Hydroelectric generating set fault early warning method and system fusing time domain and frequency domain characteristics

Abstract

The invention relates to a hydroelectric generating set fault early warning method fusing time domain and frequency domain characteristics, and the method comprises the steps: building a hydroelectric generating set time domain health model based on a least square support vector machine, calculating a relative error between a time domain characteristic actual value and a health value of a set, and taking the relative error as a time domain fault symptom index of the set; combining a wavelet transform method with a singular value decomposition algorithm to construct a frequency domain fault symptom index of the unit; and fusing to obtain a time-frequency domain fault symptom index as a unit vibration degradation index, and realizing unit vibration abnormity real-time early warning according to the unit vibration degradation index. According to the method, effective combination of model driving and data driving is realized, a unit vibration fault symptom index fusing signal time domain information and frequency domain information is constructed for fault early warning, more comprehensive unit state monitoring is realized, the early fault symptom of the unit can be found, and the method is of great significance to timely and effective unit fault early warning.

Description

Technical Field

[0001] The present invention belongs to the technical field of hydropower unit fault warning, and in particular relates to a hydropower unit fault warning method and system that integrates time domain and frequency domain characteristics. Background Art

[0002] As key equipment for hydropower energy conversion, the safety and stability of hydropower units have always been a hot topic in the power industry. As hydropower units continue to develop towards larger sizes and greater complexity, their operating status is directly related to the efficiency of hydropower energy conversion, the safety of hydropower stations and the stability of the power grid. It is of great significance to grasp the real-time operating status of hydropower units, discover fault signs in a timely manner, and make quick and accurate judgments on equipment failures.

[0003] A hydropower unit is a complex nonlinear dynamic system with highly coupled hydraulic, mechanical and electrical factors. Its operating status has a vital impact on the safety of power stations and the stability of power grids. Research on equipment fault warning is to discover fault signs in the early stage of equipment degradation and take corresponding measures to avoid irreversible failures. It plays a guiding role in accurately controlling the operating status of the unit. In traditional monitoring systems, most of them use a single eigenvalue fixed threshold alarm to realize the unit fault alarm function. The threshold alarm trigger means that the equipment has deteriorated to a certain extent and cannot achieve early warning in the true sense. At the same time, the equipment signals of large units have non-stationary nonlinear characteristics, and a single eigenvalue is difficult to show the true state of the unit. Therefore, researching and developing effective nonlinear and non-stationary signal processing methods and deeply exploring the unit status characteristic information contained in the vibration signal are of great guiding significance for accurately analyzing the unit operation status and rationally formulating operation and maintenance strategies. Summary of the invention

[0004] The purpose of the present invention is to provide a hydropower unit fault warning method that integrates time domain and frequency domain features in order to solve the above problems. The method combines the support vector machine algorithm, wavelet singular value decomposition algorithm and K cluster analysis algorithm in the field of machine learning to realize a fault warning method for hydropower units. The method makes full use of the time-frequency characteristics of the vibration signal of the hydropower unit in normal state. First, the time domain fault symptom index of the unit is constructed based on the feature detection index and LS-SVM. Then, the frequency domain fault symptom index of the unit is constructed based on wavelet singular value decomposition and cluster analysis. Finally, the fault symptom index of the fusion signal time-frequency domain information is constructed. Then, the upper and lower limits of the fault symptom index are calculated based on the Bootstrap sampling method and the Gaussian threshold method. By judging whether the index exceeds the limit, the real-time warning of abnormal vibration of the unit is realized. The present invention applies a variety of machine learning algorithms and feature extraction methods to the fault warning of hydropower units in a new way to achieve a relatively better fault warning effect of hydropower units.

[0005] In order to achieve the above object, the technical solution provided by the present invention is: The hydropower unit fault early warning method integrating time domain and frequency domain characteristics comprises the following steps: Step 1: Select a time domain parameter index, take the historical operating condition data under the unit health state as the input sample, calculate the time domain characteristics of the input sample according to the time domain parameter index, and use the time domain characteristics as the output sample to construct a unit time domain health model based on the least squares support vector machine; Step 2: Input the operating condition data in the real-time monitoring sample of the unit into the unit time domain health model, use the time domain characteristic value output by the unit time domain health model as the time domain characteristic health value of the unit vibration signal at the current moment, calculate the relative error between the actual value of the time domain characteristic of the unit vibration signal and the time domain characteristic health value according to the time domain parameter index, use the relative error as the unit's time domain fault symptom index, and use the time domain fault symptom index to perform preliminary unit fault symptom diagnosis and abnormality warning; Step 3: extract the wavelet singular value eigenvectors of the unit vibration signal under the same working condition, construct a wavelet singular value eigenvector set under the same working condition, and use a clustering algorithm to calculate the center vector of the wavelet singular value eigenvector set as the standard feature of the unit vibration signal under the corresponding working condition; Step 4: extract the wavelet singular value eigenvector of the unit vibration signal in the real-time monitoring data sample of the unit, and calculate its Euclidean distance with the center vector obtained in step 3 as the frequency domain fault symptom indicator of the unit; Step 5: transform the frequency domain fault symptom index obtained in step 4, keep the transformed frequency domain fault symptom index identical to the time domain fault symptom index, fuse the transformed frequency domain fault symptom index with the time domain fault symptom index, and construct the fault symptom index of the unit vibration signal time-frequency domain information as the unit vibration degradation index; Step 6: Uniformly sample the vibration degradation index samples under normal conditions, calculate the standard deviation of each sample, and further calculate the standard deviation estimate of the overall sample of the unit vibration degradation index based on the normal prior distribution of the unit vibration degradation index. Use the Gaussian threshold method to determine the upper and lower limits of the unit vibration degradation index. By judging whether the real-time unit vibration degradation index is within the range of the upper and lower limits, a real-time warning of unit vibration abnormality is achieved.

