A method for early warning of failure of auxiliary equipment of a hydropower station

Abstract

The application relates to the technical field of data processing and computer-aided monitoring, and discloses a hydropower station auxiliary equipment fault early warning method, main steps of which include: supplementing and denoising preprocessing of original operation data of the hydropower station auxiliary equipment; situation understanding of the preprocessed data; situation assessment of feature data of the hydropower station auxiliary equipment by constructing a model; and fault prediction according to the model constructed by the situation assessment. The application is a data-driven machine learning fault early warning method, which deeply mines the operation data features of the hydropower station auxiliary equipment, combines feature extraction, feature fusion, oversampling and machine learning methods, and predicts the operation situation of the equipment in a future period of time, and early warns possible faults, so that the accuracy, interpretability and reliability of the prediction result are ensured.

Description

Technical Field

[0001] This invention relates to the fields of data processing and computer-aided monitoring technology, and in particular to a method for early warning of faults in auxiliary equipment of hydropower stations. Background Technology

[0002] With the continuous improvement of the localization level of hydropower equipment manufacturing, the demand for safety and reliability of equipment operation in hydropower stations is also constantly increasing. Therefore, preventing failures from the source and ensuring the stable operation of equipment are urgent problems that hydropower stations need to solve. A hydropower generation system is a complex nonlinear dynamic system with water, electromechanical, and mechanical coupling. The operation of its auxiliary equipment is related to a large number of uncertain factors, and the occurrence and development of failures in auxiliary equipment have a direct impact on the main equipment.

[0003] Fault early warning methods refer to analyzing the development trend of the hydropower plant's operating status over a future period based on historical experience data and current changes in unit operation before component failure or malfunction occurs, thereby detecting component anomalies in advance. Existing fault early warning methods mainly include statistical methods, machine learning-based methods, and deep learning-based methods. Statistical methods make predictions based on statistical data and construct traditional time series model equations, such as autoregressive models (AR models) and autoregressive differential moving average models (ARIMA models). These traditional time series models are essentially simple linear models, unable to capture nonlinear relationships, and have very limited capacity to represent complex problems, making them difficult to apply to non-stationary and nonlinear hydropower plant unit signals. Machine learning-based methods generally use feature engineering and model training methods to supervise learning on a large number of historical data samples, including decision trees and support vector machines. This method can solve most complex modeling problems, but it requires manual feature engineering and therefore depends on high-quality, large-scale data samples, making it unsuitable for direct application to small-sample hydropower plant unit data with imbalanced classes. Deep learning-based methods model data scenarios by constructing network models, including Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs). However, these methods ignore the inherent characteristics of data samples, resulting in uninterpretable predictions. Furthermore, most predictions fall into the category of single-step predictions, with short prediction durations and a lack of practical significance. Summary of the Invention

[0004] The technical problem to be solved by this invention is to propose a method for early warning of faults in auxiliary equipment of hydropower stations, which deeply mines the operating data characteristics of auxiliary equipment of hydropower stations, predicts the operating status of the equipment in the future period of time, and provides early warning of possible faults.

[0005] To address the aforementioned technical problems, this invention proposes a method for early warning of faults in auxiliary equipment of hydropower stations, comprising the following main steps:

[0006] S1: Perform preprocessing such as filling gaps and noise reduction on the raw operating data of auxiliary equipment in hydropower stations;

[0007] S2: Situational understanding of preprocessed data;

[0008] S3: Construct a model to assess the situation based on the characteristic data of auxiliary equipment in hydropower stations;

[0009] S4: Fault prediction based on the model built from the situation assessment.

[0010] Furthermore, step S1 specifically includes the following steps:

[0011] S11: Initial processing of the original hydropower station auxiliary equipment data is performed using default value processing and sliding time window technology;

[0012] S12: Perform noise reduction filtering on the hydropower station auxiliary equipment operation data after subsequence segmentation.

[0013] Furthermore, step S2 specifically includes the following steps:

[0014] S21: Perform abrupt change detection on the pre-processed operating data of hydropower station auxiliary equipment, find the measurement point signals that will change abruptly, and use them as measurement points that are strongly correlated with the fault status of hydropower station auxiliary equipment.

