressure, humidity, insulation material conditions, etc. For example, in a transformer, as the temperature changes, the properties of the insulating material may change, resulting in nonlinear changes in the capacitance value. In other words, there may be a nonlinear response relationship between the timing change pattern of the capacitance and the timing change pattern of the light gas volume value. However, this nonlinear response relationship may be easily overlooked or difficult to reflect during the feature extraction process. Therefore, in the technical solution of the present application, the sequence of the local capacitance timing characteristic vectors is further compensated for the nonlinear response to obtain a sequence of the compensated local capacitance timing characteristic vectors.
[0070] Then, the last compensated local capacitance time series feature vector in the sequence of the compensated local capacitance time series feature vector is used as a query vector, and the query vector and the other compensated local capacitance time series feature vectors in the sequence of the compensated local capacitance time series feature vector are respectively input into the population divergence measurement module to obtain a capacitance semantic discrete feature vector composed of multiple capacitance semantic discrete coefficients. That is, the degree of capacitance semantic difference between the last compensated local capacitance time series feature vector and the other compensated local capacitance time series feature vectors in the sequence of the compensated local capacitance time series feature vector is quantified by the population divergence measurement module, reflecting the difference between the capacitance time series features under each local time span. In this way, the capacitance semantic discrete feature vector composed of the multiple capacitance semantic discrete coefficients can reveal whether there are abnormal fluctuations or special patterns in the capacitance value in the entire time dimension, so as to discover potential abnormal patterns in the capacitance time series data.
[0071] Accordingly, in step S124, Figure 3 As shown, the degree of capacitance semantic discreteness of the sequence of local capacitance timing feature vectors is quantified to obtain capacitance semantic discrete feature vectors, including: S1241, nonlinear response compensation is performed on the sequence of local capacitance timing feature vectors to obtain a sequence of compensated local capacitance timing feature vectors; S1242, the last compensated local capacitance timing feature vector in the sequence of compensated local capacitance timing feature vectors is used as a query vector; and, S1243, the query vector and each other compensated local capacitance timing feature vector in the sequence of compensated local capacitance timing feature vectors are respectively input into a population deviation measurement module to obtain the capacitance semantic discrete feature vector composed of multiple capacitance semantic discrete coefficients.
[0072] In one example, in step S1241, performing nonlinear response compensation on the sequence of local capacitance timing characteristic vectors to obtain a sequence of compensated local capacitance timing characteristic vectors includes: processing the local capacitance timing characteristic vectors with the following nonlinear response compensation formula to obtain the compensated local capacitance timing characteristic vectors; wherein the nonlinear response compensation formula is:
[0073]
[0074] Among them, L 0i is the eigenvalue of the i-th position in the local capacitance time series eigenvector, L 1i is the characteristic value of the i-th position in the compensated local capacitance timing characteristic vector, and A, B, C and D are adjustment parameters with different values.
[0075] In one example, in step S1243, the query vector and each other compensated local capacitance timing feature vector in the sequence of the compensated local capacitance timing feature vector are respectively input into a population divergence measurement module to obtain the capacitance semantic discrete feature vector composed of multiple capacitance semantic discrete coefficients, including: processing the query vector and each other compensated local capacitance timing feature vector in the sequence of the compensated local capacitance timing feature vector using the following population divergence measurement formula to obtain the capacitance semantic discrete feature vector composed of the multiple capacitance semantic discrete coefficients; wherein the population divergence measurement formula is:
[0076]
[0077] Among them, e i is the semantic discrete coefficient of the i-th capacitor, v q is the query vector, v i is the i-th compensated local capacitance time series feature vector in the sequence of compensated local capacitance time series feature vectors, W is the weight coefficient matrix, B is the weight coefficient vector, tanh(·) represents the hyperbolic tangent function, (·) T Represents the transpose of a vector.
[0078] Subsequently, the capacitance semantic discrete feature vector is passed through a control decision maker based on a classifier to obtain a decision result, and the decision result is used to indicate whether to start the light gas action. Accordingly, in step S125, based on the capacitance semantic discrete feature vector, determining whether to start the light gas action includes: passing the capacitance semantic discrete feature vector through a control decision maker based on a classifier to obtain a decision result, and the decision result is used to indicate whether to start the light gas action.