[0006] Preferably, in step 1, a detection index in mathematical statistics is introduced as an indicator to measure the sensitivity of the time domain characteristics of the unit vibration signal, specifically including: Assumptions and They are the vibration signals of the unit under two different operating conditions. An indicator variable of a certain time domain characteristic, and , They satisfy the normal distribution , , They are , The mean of Respectively , The standard deviation of but The larger it is, the better the distinction between two different operating states; make , then the difference signal obey The normal distribution of , the corresponding probability density function is: ; (1) In the formula , but exist The probability distribution function of the range is: ; (2) In the formula represents the cumulative probability; The recognition rate DR is defined as the degree of distinction between the two different operating states of the unit by the time domain feature index T, and the calculation formula is: ; (3) make: ; (4) Substituting into equation (1) and equation (3), using substitution integration, we get: ; (5) In the formula represents the detection index; visible, The larger the value, the stronger the signal time domain characteristics. The greater the sensitivity to different states of the unit, the higher the corresponding discrimination; The specific expression of the value is: ; (6).

[0007] Furthermore, in step 1, the unit time domain parameter indicators are screened by the detection index DI, and by selecting multiple time domain parameters for comparative analysis, the standard deviation and peak-to-peak value with the largest detection index value are selected as indicators of the unit's time domain characteristic sensitivity.

[0008] Furthermore, in step 1, the construction of the unit time domain health model based on the least squares support vector machine specifically includes: Given contains N The training set of group samples ,in , Respectively Group input and output samples, and , ; The method for support vector machine to solve regression classification problem is to construct a mathematical model in the following form: ; (7) In the formula , Respectively represent the input and output of the support vector machine; is a mapping function that maps low-dimensional input vectors to high-dimensional space, , are all constants; The least squares method is introduced into the support vector machine model to obtain the following least squares support vector machine model: ; (8) In the formula, is the weight vector matrix, represents the error, called the relaxation factor, is the error of the kth group of samples, is the regularization parameter, also known as the penalty factor; The constraints of the least squares support vector machine model are: ; (9) In the formula, is a nonlinear mapping function; The Lagrange equation is introduced to transform the optimization problem with constraints into an optimization problem without constraints, and the linear regression equation of the least squares support vector machine is obtained according to the Karush-Kuhn-Tucker condition: ; (10) Formula (10) transforms the regression and classification problem of the support vector machine into solving a linear equation. and The mapping function is obtained by the least squares method. The radial basis kernel function is selected.

[0009] Furthermore, in step 2, the historical operating condition sample data X(t) of the unit under normal operation is used as input, and the time domain feature data Y(t) screened by the detection index is used as output. The unit state health model is constructed by the least squares support vector machine, and the method is as follows: The unit time domain health model based on least squares support vector machine is: ; (11) ; In the formula , They represent the mapping functions of the least squares support vector machine corresponding to the standard deviation and peak-to-peak value of the time domain features respectively; , are the standard deviation and peak-to-peak health values ​​obtained according to the unit time-domain health model respectively; t Indicates time.

[0010] The real-time operating condition data of the unit is input into the time domain health model of the hydropower unit shown in formula (11), and the time domain characteristic health value under the current corresponding operating condition can be obtained. The relative difference between the actual value of the time domain characteristic of the vibration signal and the time domain characteristic health value is calculated, and after weighted calculation, it is used as the time domain fault symptom indicator of the unit vibration signal. The expression is: ; (12) In the formula, It is the fault symptom index of the unit vibration signal in the time domain. , They are the sign indicators corresponding to the standard deviation of time domain characteristics and the peak-to-peak value, , is the actual value of the standard deviation and peak-to-peak value of the unit vibration signal, , are the weight coefficients of the standard deviation and peak-to-peak value of the time domain characteristics respectively.

[0011] Preferably, in step 2, the weight coefficient is calculated as follows: ; (13) ; (14) In the formula , are the weight coefficients of the standard deviation and peak-to-peak value of the time domain characteristics respectively.

[0012] Furthermore, the step 3 specifically includes the following sub-steps: Step 3.1: Decompose the vibration signal of the unit using the wavelet transform method to obtain a series of wavelet decomposition coefficients; In continuous wavelet transform, the calculation formula of wavelet signal is: ; (15) In the formula, represents the wavelet mother function, is the scale factor, is the translation factor, and Continuously changing in the real number domain, we get a continuous wavelet signal ; The calculation formula of wavelet decomposition coefficient is: ; (16) In the formula, is the wavelet decomposition coefficient; represents the basic wavelet function; , Indicates the upper and lower limits of the integral; represents the space of square integrable functions; Step 3.2: Perform singular value decomposition on the wavelet decomposition coefficients obtained in step 3.1 to obtain the wavelet singular value eigenvector; The wavelet decomposition coefficients are used as the input matrix A of the singular value decomposition, the matrix The eigenvalue of ; The square root of That is the matrix The singular values ​​of For the matrix The conjugate transposed matrix of ; The formula for matrix singular value decomposition is: ; (17) ; In the formula is a diagonal matrix, Both are matrices The singular values ​​of ; U is a unitary matrix, V is an orthogonal matrix; Transform formula (17) to obtain: ; (18) Step 3.3: Compare the real-time vibration signal wavelet singular value vector with the cluster center The Euclidean distance between them is used as the fault symptom index of the unit vibration signal frequency domain, and the calculation formula is: ; (19) In the formula, It is the fault symptom index of the unit vibration signal frequency domain. for The singular value eigenvector of the vibration signal measured at each moment; It is the time domain fault symptom indicator of the unit vibration signal.