[0015] S22: Determine the data characteristics of each of the measurement points;

[0016] S23: Generate weight coefficients for the data features corresponding to the measuring points based on the topology diagram of the auxiliary equipment of the hydropower station;

[0017] S24: The overall auxiliary equipment operation data characteristics and the fault status of the corresponding time window are combined to form a supervised learning data sample to be classified.

[0018] Furthermore, step S3 specifically includes the following steps:

[0019] S31: Construct a training set for the support vector machine model based on the feature samples of auxiliary equipment operation data;

[0020] S32: Construct an SVM multi-classifier, and achieve multi-classification by combining one-to-one SVM binary classifiers;

[0021] S33: Train and optimize the binary classifier for each auxiliary equipment fault of the constructed hydropower station;

[0022] S34: Solve for the optimization objective of the binary classifier model for auxiliary equipment faults in hydropower stations;

[0023] S35: Construct a multi-classifier for faults in auxiliary equipment of hydropower stations by combining the output of the SVM binary classifier.

[0024] Further, step S25 is then performed: the supervised learning data samples are oversampled to synthesize fault samples of the minority class.

[0025] Preferably, step S21 specifically includes:

[0026] S21-1: Constructing the Order Column The order column sk is the cumulative number of times the measured value of the auxiliary equipment measuring point of the hydropower station at time i is greater than the value at time j;

[0027] S21-2: Define the operational status statistics of measuring points for auxiliary equipment in hydropower stations In the formula It is the cumulative number s k The mean, It is the cumulative number s k The variance is defined as UF1 = 0;

[0028] S21-3: Reverse the running data sequence and repeat the above steps to obtain the UF in reverse order. k r, calculate UB k =-UF k r;

[0029] S21-4: Analyze the operating status of the measuring points of the auxiliary equipment of the hydropower station, set the critical value, and set UF k and UB k The two statistical series curves and the two straight lines of the critical value are plotted on the same graph. If the intersection of the two curves is within the confidence critical value range, it indicates that the operating data series of the measuring point has abrupt change. Then the measuring point is a measuring point that is strongly correlated with the failure of the auxiliary equipment of the hydropower station.

[0030] S21-5: Following the steps above, find all the measuring points that are strongly correlated with the fault conditions of the auxiliary equipment of the hydropower station, and represent them as set V.

[0031] Preferably, the data features of the measuring point obtained in S22 include the slope change k of the measuring point over all time windows. i (2≤i≤l), mean μ and variance σ 2 .

[0032] Preferably, the specific steps of S23 include:

[0033] S23-1: Count the total number of measurement points that are strongly correlated with the fault in the system's topology graph, i.e., the total number of measurement points C in set V;

[0034] S23-2: Represent a certain measuring point in V as vi Statistics on the relationship between v in the system's topological graph i The number of directly connected measurement points that belong to set V is denoted as C. i Then the measuring point v i The weighting coefficient is The weighting coefficient of a measuring point represents the importance of that measuring point to the fault status of auxiliary equipment in a hydropower station.

[0035] S23-3: The weighting coefficient of each measuring point strongly correlated with the fault is represented by p. i (1≤i≤C), the data characteristics of the measuring point within a time window are represented as follows: The two factors are weighted and fused to obtain the overall operational characteristics of the hydropower station's auxiliary equipment within a given time window.

[0036] S23-4: Repeat the above steps in each time window to obtain the overall operating characteristics of the hydropower station's auxiliary equipment throughout the entire time series.

[0037] Preferably, step S25 specifically includes:

[0038] S25-1: For a certain fault sample x i ={X i Z i}, calculate its distance to all samples in the fault sample set using Euclidean distance as the standard, and obtain its q nearest neighbors;

[0039] S25-2: From x i Randomly select a sample x' from the q nearest neighbors. i Then generate a random number δ between 0 and 1. i Thus, a new sample is synthesized.

[0040] S25-3: Set the sampling rate N according to the sample imbalance ratio, repeat the above steps N times, and synthesize a total of N new fault samples. This yields a balanced and high-quality sample of operational data for hydropower station auxiliary equipment.