[0079] Specifically, the capacitor semantic discrete feature vector is passed through a classifier-based control decision maker to obtain a decision result, and the decision result is used to indicate whether to start the light gas action, including: using the fully connected layer of the classifier-based control decision maker to fully connect encode the capacitor semantic discrete feature vector to obtain an encoded classification feature vector; and inputting the encoded classification feature vector into the Softmax classification function of the classifier-based control decision maker to obtain the decision result.
[0080] It should be understood that the role of the classifier is to use the given categories and known training data to learn classification rules and classifiers, and then classify (or predict) unknown data. Logistic regression and SVM are often used to solve binary classification problems. For multi-class classification problems, logistic regression or SVM can also be used, but multiple binary classifications are required to form a multi-classification. However, this is prone to errors and is not efficient. Commonly used multi-classification methods include the Softmax classification function.
[0081] Further, in the technical solution of the present application, the transformer fault diagnosis method based on gas online monitoring also includes a training step: training the capacitor timing feature capture module based on the one-dimensional convolutional neural network model, the population separation measurement module and the control decision maker based on the classifier. It should be understood that the training step plays an important role in the transformer fault diagnosis method based on gas online monitoring, mainly including training the capacitor timing feature capture module, the population separation measurement module and the control decision maker based on the classifier of the one-dimensional convolutional neural network model. The following are the effects of the training step: 1. Improve model accuracy: Training the capacitor timing feature capture module, the population separation measurement module and the control decision maker based on the classifier of the one-dimensional convolutional neural network model can help the model learn the patterns and features in the data and improve the model's diagnostic accuracy for transformer faults. 2. Optimize parameters: Through parameter optimization during the training process, the weight and bias of the model can be adjusted to make it better adapt to actual data and improve the performance and generalization ability of the model. 3. Reduce the risk of overfitting: Training the model can help reduce the risk of overfitting by learning the patterns and features of the data on the training set, making the model more generalizable on new data. 4. Enhance model stability: Trained models are usually more stable and can better cope with the needs of transformer fault diagnosis under different data conditions. 5. Optimize the decision process: Training the classifier-based control decision maker can optimize the decision process, making it more targeted and accurate, thereby better guiding the start of light gas action. 6. Improve system performance: The training step helps to improve the performance of the entire transformer fault diagnosis system based on gas online monitoring, so that it can more accurately and stably diagnose transformer faults and take corresponding measures in practical applications. In summary, the training step plays a vital role in model accuracy, parameter optimization, reducing overfitting risks, enhancing stability, optimizing the decision process, and improving overall system performance.
[0082] In one example, the training step includes: acquiring training data, the training data including a time series of training capacitance values collected by a capacitance sensor, and a real value of whether a light gas action is started; arranging the time series of training capacitance values into a training capacitance timing input vector according to a time dimension; vector segmenting the training capacitance timing input vector to obtain a sequence of training local capacitance timing input vectors; passing the sequence of training local capacitance timing input vectors through the capacitance timing feature capture module based on the one-dimensional convolutional neural network model to obtain a sequence of training local capacitance timing feature vectors; performing nonlinear response compensation on the sequence of training local capacitance timing feature vectors to obtain a sequence of training compensated local capacitance timing feature vectors; and transmitting the training compensated local capacitance timing feature vectors to the capacitor timing feature capture module based on the one-dimensional convolutional neural network model. The last trained compensated local capacitance time series feature vector in the sequence of the trained compensated local capacitance time series feature vectors is used as a query vector, and the query vector and other trained compensated local capacitance time series feature vectors in the sequence of the trained compensated local capacitance time series feature vectors are respectively input into the population divergence measurement module to obtain a training capacitance semantic discrete feature vector composed of multiple capacitance semantic discrete coefficients; the training capacitance semantic discrete feature vector is passed through the classifier-based control decision maker to obtain a classification loss function value; and the classification loss function value is used to train the capacitance timing feature capture module based on the one-dimensional convolutional neural network model, the population divergence measurement module and the classifier-based control decision maker, wherein, in each round of iteration of the training, the training capacitance semantic discrete feature vector is iteratively optimized.