[0013] Furthermore, in step 5, the unit vibration signal time domain fault symptom index and the unit vibration signal frequency domain fault symptom index are weightedly fused to obtain an overall time-frequency domain fault symptom index; The formula for weighted fusion of unit vibration signal time domain characteristics and unit vibration signal frequency domain characteristics is: ; (20) In the formula Indicates fault symptom indicators in the time-frequency domain; ;(twenty one) ;(twenty two) In the formula , are all weight coefficients.

[0014] Preferably, the step 6 specifically includes: Step 6.1: Construct the normal operating state of the unit under similar working conditions Fault symptom indicator samples independent and identically distributed at each moment ;in Respectively represent the fault symptom indicator variables at the 1st, 2nd, …, mth moments; Step 6.2: Sample fault symptom indicators obtained in step 6.1 Sampling with replacement is performed, and the independent sampling capacity is Bootstrap Failure Indicator Subsample , , are all fault symptom indicator sub-sample variables obtained by sampling, repeated times; calculate the statistic of each Bootstrap fault symptom indicator subsample, that is, the mean, recorded as ; Step 6.3: Get the value obtained in step 6.2 Subsample statistics , calculate its standard deviation ; Under known normal conditions, the vibration fault symptom indicators of the unit under the same working condition are generally close to normal distribution ,in represents the mean of the unit vibration fault symptom index under the same working condition; then the sub-sample mean obtained by Bootstrap sampling satisfies the normal distribution , so the standard deviation of the overall sample of fault symptom indicators is ; Step 6.4: Use the threshold method to determine the upper and lower limits of the fault symptom indicators; According to the Gaussian threshold method, the probability that the value of the overall sample standard deviation of the fault symptom indicator falls in the interval (μ-3s, μ+3s) is 99.74%, where μ represents the mean. Therefore, μ+3s and μ-3s are respectively used as the upper and lower limits of normal operation of the monitored quantity. By judging whether the indicator exceeds the limit, real-time warning of abnormal unit vibration can be achieved.

[0015] Compared with the prior art, the beneficial effects of the present invention include: 1) The present invention provides a fault warning method for hydropower units based on the fusion of LS-SVM time domain features and wavelet singular value decomposition frequency domain features. Aiming at the problem that hydropower units currently lack an effective fault symptom warning mechanism, the present invention combines feature detection with feature extraction algorithms in the field of machine learning to construct a unit vibration fault symptom indicator that fuses signal time domain information and frequency domain information for fault warning, thereby achieving more comprehensive unit status monitoring. The fault symptom indicator that comprehensively considers time-frequency information is more sensitive to fault reactions and can detect early fault signs of the unit, which is of great significance for timely and effective unit fault warning.

[0016] 2) The present invention adopts the Bootstrap method to sample the vibration degradation index samples under normal conditions, calculates the standard deviation of each Bootstrap subsample, obtains the overall standard deviation estimate based on the normal prior distribution of the degradation index, and then uses the Gaussian threshold method to determine the upper and lower limits of the degradation index. By judging whether the index exceeds the limit, the real-time warning of the unit vibration abnormality is realized, so that the warning threshold can be dynamically adjusted according to the unit status, thereby improving the sensitivity and accuracy of the warning.

[0017] 3) The present invention uses historical health data to establish a time-domain health model, and uses real-time monitoring data for verification and early warning, thereby achieving an effective combination of model-driven and data-driven. Through real-time monitoring and early warning, potential faults of the unit can be discovered and handled in a timely manner, which is conducive to improving the reliability of the unit operation.

[0018] 4) The present invention can perform real-time online fast-line diagnosis and identification of unit fault signs according to the unit's time-domain fault sign index value or frequency-domain fault sign index value, which is conducive to early detection and treatment of potential unit faults and improving the safety and stability of unit operation.

[0019] 5) The present invention adopts an incremental database and regularly uses the unit fault case data obtained by the method of the present invention to retrain the hypersphere model used to determine the time domain fault symptom indicator threshold and the frequency domain fault symptom indicator threshold, thereby improving the adaptability of the time domain fault symptom indicator threshold or the frequency domain fault symptom indicator threshold to the changes of the unit over time, being able to adapt to the latest changes in the operating conditions of the unit, and ensuring the accuracy and precision of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0021] Figure 1 The figure is a flow chart of a hydropower unit fault warning method according to an embodiment of the present invention.

[0022] Figure 2 Schematic diagram of the flow of time domain symptom indicators of a computer group in an embodiment of the present invention.

[0023] Figure 3 Schematic diagram of the flow of frequency domain symptom indicators of a computer group in an embodiment of the present invention.

[0024] Figure 4 The figure is a schematic diagram of a flow chart of performing unit fault warning according to the unit vibration degradation index in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] Embodiment 1 like Figure 1 As shown in FIG. 1 , a hydropower unit fault early warning method integrating time domain and frequency domain features includes: Step 1: Define the detection index, select the time domain parameter index with the largest detection index, use the historical operating condition data of the unit in a healthy state as the input sample, calculate the time domain characteristics of the input sample according to the time domain parameter index, and use the time domain characteristics as the output sample to construct a unit time domain health model based on the least squares support vector machine; and They are the vibration signals of the unit under two different operating conditions. An indicator variable of a certain time domain characteristic, and , They satisfy the normal distribution , , They are , The mean of Respectively , The standard deviation of but The larger it is, the better the distinction between two different operating states; make , then the difference signal obey The normal distribution of , the corresponding probability density function is: ; (1) In the formula , but exist The probability distribution function of the range is: ; (2) In the formula represents the cumulative probability; The recognition rate DR is defined as the degree of distinction between the two different operating states of the unit by the time domain feature index T, and the calculation formula is: ; (3) make: ; (4) Substituting into equation (1) and equation (3), using substitution integration, we get: ; (5) In the formula represents the detection index; visible, The larger the value, the stronger the signal time domain characteristics. The greater the sensitivity to different states of the unit, the higher the corresponding discrimination; The specific expression of the value is: ; (6).