[0041] The beneficial effects of this invention are:

[0042] (1) A fault early warning method specifically for auxiliary equipment of hydropower stations is proposed, which makes up for the shortcomings of existing fault prediction technology in the power field.

[0043] (2) In the scenario of predicting the failure of auxiliary equipment of hydropower station, we will conduct a detailed situational understanding, deeply explore the operating data characteristics of auxiliary equipment of hydropower station, fully extract, analyze, merge and process the data, clarify the overall characteristics of the data, lay the foundation for situational awareness, and ensure the reliability of the prediction results.

[0044] (3) Based on high-quality data samples, a model is constructed to accurately assess the operation status of auxiliary equipment in hydropower stations, helping managers to grasp the operation status of the equipment, predict its operation trend in the future, and provide early warning of possible faults, which has practical application value. Attached Figure Description

[0045] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] Figure 1 This is a schematic diagram of the sliding time window technique.

[0047] Figure 2 This is a statistical curve for mutation point detection.

[0048] Figure 3 This is a topology diagram of auxiliary equipment in a hydropower station.

[0049] Figure 4 This is a topological diagram showing the location relationships of the measurement points.

[0050] Figure 5 This is a flowchart for situation assessment.

[0051] Figure 6 A flowchart for early warning of faults in auxiliary equipment of a hydropower station.

[0052] Figure 7 This is a test diagram for real-time fault early warning of oil level measuring points in oil leak tanks and pressure measuring points in pressure tanks.

[0053] Figure 8 This is a test diagram for real-time fault early warning of water level measuring points in a water tank. Detailed Implementation

[0054] This invention relates to a method for early warning of faults in auxiliary equipment of a hydropower station, used to predict the operating status of auxiliary equipment of a hydropower station over a future period of time and to provide early warning of possible faults, including the following steps S1 to S4:

[0055] S1: Preprocessing the raw operating data of the hydropower station's auxiliary equipment. Preprocessing the raw data lays the foundation for the situation understanding process and includes the following steps S11 and S12.

[0056] S11: Initial processing of the raw hydropower station auxiliary equipment data is performed using default value processing and sliding time window techniques. First, the collected raw hydropower station auxiliary equipment operating data is represented as f(t). Since f(t) is a continuous time-series signal, when the data signal has a default value due to acquisition, observation, or recording errors, the data value of the previous moment is used to fill in the default value of the current moment to ensure the integrity and continuity of the time-series data signal. The f(t) signal after default value processing is represented as y(t).

[0057] Furthermore, a sliding time window technique is used to segment the time series data y(t) into subsequences. If the complete time span covered by the time series data y(t) is n (where n represents the complete time span of the time series data in seconds), then y(t) = (y(t1), y(t2), ..., y(t...). n ), where y(t) i ) for t i Data recorded at any time, collection time t i Strictly incremental, the time series is divided into equal-length segments using a sliding window of length l (l is a user-defined parameter representing the time span of the sliding window for the time series data, in seconds, and l is much smaller than n). Then, the window moves forward by one step r each time, continuously sliding... Next, forming a common Several equal-length subsequence segments, such as Figure 1 As shown. Therefore, the time-series operation data of the hydropower station auxiliary equipment after initial preprocessing and subsequence segmentation can be obtained, represented as... in It represents a subsequence within a time window.

[0058] S12: Denoising filtering is performed on the hydropower station auxiliary equipment operation data Y(t) after subsequence segmentation. The hydropower station auxiliary equipment system is affected by hydraulic factors, strong electromagnetic fields, and various noises within the plant, resulting in a large amount of noise in the collected signals. Therefore, it is necessary to suppress the noise to the maximum extent possible and restore the true signal. This specific implementation provides a noise reduction example based on the least squares principle, which modulates the signal within a sliding time window. To perform polynomial fitting and achieve smoothing filtering, the specific steps are as follows:

[0059] S12-1: Each sliding time window The length is l, where each measurement point of the timing operation data of the auxiliary equipment of the hydropower station is represented as follows: A polynomial of degree k-1 (where k is a user-defined parameter, and k-1 represents the highest power of the polynomial to be fitted) is used to fit the data within the window, i.e. Where x is the original data to be fitted, a is the polynomial coefficient to be solved, and y is the output data after smoothing and filtering the original data;

