Comprehensive degradation evaluation method, system and equipment of hydroelectric generating set shaft system and medium

Abstract

The invention provides a comprehensive degradation evaluation method, system and equipment for a shaft system of a hydroelectric generating set and a medium. The method comprises the following steps: acquiring a vibration signal health sample set of each key measuring point of the shaft system in a healthy operation process of the hydroelectric generating set, and preprocessing the vibration signal health sample set; a vibration signal multi-dimensional feature vector fusing typical time domain features, frequency domain features, multi-scale attention entropy and SDAE model depth features is constructed, a Mean-Shift density clustering method is combined to construct performance degradation degree indexes of all key measuring points, weighted summation is carried out on the performance degradation degree indexes of all the key measuring points, and the comprehensive degradation index of the hydroelectric generating set shafting can be obtained. Potential information of the vibration signals is comprehensively excavated, degradation information of multiple key measuring points is fused to construct the comprehensive degradation index of the hydroelectric generating set shaft system, more comprehensive description of the operation state of the hydroelectric generating set is achieved, operation and maintenance personnel can master the operation condition of the hydroelectric generating set in real time, and safe and stable operation of the hydroelectric generating set is effectively guaranteed.

Description

Technical Field

[0001] The invention belongs to the technical field of electric power engineering, and relates to an evaluation method, system, equipment and medium for a hydropower unit. Background Art

[0002] As a key equipment for hydropower energy conversion in power systems, the safety and stability of hydropower units are very important. With the development of hydropower in power systems, the component structure of hydropower units is becoming more and more complex, and the amount of status monitoring information of hydropower units is constantly expanding, which puts higher requirements on the safety of hydropower units. Mastering the real-time operating status of hydropower units and timely discovering signs of faults are of great significance to ensuring the safety of hydropower units and improving the comprehensive benefits of power systems.

[0003] The shaft system of a hydropower unit contains several guide bearings, which are key components that affect the operating stability of the hydropower unit. The swing monitoring data contains a large amount of important information that characterizes the operating status of the shaft system. When the hydropower unit is operating abnormally or fails, the vibration signal of the swing monitoring data will show corresponding dynamic mutation behavior. By analyzing the swing monitoring data and constructing a degradation index that effectively characterizes the operating status of the shaft system of the hydropower unit, it is helpful to timely grasp the operating status of the power station unit. However, in the currently constructed hydropower unit equipment degradation assessment model, the vibration signal feature extraction usually processes and analyzes the sequence from a single perspective, which makes it difficult to fully explore the potential information of the vibration signal and effectively ensure the safe and stable operation of the hydropower unit. Summary of the invention

[0004] In order to solve the problems described in the background technology that the currently constructed degradation assessment of hydropower unit equipment is difficult to fully explore the potential information of vibration signals and the safe and stable operation of hydropower units is difficult to effectively ensure, the present invention provides a comprehensive degradation assessment method, system, equipment and medium for the shaft system of a hydropower unit.

[0005] The method of the present invention comprises:

[0006] Obtain a healthy sample set of vibration signals of key measuring points of the shaft system of the hydropower unit under stable operating conditions and during the healthy operation of the hydropower unit;

[0007] The healthy sample set of vibration signals is preprocessed by using complementary set empirical mode decomposition method and wavelet threshold denoising method to obtain the preprocessed healthy sample set of vibration signals.

[0008] The preprocessed vibration signal healthy sample set is analyzed to obtain typical time domain features, typical frequency domain features and multi-scale attention entropy features. The unsupervised features in the preprocessed vibration signal healthy sample set are mined through stacked denoising autoencoders. The multi-dimensional fusion feature set of healthy samples of key measuring points of the hydropower unit shaft system is established by combining typical time domain features, typical frequency domain features, multi-scale attention entropy features and unsupervised features.

[0009] Based on the Mean-Shift density clustering method, the cluster center of the multi-dimensional fusion feature set of the health samples of each key measuring point of the hydropower unit shaft system is calculated, and the cluster center is used as the health benchmark feature vector of each key measuring point.