[0026] The unit's time domain parameter indicators are screened through the detection index DI, and by selecting multiple time domain parameters for comparative analysis, the standard deviation and peak-to-peak value with the largest detection index value are selected as indicators of the unit's time domain characteristic sensitivity.

[0027] Construct a unit time domain health model based on least squares support vector machine, including: Given contains N The training set of group samples ,in , Respectively Group input and output samples, and , ; The method for support vector machine to solve regression classification problem is to construct a mathematical model in the following form: ; (7) In the formula , Respectively represent the input and output of the support vector machine; is a mapping function that maps low-dimensional input vectors to high-dimensional space, , are all constants; The least squares method is introduced into the support vector machine model to obtain the following least squares support vector machine model: ; (8) In the formula, is the weight vector matrix, represents the error, called the relaxation factor, is the error of the kth group of samples, is the regularization parameter, also known as the penalty factor; The constraints of the least squares support vector machine model are: ; (9) In the formula, is a nonlinear mapping function; The Lagrange equation is introduced to transform the optimization problem with constraints into an optimization problem without constraints, and the linear regression equation of the least squares support vector machine is obtained according to the Karush-Kuhn-Tucker condition: ; (10) Formula (10) transforms the regression and classification problem of the support vector machine into solving a linear equation. and The mapping function is obtained by the least squares method. The radial basis kernel function is selected.

[0028] Step 2: Input the operating condition data in the real-time monitoring sample of the unit into the unit time domain health model, use the time domain characteristic value output by the unit time domain health model as the time domain characteristic health value of the unit vibration signal at the current moment, calculate the relative error between the actual value of the time domain characteristic of the unit vibration signal and the time domain characteristic health value according to the time domain parameter index, and use the relative error as the unit's time domain fault symptom indicator, such as Figure 2 shown.

[0029] The unit historical operating condition sample data X(t) under normal operating conditions is used as input, and the time domain feature data Y(t) screened by the detection index is used as output. The unit state health model is constructed by the least squares support vector machine. The method is as follows: The unit time domain health model based on least squares support vector machine is: ; (11) ; In the formula , They represent the mapping functions of the least squares support vector machine corresponding to the standard deviation and peak-to-peak value of the time domain features respectively; , are the standard deviation and peak-to-peak health values ​​obtained according to the unit time-domain health model respectively; t Indicates time.

[0030] The real-time operating condition data of the unit is input into the time domain health model of the hydropower unit shown in formula (11), and the time domain characteristic health value under the current corresponding operating condition can be obtained. The relative difference between the actual value of the time domain characteristic of the vibration signal and the time domain characteristic health value is calculated, and after weighted calculation, it is used as the time domain fault symptom indicator of the unit vibration signal. The expression is: ; (12) In the formula, It is the fault symptom index of the unit vibration signal in the time domain. , They are the sign indicators corresponding to the standard deviation of time domain characteristics and the peak-to-peak value, , is the actual value of the standard deviation and peak-to-peak value of the unit vibration signal, , are the weight coefficients of the standard deviation and peak-to-peak value of the time domain characteristics respectively.

[0031] The calculation formula of the weight coefficient is: ; (13) ; (14) In the formula , are the weight coefficients of the standard deviation and peak-to-peak value of the time domain characteristics respectively.

[0032] Step 3: extract the wavelet singular value eigenvectors of the unit vibration signal under the same working condition, construct a wavelet singular value eigenvector set under the same working condition, and use the K-means clustering algorithm to calculate the center vector of the wavelet singular value eigenvector set as the standard feature of the unit vibration signal under the corresponding working condition; In the embodiment, a unit vibration signal feature extraction method combining a wavelet transform method and a singular value decomposition method is used to extract features of historical vibration signals in a healthy working state of the unit, and a wavelet singular value feature vector set containing different frequency components is obtained. A K-means algorithm is used to obtain a healthy cluster center C of the singular value feature vector set. The obtained singular value vector is the signal feature vector. The Euclidean distance between the real-time vibration signal wavelet singular value vector and the cluster center C is used as a frequency domain fault symptom indicator of the method, such as Figure 3 shown.

[0033] The specific process of step 3 is: Step 3.1: Decompose the vibration signal of the unit using the wavelet transform method to obtain a series of wavelet decomposition coefficients; In continuous wavelet transform, the calculation formula of wavelet signal is: ; (15) In the formula, represents the wavelet mother function, is the scale factor, is the translation factor, and Continuously changing in the real number domain, we get a continuous wavelet signal ; The calculation formula of wavelet decomposition coefficient is: ; (16) In the formula, is the wavelet decomposition coefficient; represents the basic wavelet function; , Indicates the upper and lower limits of the integral; represents the space of square integrable functions; Step 3.2: Perform singular value decomposition on the wavelet decomposition coefficients obtained in step 3.1 to obtain the wavelet singular value eigenvector; The wavelet decomposition coefficients are used as the input matrix A of the singular value decomposition. ,matrix The eigenvalue of ; The square root of That is the matrix The singular values ​​of For the matrix The conjugate transposed matrix of ; The formula for matrix singular value decomposition is: ; (17) ; In the formula is a diagonal matrix, Both are matrices The singular values ​​of ; U is a unitary matrix, , V is an orthogonal matrix, ; Transform formula (17) to obtain: ; (18) Step 3.3: Compare the real-time vibration signal wavelet singular value vector with the cluster center The Euclidean distance between them is used as the fault symptom index of the unit vibration signal frequency domain, and the calculation formula is: ; (19) In the formula, It is the fault symptom index of the unit vibration signal frequency domain. for The singular value eigenvector of the vibration signal measured at each moment; It is the time domain fault symptom indicator of the unit vibration signal.