[0060] S12-2: The fitting parameters were determined by fitting using the least squares method, resulting in l equations within each window, forming a system of k linear equations. Converted into matrix representation, Y l×1 =X l×k ·A k×1 +B l×1 ;

[0061] S12-3: The least squares solution for parameter A is... Therefore, the output value after filtering and denoising the original is By analogy, we can obtain the total The output values ​​within a sliding time window enable smooth filtering of the operating data of hydropower station auxiliary equipment.

[0062] The operating data of hydropower station auxiliary equipment obtained after S12 noise reduction filtering is represented as follows: It provides the necessary foundational data for the situational understanding process.

[0063] S2: Situational understanding is performed on the preprocessed data. The overall method for situational understanding is to mine the operational data characteristics of hydropower station auxiliary equipment, identify signals that may cause abrupt changes as measurement point data strongly correlated with faults, select the slope change, variance, etc. of the measurement point within a time window as its data features, and further analyze, weightedly fuse, and oversample these features to generate balanced, high-quality operational data samples of hydropower station auxiliary equipment, which serve as the basis for situational awareness. Specifically, this includes the following steps:

[0064] S21: Pre-processed operating data of auxiliary equipment in hydropower stations Sudden change point detection is performed to identify measurement point signals prone to sudden changes, which are then considered as measurement points strongly correlated with the fault conditions of hydropower station auxiliary equipment. Time series data for each measurement point are then analyzed. To obtain a running data sequence x with a time length of L, the specific steps are as follows:

[0065] S21-1: Constructing the Order Column Order column s k It is the cumulative number of times the measured value of the auxiliary equipment measuring point of the hydropower station at time i is greater than the value at time j;

[0066] S21-2: Define the operational status statistics of measuring points for auxiliary equipment in hydropower stations In the formula It is the cumulative number s k The mean, It is the cumulative number sk The variance is defined as UF1 = 0;

[0067] S21-3: Reverse the running data sequence x and repeat the above steps to obtain the UF in reverse order. k r, calculate UB k =-UF k r;

[0068] S21-4: Analyze the operating status of the measuring points of the auxiliary equipment in the hydropower station and set the critical value U. 0.05 =±1.96, UF k and UB k The two statistical series curves and the two straight lines ±1.96 are plotted on the same graph. If the intersection of the two curves is within the confidence level interval [-1.96, 1.96], it indicates that the operational data series of that measurement point has abrupt changes. Figure 2 As shown, this measuring point is strongly correlated with the failure status of auxiliary equipment in hydropower stations.

[0069] S21-5: Following the steps above, find all the measuring points that are strongly correlated with the fault conditions of the auxiliary equipment of the hydropower station, and represent them as set V.

[0070] S22: For the set of measuring points V obtained from S21 that are strongly correlated with the faults of hydropower station auxiliary equipment, determine the data characteristics of each measuring point. (This refers to the time-series operation data of hydropower station auxiliary equipment.) Subsequence within a time window The slope between any two adjacent data points in this subsequence Therefore in the subsequence There are l-1 slope values ​​in total, and the subsequence The mean can be expressed as variance is This allows us to obtain the slope change k of measuring points strongly correlated with the fault status of hydropower station auxiliary equipment over all time windows. i (2≤i≤l), mean μ and variance σ 2 As data features of measurement points in set V.

[0071] S23: Based on the topology diagram of the auxiliary equipment of the hydropower station, such as Figure 3 As shown, the topological location relationship of the measuring points strongly correlated with the fault can be known, such as... Figure 4 As shown. Then, weight coefficients are generated for the data features corresponding to the measurement points. The specific steps are as follows:

[0072] S23-1: Count the total number of measurement points that are strongly correlated with the fault in the system's topology graph, which is the total number of measurement points in set V obtained in S21, denoted as C;

[0073] S23-2: Represent a certain measuring point in V as v i Statistics are shown in the system's topological relationship graph and v i The number of directly connected measurement points that are strongly correlated with the fault (belonging to set V) is denoted as C. i Then the measuring point v i The weighting coefficient is The weighting coefficient of the measuring point represents the importance of the measuring point to the fault status of the auxiliary equipment of the hydropower station.