[0083] In the above technical scheme, the sequence of the trained compensated local capacitance timing feature vectors expresses the timing correlation characteristics of the training capacitance value in the local time domain after correction by nonlinear timing response. Thus, the last trained compensated local capacitance timing feature vector in the sequence of the trained compensated local capacitance timing feature vectors is used as a query vector, and the query vector and each other trained compensated local capacitance timing feature vector in the sequence of the trained compensated local capacitance timing feature vectors are respectively input into the population divergence measurement module. Then, the training capacitance semantic discrete feature vector composed of multiple capacitance semantic discrete coefficients can be understood as the semantic discrete expression of the local time domain timing feature in the global time domain.
[0084] However, considering that the training capacitance value itself has an uneven distribution of local time domain features in the global time domain, and since the nonlinear time response correction emphasizes the main trend of the distribution of time-related features in each training local time domain, it will cause significant differences in the eigenvalue-by-eigenvalue feature distribution of the semantic discrete representation of the time-series features of each local time domain in the global time domain, thereby causing inconsistency in the local feature distribution of the training capacitance semantic discrete feature vector based on the eigenvalue numerical distribution, affecting the convergence speed of the training capacitance semantic discrete feature vector through the class regression iteration of the classifier.
[0085] Therefore, the applicant of the present application iteratively optimizes the training capacitor semantic discrete feature vector each time the training capacitor semantic discrete feature vector is iteratively classified and regressed by the classifier. Accordingly, in one example, in each round of iteration of the training, the training capacitor semantic discrete feature vector is iteratively optimized using the following optimization formula to obtain the optimized training capacitor semantic discrete feature vector; wherein the optimization formula is:
[0086]
[0087] Among them, v i is the i-th eigenvalue of the training capacitor semantic discrete feature vector, φ(v i ) represents the probabilistic function of the eigenvalue, that is, the eigenvalue v i is a probabilistic function mapped to the interval [0,1], p is the class probability value obtained by the classifier, L is the length of the training capacitor semantic discrete feature vector, and α is a weight hyperparameter, exp[·] represents the exponential operation of the value, and the exponential operation of the value represents the calculation of the natural exponential function value with the value as the power, v ′i is the i-th eigenvalue of the optimized training capacitor semantic discrete feature vector.
[0088] Here, the training is performed by taking the Bernoulli probability likelihood distribution as the authoritative meta-entity and the feature value v of the capacitor semantic discrete feature vector i Based on the high-level class decision of Bernoulli probability distribution (p,1-p), and introducing the probability density space mapping allocation relative to the whole feature set, the eigenvalue v of the training capacitor semantic discrete feature vector is determined. i The key insight with respect to the class probability mapping is to implement the eigenvalues v of the training capacitive semantic discrete feature vector i Based on the correctness of the matching function of class probability distribution (p,1-p) and class convergence, the superiority of convergence performance can be used to promote sample iterative convergence and improve the classification training speed.
[0089] In summary, the transformer fault diagnosis method based on online monitoring of gas based on the embodiment of the present application is explained, which can identify abnormal conditions that may exist inside the transformer and realize monitoring and fault diagnosis of the internal conditions of the transformer.
[0090] Figure 4 FIG. 1 is a block diagram of a transformer fault diagnosis system 100 based on online gas monitoring according to an embodiment of the present application. Figure 4 As shown, according to an embodiment of the present application, a transformer fault diagnosis system 100 based on online monitoring of gas includes: a volume detection module 110, used to monitor changes in the light gas volume value; an action initiation module 120, used to determine whether to initiate a light gas action based on changes in the light gas volume value; a component quantitative analysis module 130, used to perform component quantitative analysis on the gas to obtain an analysis result in response to initiating a light gas action; and a fault analysis module 140, used to determine whether to generate a transformer fault alarm based on the analysis result.
[0091] Here, those skilled in the art can understand that the specific functions and operations of each module in the above-mentioned transformer fault diagnosis system based on gas online monitoring 100 have been referred to above. Figures 1 to 3 The description of the transformer fault diagnosis method based on online monitoring of gas has been introduced in detail, and therefore, its repeated description will be omitted.