[0034] Furthermore, in step 5, the unit vibration signal time domain fault symptom index and the unit vibration signal frequency domain fault symptom index are weightedly fused to obtain an overall time-frequency domain fault symptom index; The formula for weighted fusion of unit vibration signal time domain characteristics and unit vibration signal frequency domain characteristics is: ; (20) In the formula Indicates fault symptom indicators in the time-frequency domain; ;(twenty one) ;(twenty two) In the formula , are all weight coefficients.

[0035] Step 4: Extract the wavelet singular value eigenvector of the unit vibration signal in the real-time monitoring data sample of the unit, and calculate its Euclidean distance with the center vector obtained in step 3 as the frequency domain fault symptom indicator of the unit.

[0036] Step 5: Transform the frequency domain fault symptom index obtained in step 4. The transformed frequency domain fault symptom index and the time domain fault symptom index are kept identical. The transformed frequency domain fault symptom index and the time domain fault symptom index are fused to construct a fault symptom index of the time-frequency domain information of the unit vibration signal as the unit vibration degradation index.

[0037] Step 6: uniformly sample the vibration degradation index samples under normal conditions, calculate the standard deviation of each sample, and further calculate the standard deviation estimate of the overall sample of the unit vibration degradation index based on the normal prior distribution of the unit vibration degradation index. Use the Gaussian threshold method to determine the upper and lower limits of the unit vibration degradation index. By judging whether the real-time unit vibration degradation index is within the range of the upper and lower limits, a real-time warning of abnormal unit vibration is achieved, such as Figure 4 shown.

[0038] Step 6.1: Construct the normal operating state of the unit under similar working conditions Fault symptom indicator samples independent and identically distributed at each moment ;in Respectively represent the fault symptom indicator variables at the 1st, 2nd, …, mth moments; Step 6.2: Sample fault symptom indicators obtained in step 6.1 Sampling with replacement is performed, and the independent sampling capacity is Bootstrap Failure Indicator Subsample , , are all fault symptom indicator sub-sample variables obtained by sampling, repeated times; calculate the statistic of each Bootstrap fault symptom indicator subsample, that is, the mean, recorded as ; Step 6.3: Get the value obtained in step 6.2 Subsample statistics , calculate its standard deviation ; Under known normal conditions, the vibration fault symptom indicators of the unit under the same working condition are generally close to normal distribution ,in represents the mean of the unit vibration fault symptom index under the same working condition; then the sub-sample mean obtained by Bootstrap sampling satisfies the normal distribution , so the standard deviation of the overall sample of fault symptom indicators is ; Step 6.4: Use the threshold method to determine the upper and lower limits of the fault symptom indicators; According to the Gaussian threshold method, the probability that the value of the overall sample standard deviation of the fault symptom indicator falls in the interval (μ-3s, μ+3s) is 99.74%, where μ represents the mean. Therefore, μ+3s and μ-3s are respectively used as the upper and lower limits of normal operation of the monitored quantity. By judging whether the indicator exceeds the limit, real-time warning of abnormal unit vibration can be achieved.

[0039] Embodiment 2 The difference from Example 1 is that in this embodiment, the real-time time-domain fault symptom index value of the unit is compared with the time-domain fault symptom index threshold to achieve real-time online rapid diagnosis of unit fault symptoms, and the unit time-domain fault symptom index value of the unit failure case obtained by unit degradation index diagnosis and identification is used, and the time-domain fault symptom index threshold is determined using a hypersphere model.

[0040] The unit failure warning system of the embodiment includes: Time domain health value calculation module: constructs a time domain health model of the unit based on the least squares support vector machine, and uses the time domain health model of the unit to calculate the time domain characteristic health value of the unit under the current operating condition; Time domain index calculation module: used to calculate the relative error between the actual value of the time domain characteristic of the vibration signal of the computer group and the health value of the time domain characteristic, and obtain the time domain fault symptom index value of the unit; Time domain diagnosis module: sets the time domain fault symptom index threshold of the unit, compares the real-time unit time domain fault symptom index value obtained by the time domain index calculation module with the time domain fault symptom index threshold, and if the real-time unit time domain fault symptom index value exceeds the threshold, it is judged that the unit has a fault symptom, and sends a unit fault warning signal through the abnormal alarm module; Standard frequency domain feature module: Combine the wavelet transform method with the singular value decomposition method to extract the wavelet singular value feature vectors of the unit vibration signal under the same working condition, construct the wavelet singular value feature vector set under the same working condition, and calculate the center vector of the wavelet singular value feature vector set as the standard feature of the unit vibration signal under the corresponding working condition; Frequency domain index calculation module: obtains the wavelet singular value eigenvector of the unit vibration signal in the unit real-time monitoring sample, calculates the distance from the standard frequency domain feature of the unit vibration signal, and obtains the frequency domain fault symptom index value; Unit degradation index calculation module: integrates the frequency domain fault symptom index with the time domain fault symptom index to calculate the real-time unit vibration degradation index value; Unit degradation index threshold module: Based on the fault symptom index sample data under normal operation of the unit, the standard deviation of the overall sample of the fault symptom index is obtained using the sampling method, and the upper and lower limits of the fault symptom index are determined using the Gaussian threshold method; Abnormal alarm module: According to the upper and lower limits of the fault symptom index output by the unit degradation index threshold module, it is judged in real time whether the unit vibration degradation index value exceeds the limit. If the unit vibration degradation index value exceeds the upper and lower limits of the fault symptom index, a unit fault warning signal is issued, and the corresponding unit degradation index value, unit time domain fault symptom index value and unit frequency domain fault symptom index value are stored in the fault library as unit fault case data; or the judgment result of the time domain diagnosis module is that the unit has a fault symptom, and a unit fault warning signal is issued; Fault library: an incremental database is used to store the unit fault case data diagnosed by the method of the present invention, wherein the unit fault case data includes unit vibration signals, unit degradation index values, unit time domain fault symptom index values, and unit frequency domain fault symptom index values.