[0074] S23-3: The weighting coefficient of each measuring point strongly correlated with the fault is represented by p. i (1≤i≤C), the data characteristics of the measuring point within a time window are represented as follows: The two factors are weighted and fused to obtain the overall operational characteristics of the hydropower station's auxiliary equipment within a given time window.

[0075] S23-4: Repeat the above steps in each time window to obtain the overall operating characteristics of the hydropower station's auxiliary equipment throughout the entire time series.

[0076] S24: The overall operational characteristics of the hydropower station auxiliary equipment obtained in S23 and the corresponding time window for fault status Combined, where Z i =0 indicates that there are no faults in the auxiliary equipment of the hydropower station, Z i =1 indicates that a fault has occurred in an auxiliary device of a hydropower station, forming a supervised learning data sample to be classified.

[0077] S25: Due to the insufficient number of fault sample data and the severe imbalance between sample classes, it is necessary to oversample the samples to synthesize fault samples of the minority class. The specific steps are as follows:

[0078] S25-1: For a certain fault sample x i ={X i Z i}, calculate its distance to all samples in the fault sample set using Euclidean distance as the standard, and obtain its q nearest neighbors;

[0079] S25-2: From x i Randomly select a sample x' from the q nearest neighbors. i Then generate a random number δ between 0 and 1. i Thus, a new sample is synthesized.

[0080] S25-3: Set the sampling rate N according to the sample imbalance ratio, repeat the above steps N times, and synthesize a total of N new fault samples. Ultimately, a balanced and high-quality sample of hydropower station auxiliary equipment operation data can be obtained, denoted as M. Through in-depth analysis of the operational data characteristics of hydropower station auxiliary equipment, and by fully extracting, analyzing, merging, and processing the data, a detailed situational understanding was achieved, laying the foundation for the next step of situational assessment and ensuring the reliability of the prediction results.

[0081] S3: Construct a model to assess the situation based on the characteristic data of the hydropower station's auxiliary equipment. Considering the computational speed of the final model, an efficient support vector machine model is selected for training. The characteristic data is input into the model and fitted with the fault situation.

[0082] Step S3 involves constructing a Support Vector Machine (SVM) model to assess the situation of the auxiliary equipment of the hydropower station, specifically including the following steps:

[0083] S31: In step S2, a balanced feature sample M of the hydropower station auxiliary equipment operation data was obtained, which can then be used to construct the training set of the support vector machine model, represented as T={(M1,y1),(M2,y2),...,(M m ,y m )}, where m represents the total number of data samples, M i (1≤i≤m) represents the overall operational characteristics of the auxiliary equipment of the hydropower station in the i-th time window, y i (1≤i≤m) represents the fault status of the hydropower station's auxiliary equipment at the end of the i-th time window, y i =0, 1, 2, 3 represent no fault in the auxiliary equipment of the hydropower station, fault in the high-pressure gas system, fault in the oil system, and fault in the drainage system, respectively.

[0084] S32: The fault condition assessment problem for auxiliary equipment in hydropower stations is a multi-classification problem, requiring the construction of an SVM multi-classifier. This multi-classification is achieved by combining one-to-one SVM binary classifiers. Six SVM binary classifiers are constructed for four categories: no faults, high-pressure gas system faults, oil system faults, and drainage system faults. These are, in order, a no-fault to high-pressure gas system fault classifier (SVM). 0-1 Fault-free oil system fault classifier SVM 0-2 Fault-free drainage system fault classifier SVM 0-3 High-pressure gas system fault-oil system fault classifier SVM 1-2 High-pressure gas system fault-drainage system fault classifier SVM 1-3 And the oil system failure-drainage system failure classifier SVM 2-3 .