[0092] As described above, the transformer fault diagnosis system 100 based on online gas monitoring according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a transformer fault diagnosis algorithm based on online gas monitoring. In one example, the transformer fault diagnosis system 100 based on online gas monitoring according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the transformer fault diagnosis system 100 based on online gas monitoring can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the transformer fault diagnosis system 100 based on online gas monitoring can also be one of the many hardware modules of the wireless terminal.
[0093] Alternatively, in another example, the transformer fault diagnosis system based on online gas monitoring 100 and the wireless terminal may also be separate devices, and the transformer fault diagnosis system based on online gas monitoring 100 may be connected to the wireless terminal via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.
[0094] Figure 5FIG. 1 is an application scenario diagram of a transformer fault diagnosis method based on online gas monitoring according to an embodiment of the present application. Figure 5 As shown, in this application scenario, first, a time series of capacitance values collected by a capacitance sensor is obtained (for example, Figure 5 D), then, the time series of the capacitance value is input to a server (for example, Figure 5 In S) shown in , the server can use the transformer fault diagnosis algorithm based on online gas monitoring to process the time series of the capacitance value to obtain a decision result indicating whether to start the light gas action.
[0095] In addition, it will be appreciated by those skilled in the art that various aspects of the present application may be illustrated and described by a number of patentable categories or situations, including any new and useful process, machine, product or combination of substances, or any new and useful improvements thereto. Accordingly, various aspects of the present application may be performed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software may all be referred to as "data blocks", "modules", "engines", "units", "components" or "systems". In addition, various aspects of the present application may be represented as a computer product located in one or more computer-readable media, which includes computer-readable program code.
[0096] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an idealized or extremely formal sense, unless explicitly defined as such herein.
[0097] The above is an explanation of the present application and should not be considered as a limitation thereof. Although several exemplary embodiments of the present application are described, it will be readily understood by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present application. Therefore, all of these modifications are intended to be included within the scope of the present application as defined by the claims. It should be understood that the above is an explanation of the present application and should not be considered to be limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present application is defined by the claims and their equivalents.
Claims
1. A transformer fault diagnosis method based on online monitoring of gas, characterized in that: include: Monitor the changes in light gas volume value; Determining whether to start a light gas action based on the change in the light gas volume value; In response to starting the light gas action, performing a quantitative component analysis on the gas to obtain an analysis result; as well as Based on the analysis result, determining whether to generate a transformer fault alarm; Determining whether to start a light gas action based on the change in the light gas volume value includes: Acquire a time series of capacitance values collected by a capacitance sensor; Performing data regularization on the time series of the capacitance values to obtain a sequence of local capacitance timing input vectors; Performing capacitance timing feature extraction on the sequence of local capacitance timing input vectors to obtain a sequence of local capacitance timing feature vectors; quantizing the degree of capacitance semantic discreteness of the sequence of the local capacitance time series feature vectors to obtain a capacitance semantic discrete feature vector; and Based on the capacitance semantic discrete feature vector, it is determined whether to start the light gas action.
2. The transformer fault diagnosis method based on online gas monitoring according to claim 1 is characterized in that: The time series of the capacitance values is subjected to data regularization to obtain a sequence of local capacitance timing input vectors, including: Arranging the time series of the capacitance values into a capacitance time series input vector according to the time dimension; and The capacitance timing input vector is vector-segmented to obtain a sequence of the local capacitance timing input vectors.
3. The transformer fault diagnosis method based on online gas monitoring according to claim 2 is characterized in that: Performing capacitance timing feature extraction on the sequence of local capacitance timing input vectors to obtain a sequence of local capacitance timing feature vectors includes: The sequence of the local capacitance timing input vectors is passed through a capacitance timing feature capture module based on a one-dimensional convolutional neural network model to obtain a sequence of the local capacitance timing feature vectors.