[0041] In the embodiment, the hypersphere model used to determine the time domain fault symptom indicator threshold is regularly retrained using the unit fault case data samples in the fault library, so as to realize the dynamic adaptive update of the time domain fault symptom indicator threshold.

[0042] Embodiment 3 The difference from Example 1 is that in this embodiment, a threshold value of the frequency domain fault symptom index of the unit is set, and the real-time frequency domain fault symptom index value of the unit is compared with the frequency domain fault symptom index threshold value to achieve real-time online rapid diagnosis of unit fault symptoms, and the unit frequency domain fault symptom index value of the unit fault case obtained by unit degradation index diagnosis and identification is used, and the frequency domain fault symptom index threshold value is determined using a hypersphere model.

[0043] The unit failure warning system of the embodiment includes: Time domain health value calculation module: constructs a time domain health model of the unit based on the least squares support vector machine, and uses the time domain health model of the unit to calculate the time domain characteristic health value of the unit under the current operating condition; Time domain index calculation module: used to calculate the relative error between the actual value of the time domain characteristic of the vibration signal of the computer group and the health value of the time domain characteristic, and obtain the time domain fault symptom index value of the unit; Standard frequency domain feature module: Combine the wavelet transform method with the singular value decomposition method to extract the wavelet singular value feature vectors of the unit vibration signal under the same working condition, construct the wavelet singular value feature vector set under the same working condition, and calculate the center vector of the wavelet singular value feature vector set as the standard feature of the unit vibration signal under the corresponding working condition; Frequency domain index calculation module: obtains the wavelet singular value eigenvector of the unit vibration signal in the unit real-time monitoring sample, calculates the distance from the standard frequency domain feature of the unit vibration signal, and obtains the frequency domain fault symptom index value; Frequency domain diagnosis module: sets the unit frequency domain fault symptom index threshold, compares the real-time unit frequency domain fault symptom index value obtained by the frequency domain index calculation module with the frequency domain fault symptom index threshold, and if the real-time unit frequency domain fault symptom index value exceeds the threshold, it is judged that the unit has fault signs, and sends a unit fault warning signal through the abnormal alarm module; Unit degradation index calculation module: integrates the frequency domain fault symptom index with the time domain fault symptom index to calculate the real-time unit vibration degradation index value; Unit degradation index threshold module: Based on the fault symptom index sample data under normal operation of the unit, the standard deviation of the overall sample of the fault symptom index is obtained using the sampling method, and the upper and lower limits of the fault symptom index are determined using the Gaussian threshold method; Abnormal alarm module: According to the upper and lower limits of the fault symptom index output by the unit degradation index threshold module, it is judged in real time whether the unit vibration degradation index value exceeds the limit. If the unit vibration degradation index value exceeds the upper and lower limits of the fault symptom index, a unit fault warning signal is issued, and the corresponding unit degradation index value, unit time domain fault symptom index value and unit frequency domain fault symptom index value are stored in the fault library as unit fault case data; or the judgment result of the time domain diagnosis module is that the unit has a fault symptom, and a unit fault warning signal is issued; Fault database: used to store unit fault case data, the unit fault case data includes unit vibration signals, unit degradation index values, unit time domain fault symptom index values ​​and unit frequency domain fault symptom index values.

[0044] In the embodiment, the hypersphere model used to determine the frequency domain fault symptom indicator threshold is regularly retrained using the unit fault case data samples in the fault library, so as to realize the dynamic adaptive update of the frequency domain fault symptom indicator threshold.

Claims

1. A hydropower unit fault early warning method integrating time domain and frequency domain features, characterized in that: The following steps are involved: Step 1: Select a time domain parameter index, take the historical operating condition data under the unit health state as the input sample, calculate the time domain characteristics of the input sample according to the time domain parameter index, and use the time domain characteristics as the output sample to construct a unit time domain health model based on the least squares support vector machine; Step 2: Input the operating condition data in the real-time monitoring sample of the unit into the unit time domain health model, use the time domain characteristic value output by the unit time domain health model as the time domain characteristic health value of the unit vibration signal at the current moment, calculate the relative error between the actual value of the time domain characteristic of the unit vibration signal and the time domain characteristic health value according to the time domain parameter index, use the relative error as the unit's time domain fault symptom index, and use the time domain fault symptom index to perform preliminary unit fault symptom diagnosis and abnormality warning; Step 3: extract the wavelet singular value eigenvectors of the unit vibration signal under the same working condition, construct a wavelet singular value eigenvector set under the same working condition, and use a clustering algorithm to calculate the center vector of the wavelet singular value eigenvector set as the standard feature of the unit vibration signal under the corresponding working condition; Step 4: extract the wavelet singular value eigenvector of the unit vibration signal in the real-time monitoring data sample of the unit, and calculate its Euclidean distance with the center vector obtained in step 3 as the frequency domain fault symptom indicator of the unit; Step 5: transform the frequency domain fault symptom index obtained in step 4, keep the transformed frequency domain fault symptom index identical to the time domain fault symptom index, fuse the transformed frequency domain fault symptom index with the time domain fault symptom index, and construct the fault symptom index of the unit vibration signal time-frequency domain information as the unit vibration degradation index; Step 6: Uniformly sample the vibration degradation index samples under normal conditions, calculate the standard deviation of each sample, and further calculate the standard deviation estimate of the overall sample of the unit vibration degradation index based on the normal prior distribution of the unit vibration degradation index. Use the Gaussian threshold method to determine the upper and lower limits of the unit vibration degradation index. By judging whether the real-time unit vibration degradation index is within the range of the upper and lower limits, a real-time warning of unit vibration abnormality is achieved.