[0085] S33: Train and optimize the binary classifier for each hydropower station auxiliary equipment fault. Using the feature samples M of hydropower station auxiliary equipment operation data, construct an SVM binary classifier. For data samples of class u (0≤u≤3) and class v (0≤v≤3), the classification function is f(M)=ω T φ(M)+b, the objective function to be optimized is Where C represents the penalty factor, ζ i This represents a slack variable.

[0086] S34: Solve for the optimization objective of the binary classifier model for auxiliary equipment faults in the hydropower station, i.e., solve for each of the six SVM binary classifiers. The conditional extremum problem in S33 is transformed into an unconstrained quadratic programming problem, with the dual optimization objective being... The inner product of the mapping functions is the kernel function K(M). i M j )=φ(M i ) T φ(M j ), choose the polynomial kernel function K(M) i M j )=(M i ·M j +c) d The optimal solution α is obtained by solving the problem. * This leads to the derivation of the optimal solution ω. * b * By combining the target classification hyperplane with the target classification hyperplane, six SVM binary classification results for the fault status of auxiliary equipment in hydropower stations can be obtained.

[0087] S35: Construct a multi-classifier for auxiliary equipment faults in hydropower stations by combining the outputs of six SVM binary classifiers. A voting strategy is adopted; for the data samples of class u and class v mentioned above, if the final solution (ω... uv ) T φ(x i )+b uv If the value is greater than 0, then the sample x will be... i The votes for class u are incremented by one; otherwise, the votes for class v are incremented by one, and finally x is... i The sample was categorized into the class with the most votes, thus determining the fault condition category and enabling a situational assessment of the hydropower station's auxiliary equipment operational data. The overall situational assessment process is as follows: Figure 5 As shown.

[0088] Based on the high-quality hydropower station auxiliary equipment data sample M obtained through situational understanding in S2, this step trains a multi-class support vector machine model, which classifies four types of sample datasets: no faults, high-pressure gas system faults, oil system faults, and drainage system faults. This enables accurate situational assessment of the operation of hydropower station auxiliary equipment, helping managers to grasp the equipment's operating status in real time.

[0089] S4: Fault prediction based on the model built from the situation assessment. The multi-class support vector machine model trained in the S3 situation assessment is used to predict the future fault conditions of the hydropower station's auxiliary equipment in real time, so as to provide early warning before the actual occurrence of the fault. This step is the ultimate goal of situation awareness.

[0090] Integrating all steps involved in S1 to S4, the overall process for early warning of auxiliary equipment failures in hydropower stations is as follows: Figure 6 As shown, based on data preprocessing, situation understanding, and situation assessment, this step uses a trained SVM prediction model to test the test samples in order to achieve real-time early warning of faults.

[0091] Situation prediction was performed using test samples, and relevant measurement point data was extracted. The test results are as follows: Figure 7 , Figure 8 As shown. Figure 7 , Figure 8 As can be seen, the well-trained SVM prediction model can provide fault warnings based on the real-time changes in the measurement point values, and successfully detect potential future faults.

[0092] Finally, it should be noted that the above specific embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for early warning of faults in auxiliary equipment of a hydropower station, characterized in that, The main steps include: S1: Perform preprocessing such as filling gaps and noise reduction on the raw operating data of auxiliary equipment in hydropower stations; S2: Perform situational understanding on the preprocessed data. The process includes the following steps: S21: Detecting abrupt changes in the preprocessed hydropower station auxiliary equipment operation data to identify measurement point signals that are prone to abrupt changes, which are then used as measurement points strongly correlated with the fault conditions of the hydropower station auxiliary equipment; S22: Determining the data characteristics of each measurement point; S23: Generating weight coefficients for the data characteristics corresponding to the measurement points based on the topological relationship diagram of the hydropower station auxiliary equipment; S24: Combining the overall auxiliary equipment operation data characteristics with the fault conditions of the corresponding time window to form supervised learning data samples to be classified. S3: Build a model to assess the situation based on the characteristic data of hydropower station auxiliary equipment. The process includes the following steps: S31: Construct a training set for the Support Vector Machine (SVM) model based on the feature samples of auxiliary equipment operation data; S32: Construct an SVM multi-classifier and achieve multi-classification by combining one-to-one SVM binary classifiers; S33: Train and optimize each constructed hydropower station auxiliary equipment fault binary classifier separately; S34: Solve for the optimization objective of the hydropower station auxiliary equipment fault binary classifier model; S35: Construct a hydropower station auxiliary equipment fault multi-classifier by combining the output results of the SVM binary classifier. S4: Fault prediction based on the model built from the situation assessment.