4. The transformer fault diagnosis method based on online gas monitoring according to claim 3 is characterized in that: Quantifying the degree of capacitance semantic discreteness of the sequence of the local capacitance time series feature vector to obtain a capacitance semantic discrete feature vector includes: Performing nonlinear response compensation on the sequence of local capacitance timing characteristic vectors to obtain a sequence of compensated local capacitance timing characteristic vectors; Using the last compensated local capacitance time series feature vector in the sequence of compensated local capacitance time series feature vectors as a query vector; and The query vector and other compensated local capacitance time series feature vectors in the sequence of compensated local capacitance time series feature vectors are respectively input into a population divergence measurement module to obtain the capacitance semantic discrete feature vector composed of multiple capacitance semantic discrete coefficients.
5. The transformer fault diagnosis method based on online gas monitoring according to claim 4 is characterized in that: Performing nonlinear response compensation on the sequence of local capacitance time series characteristic vectors to obtain a sequence of compensated local capacitance time series characteristic vectors, comprising: The local capacitance timing characteristic vector is processed by the following nonlinear response compensation formula to obtain the compensated local capacitance timing characteristic vector; wherein the nonlinear response compensation formula is: Among them, L 0i is the eigenvalue of the i-th position in the local capacitance time series eigenvector, L 1i is the characteristic value of the i-th position in the compensated local capacitance timing characteristic vector, and A, B, C and D are adjustment parameters with different values.
6. The transformer fault diagnosis method based on online gas monitoring according to claim 5 is characterized in that: The query vector and other compensated local capacitance time series feature vectors in the sequence of the compensated local capacitance time series feature vector are respectively input into a population divergence measurement module to obtain the capacitance semantic discrete feature vector composed of a plurality of capacitance semantic discrete coefficients, including: The query vector and other compensated local capacitance time series feature vectors in the sequence of the compensated local capacitance time series feature vector are processed using the following population separation measurement formula to obtain the capacitance semantic discrete feature vector composed of the multiple capacitance semantic discrete coefficients; wherein the population separation measurement formula is: Among them, e i is the semantic discrete coefficient of the i-th capacitor, v q is the query vector, v i is the i-th compensated local capacitance time series feature vector in the sequence of compensated local capacitance time series feature vectors, W is the weight coefficient matrix, B is the weight coefficient vector, tanh(·) represents the hyperbolic tangent function, (·) T Represents the transpose of a vector.
7. The transformer fault diagnosis method based on online gas monitoring according to claim 6 is characterized in that: Determining whether to start a light gas action based on the capacitance semantic discrete feature vector includes: The capacitance semantic discrete feature vector is passed through a control decision maker based on a classifier to obtain a decision result, and the decision result is used to indicate whether to start a light gas action.
8. The transformer fault diagnosis method based on online gas monitoring according to claim 7 is characterized in that: The method also includes a training step: training the capacitor timing feature capture module based on the one-dimensional convolutional neural network model, the population separation measurement module and the classifier-based control decision maker.
9. The transformer fault diagnosis method based on online gas monitoring according to claim 8 is characterized in that: The training step comprises: Acquire training data, where the training data includes a time series of training capacitance values collected by the capacitance sensor and a real value of whether to start a light gas action; Arranging the time series of the training capacitance values into a training capacitance timing input vector according to the time dimension; Performing vector segmentation on the training capacitance timing input vector to obtain a sequence of training local capacitance timing input vectors; Passing the sequence of the training local capacitance timing input vectors through the capacitance timing feature capture module based on the one-dimensional convolutional neural network model to obtain a sequence of training local capacitance timing feature vectors; Performing nonlinear response compensation on the sequence of training local capacitance time series feature vectors to obtain a sequence of training compensated local capacitance time series feature vectors; The last trained compensated local capacitance time series feature vector in the sequence of the trained compensated local capacitance time series feature vectors is used as a query vector, and the query vector and other trained compensated local capacitance time series feature vectors in the sequence of the trained compensated local capacitance time series feature vectors are respectively input into the population divergence measurement module to obtain a trained capacitance semantic discrete feature vector composed of a plurality of capacitance semantic discrete coefficients; Passing the training capacitor semantic discrete feature vector through the classifier-based control decision maker to obtain a classification loss function value; and The capacitor timing feature capture module based on the one-dimensional convolutional neural network model, the population separation measurement module and the classifier-based control decision maker are trained using the classification loss function value, wherein in each round of iteration of the training, the training capacitor semantic discrete feature vector is iteratively optimized.