2. The hydropower unit fault early warning method integrating time domain and frequency domain features according to claim 1 is characterized in that: In step 1, the detection index in mathematical statistics is introduced as an indicator to measure the sensitivity of the time domain characteristics of the unit vibration signal, which specifically includes: Assumptions and They are the vibration signals of the unit under two different operating conditions. An indicator variable of a certain time domain characteristic, and , They satisfy the normal distribution , , They are , The mean of Respectively , The standard deviation of but The larger it is, the better the distinction between two different operating states; make , then the difference signal obey The normal distribution of , the corresponding probability density function is: ；（1） In the formula ,but exist The probability distribution function of the range is: ；（2） In the formula represents the cumulative probability; The recognition rate DR is defined as the degree of distinction between the two different operating states of the unit by the time domain feature index T, and the calculation formula is: ；（3） make: ；（4） In the formula is an intermediate variable; Substituting into equation (1) and equation (3), using substitution integration, we get: ；（5） In the formula represents the detection index; Detection Index The expression is: ；（6）。 3. The hydropower unit fault early warning method integrating time domain and frequency domain features according to claim 2 is characterized in that: In step 1, the unit time domain parameter index is screened by the detection index DI, and by selecting multiple time domain parameters for comparative analysis, the standard deviation and peak-to-peak value with the largest detection index value are selected as indicators of the unit's time domain characteristic sensitivity.

4. The hydropower unit fault early warning method integrating time domain and frequency domain features according to claim 3 is characterized in that: In step 1, the construction of a unit time domain health model based on a least squares support vector machine specifically includes: Given contains N The training set of group samples ,in , Respectively Group input and output samples; Based on the support vector machine, the following mathematical model is constructed: ；（7） In the formula , Respectively represent the input and output of the support vector machine; is a mapping function that maps low-dimensional input vectors to high-dimensional space, , are all constants; The least squares method is introduced into the support vector machine model to obtain the following least squares support vector machine model: ；（8） In the formula, is the weight vector matrix, Indicates the error, is the error of the kth group of samples, is the regularization parameter; The constraints of the least squares support vector machine model are: ；（9） In the formula, is a nonlinear mapping function; The Lagrange equation is introduced to transform the optimization problem with constraints into an optimization problem without constraints, and the linear regression equation of the least squares support vector machine is obtained according to the Karush-Kuhn-Tucker condition: ；（10） Formula (10) transforms the regression and classification problem of the support vector machine into solving a linear equation. and The mapping function is obtained by the least squares method. The radial basis kernel function is selected.

5. The hydropower unit fault early warning method integrating time domain and frequency domain features according to claim 4 is characterized in that: In step 2, the historical operating condition sample data X(t) of the unit under normal operating conditions is used as input, and the time domain feature data Y(t) screened by the detection index is used as output, and the unit state health model is constructed by the least squares support vector machine; The unit time domain health model of the least squares support vector machine is: ；（11） ； In the formula , They represent the mapping functions of the least squares support vector machine corresponding to the standard deviation and peak-to-peak value of the time domain features respectively; , are the standard deviation and peak-to-peak health values ​​obtained according to the unit time-domain health model respectively; t represents time; The real-time operating condition data of the unit is input into the time domain health model of the hydropower unit shown in formula (11), and the time domain characteristic health value under the current corresponding operating condition can be obtained. The relative difference between the actual value of the time domain characteristic of the vibration signal and the time domain characteristic health value is calculated, and after weighted calculation, it is used as the time domain fault symptom indicator of the unit vibration signal. The expression is: ；（12） In the formula, It is the fault symptom index of the unit vibration signal in the time domain. , They are the sign indicators corresponding to the standard deviation of time domain characteristics and the peak-to-peak value, , is the actual value of the standard deviation and peak-to-peak value of the unit vibration signal, , are the weight coefficients of the standard deviation and peak-to-peak value of the time domain characteristics respectively.

6. The hydropower unit fault early warning method integrating time domain and frequency domain features according to claim 5 is characterized in that: In step 2, the calculation formula of the weight coefficient is: ；（13） ； （14） In the formula , are the weight coefficients of the standard deviation and peak-to-peak value of the time domain characteristics respectively.