2. The method for early warning of faults in auxiliary equipment of a hydropower station according to claim 1, characterized in that, Step S1 specifically includes the following steps: S11: Initial processing of the original hydropower station auxiliary equipment data is performed using default value processing and sliding time window technology; S12: Perform noise reduction filtering on the hydropower station auxiliary equipment operation data after subsequence segmentation.

3. The method for early warning of faults in auxiliary equipment of a hydropower station according to claim 1, characterized in that, Step S25 follows step S24: oversampling is performed on the supervised learning data samples to synthesize fault samples of the minority class.

4. The method for early warning of faults in auxiliary equipment of a hydropower station according to claim 1, characterized in that, The specific steps in S21 include: S21-1: Constructing the Order Column Order column s k The measured value x at the measuring point of the auxiliary equipment of the hydropower station at time i. i Greater than the value x at time j j The cumulative number of items; S21-2: Define the operational status statistics of measuring points for auxiliary equipment in hydropower stations In the formula It is the cumulative number s k The mean of the data, where L is the total number of samples in the measurement data. It is the cumulative number s k The variance is defined as UF1 = 0; S21-3: Reverse the running data sequence and repeat the above steps to obtain the UF in reverse order. k r, calculate UB k =-UF k r; S21-4: Analyze the operating status of the measuring points of the auxiliary equipment of the hydropower station, set the critical value, and set UF k and UB k The two statistical series curves and the two straight lines of the critical value are plotted on the same graph. If the intersection of the two curves is within the confidence critical value range, it indicates that the operating data series of the measuring point has abrupt change. Then the measuring point is a measuring point that is strongly correlated with the failure of the auxiliary equipment of the hydropower station. S21-5: Following the steps above, find all the measuring points that are strongly correlated with the fault conditions of the auxiliary equipment of the hydropower station, and represent them as set V.

5. The method for early warning of faults in auxiliary equipment of a hydropower station according to claim 4, characterized in that, The data characteristics of the measuring point obtained in S22 include the slope change k of the measuring point over all time windows. i (2≤i≤l), mean μ and variance σ 2 , where l is the total number of all time windows.

6. The method for early warning of faults in auxiliary equipment of a hydropower station according to claim 5, characterized in that, The specific steps of S23 include: S23-1: Count the total number of measurement points that are strongly correlated with the fault in the system's topology graph, i.e., the total number of measurement points C in set V; S23-2: Represent a certain measuring point in V as v i Statistics on the relationship between v in the system's topological graph i The number of directly connected measurement points that belong to set V is denoted as C. i Then the measuring point v i The weighting coefficient is The weighting coefficient of a measuring point represents the importance of that measuring point to the fault status of auxiliary equipment in a hydropower station. S23-3: The weighting coefficient of each measuring point strongly correlated with the fault is represented by p. i (1≤i≤C), the data characteristics of the measuring point within a time window are represented as follows: The two factors are weighted and fused to obtain the overall operational characteristics of the hydropower station's auxiliary equipment within a given time window. S23-4: Repeat the above steps in each time window to obtain the overall operating characteristics of the hydropower station's auxiliary equipment throughout the entire time series.

7. The method for early warning of faults in auxiliary equipment of a hydropower station according to claim 3, characterized in that, Step S25 specifically includes: S25-1: For a certain fault sample x i ={X i Z i }, calculate its distance to all samples in the fault sample set using Euclidean distance as the standard, and obtain its q nearest neighbors; S25-2: From x i Randomly select a sample x′ from the q nearest neighbors. i Then generate a random number δ between 0 and 1. i Thus, a new sample is synthesized. S25-3: Set the sampling rate N according to the sample imbalance ratio, repeat the above steps N times, and synthesize a total of N new fault samples. This yields a balanced and high-quality sample of operational data for hydropower station auxiliary equipment.

Images