7. The hydropower unit fault early warning method integrating time domain and frequency domain features according to claim 1, 2, 3, 4, 5 or 6 is characterized in that: The step 3 specifically includes the following sub-steps: Step 3.1: Decompose the vibration signal of the unit using the wavelet transform method to obtain a series of wavelet decomposition coefficients; In continuous wavelet transform, the calculation formula of wavelet signal is: ； （15） In the formula, represents the wavelet mother function, is the scale factor, is the translation factor, and Continuously changing in the real number domain, we get a continuous wavelet signal ; The calculation formula of wavelet decomposition coefficient is: ；（16） In the formula, is the wavelet decomposition coefficient; represents the basic wavelet function; , Indicates the upper and lower limits of the integral; Step 3.2: Perform singular value decomposition on the wavelet decomposition coefficients obtained in step 3.1 to obtain the wavelet singular value eigenvector; The wavelet decomposition coefficients are used as the input matrix A of the singular value decomposition, the matrix The eigenvalue of ; The square root of That is the matrix The singular values ​​of For the matrix The conjugate transposed matrix of ; The formula for matrix singular value decomposition is: ；（17） ； In the formula is a diagonal matrix, Both are matrices The singular values ​​of ; U is a unitary matrix, V is an orthogonal matrix; Transform formula (17) to obtain: ；（18） Step 3.3: Compare the real-time vibration signal wavelet singular value vector with the cluster center The Euclidean distance between them is used as the fault symptom index of the unit vibration signal frequency domain, and the calculation formula is: ；（19） In the formula, It is the fault symptom index of the unit vibration signal frequency domain. for The singular value eigenvector of the vibration signal measured at each moment; It is the time domain fault symptom indicator of the unit vibration signal.

8. The hydropower unit fault early warning method integrating time domain and frequency domain features according to claim 7 is characterized in that: In the step 5, the unit vibration signal time domain fault symptom index and the unit vibration signal frequency domain fault symptom index are weightedly fused to obtain an overall time-frequency domain fault symptom index; The formula for weighted fusion of unit vibration signal time domain characteristics and unit vibration signal frequency domain characteristics is: ； （20） In the formula Indicates fault symptom indicators in the time-frequency domain; ；（21） ；（22） In the formula , are all weight coefficients.

9. The hydropower unit fault early warning method integrating time domain and frequency domain features according to claim 8 is characterized in that: The step 6 specifically includes: Step 6.1: Construct the normal operating state of the unit under similar working conditions Fault symptom indicator samples independent and identically distributed at each moment ;in Respectively represent the fault symptom indicator variables at the 1st, 2nd, …, mth moments; Step 6.2: Sample fault symptom indicators obtained in step 6.1 Sampling with replacement is performed, and the independent sampling capacity is Bootstrap Failure Indicator Subsample , , are all fault symptom indicator subsample variables obtained by sampling, repeated times; calculate the statistic of each Bootstrap fault symptom indicator subsample, that is, the mean, recorded as ; Step 6.3: Get the value obtained in step 6.2 Subsample statistics , calculate its standard deviation ; Under known normal conditions, the vibration fault symptom indicators of the unit under the same working condition are generally close to normal distribution ,in represents the mean of the unit vibration fault symptom index under the same working condition; then the sub-sample mean obtained by Bootstrap sampling satisfies the normal distribution , so the standard deviation of the overall sample of fault symptom indicators is ; Step 6.4: Use the threshold method to determine the upper and lower limits of the fault symptom indicators; According to the Gaussian threshold method, the probability that the value of the overall sample standard deviation of the fault symptom indicator falls in the interval (μ-3s, μ+3s) is 99.74%, where μ represents the mean. Therefore, μ+3s and μ-3s are respectively used as the upper and lower limits of normal operation of the monitored quantity. By judging whether the indicator exceeds the limit, real-time warning of abnormal unit vibration can be achieved.

10. The system of the hydropower unit fault early warning method according to claim 1 or 2 or 3 or 5 or 6 or 8 or 9, characterized in that: The system comprises: Time domain health value calculation module: constructs a time domain health model of the unit based on the least squares support vector machine, and uses the time domain health model of the unit to calculate the time domain characteristic health value of the unit under the current operating condition; Time domain index calculation module: used to calculate the relative error between the actual value of the time domain characteristic of the vibration signal of the computer group and the health value of the time domain characteristic, and obtain the time domain fault symptom index value of the unit; Time domain diagnosis module: sets the time domain fault symptom index threshold of the unit, compares the real-time unit time domain fault symptom index value obtained by the time domain index calculation module with the time domain fault symptom index threshold, and if the real-time unit time domain fault symptom index value exceeds the threshold, it is judged that the unit has a fault symptom, and sends a unit fault warning signal through the abnormal alarm module; Standard frequency domain feature module: Combine the wavelet transform method with the singular value decomposition method to extract the wavelet singular value feature vectors of the unit vibration signal under the same working condition, construct the wavelet singular value feature vector set under the same working condition, and calculate the center vector of the wavelet singular value feature vector set as the standard feature of the unit vibration signal under the corresponding working condition; Frequency domain index calculation module: obtains the wavelet singular value eigenvector of the unit vibration signal in the unit real-time monitoring sample, calculates the distance from the standard frequency domain feature of the unit vibration signal, and obtains the frequency domain fault symptom index value; Unit degradation index calculation module: integrates the frequency domain fault symptom index with the time domain fault symptom index to calculate the real-time unit vibration degradation index value; Unit degradation index threshold module: Based on the fault symptom index sample data under normal operation of the unit, the standard deviation of the overall sample of the fault symptom index is obtained using the sampling method, and the upper and lower limits of the fault symptom index are determined using the Gaussian threshold method; Abnormal alarm module: According to the upper and lower limits of the fault symptom index output by the unit degradation index threshold module, it is judged in real time whether the unit vibration degradation index value exceeds the limit. If the unit vibration degradation index value exceeds the upper and lower limits of the fault symptom index, a unit fault warning signal is issued, and the corresponding unit degradation index value, unit time domain fault symptom index value and unit frequency domain fault symptom index value are stored in the fault library as unit fault case data; or the judgment result of the time domain diagnosis module is that the unit has a fault symptom, and a unit fault warning signal is issued; Fault database: used to store unit fault case data, the unit fault case data includes unit vibration signals, unit degradation index values, unit time domain fault symptom index values ​​and unit frequency domain fault symptom index values.

Images