[0010] The vibration signals of each key measuring point of the shaft system during the operation of the hydropower unit are collected in real time, and the multi-dimensional fusion feature vectors of the vibration signals of each key measuring point of the shaft system collected in real time are extracted. The relative Euclidean distance between the multi-dimensional fusion feature vectors of the vibration signals of each key measuring point of the shaft system collected in real time and the corresponding healthy reference feature vectors is calculated as the performance degradation index of each key measuring point of the shaft system of the hydropower unit;

[0011] The performance degradation index of each key measuring point of the hydropower unit shaft system is weighted and summed to calculate the comprehensive degradation index of the hydropower unit shaft system. The comprehensive degradation index of the hydropower unit shaft system is compared with the health limit of the comprehensive degradation assessment index to obtain the comprehensive degradation assessment result of the hydropower unit shaft system.

[0012] Furthermore, the preprocessing method of the vibration signal healthy sample set includes:

[0013] For the healthy sample set of vibration signals of key measuring points of the shaft system of the hydropower unit during healthy operation, the complementary set empirical mode decomposition method is used to decompose the vibration signal step by step into several relatively stable intrinsic mode functions IMF and a standard residual.

[0014] The mutual correlation coefficient between each IMF component and the original vibration signal is calculated by formula (1) to determine the noise component and effective component in the intrinsic mode function IMF. Formula (1) is as follows:

[0015]

[0016] Where i represents the order number of the intrinsic mode function IMF, and IMF i represents the i-th order eigenmode function, and y represents the original vibration signal;

[0017] The wavelet threshold denoising method is used to denoise the noise component, and the noise component after denoising is superimposed with the effective component and in the original vibration signal to obtain the preprocessed vibration signal healthy sample set.

[0018] Furthermore, the typical time domain features and the typical frequency domain features are obtained by extracting vibration signals based on the amplitude domain analysis method and Fourier transform theory;

[0019] The typical time domain features include: standard deviation T1, peak-to-peak value T2, kurtosis T3, skewness T4 and root mean square value T5;

[0020] The typical frequency domain features include: frequency amplitude mean F1, frequency centroid value F2, frequency standard deviation F3, frequency variation coefficient F4 and frequency slope F5;

[0021] The method for obtaining the multi-scale attention entropy feature is:

[0022] For the preprocessed vibration signal healthy sample set, given a time series signal X with a length of N = {x1, x2, ..., x N}, and then the coarse-grained time series is obtained by using formula (2), namely:

[0023]

[0024] Where τ is the scale factor, ranging from 1 to τ max , τ max is the maximum scale factor; Y i (τ) is a coarse-grained time series at scale τ; when τ = 1, the coarse-grained component is the original signal; when τ > 1, the original signal is divided into multiple coarse-grained sequences of length [N / τ] after coarse-grained processing;

[0025] Compute the coarse-grained time series Y at scale τ i (τ) The attention entropy value is defined as MATE(X,τ) of the time series signal X at this scale:

[0026] MATE(X,τ)=ATE(Y i (τ) )(3),

[0027] In the formula, ATE(Y i (τ) ) represents the sequence Y i (τ) The attention entropy value of

[0028] Repeat the calculation until τ = τ max , the multi-scale attention entropy of the signal is the set of MATE(X,τ), that is:

[0029] MATE={MATE(X,τ)},τ=1,2,...,τ max (4);

[0030] The unsupervised feature mining method includes:

[0031] Multiple denoising autoencoders are stacked in sequence to form a stacked denoising autoencoder. Each denoising autoencoder is trained independently in a bottom-up order. The lower denoising autoencoder uses the implicit representation of the denoising autoencoder trained in the upper layer as a pure input, and is trained layer by layer until the highest hidden layer outputs the deepest high-order features of the original input signal, that is, unsupervised features.

[0032] Given an input vector X, the denoising autoencoder transforms by random mapping Add random noise to the input data to get "contaminated" data q D is represented as a binomial random hidden noise; therefore, the encoded output H of the denoising autoencoder and the reconstructed decoded output Z are:

[0033]

[0034] Z = g(H) = s(W'H + b2) (6),

[0035] Where W∈R m×n , W'∈R n×m are the connection weights of the encoding network and the decoding network respectively; b1∈R m×1 , b2∈R n ×1 are the biases of the encoding network and the decoding network respectively; s(x) is the activation function, which uses the Sigmoid function, that is, s(x) = 1 / (1+e -x );

[0036] During the denoising autoencoder training process, the model weights and biases are iteratively updated through the back-propagation algorithm to minimize the reconstruction error between the reconstructed output Y and the original noise-free input X;

[0037] Normalize various feature indicators, combine typical time domain features (T1, T2, ..., T5), typical frequency domain features (F1, F2, ..., F5), multi-scale attention entropy features (M1, M2, ..., M n ) and unsupervised features (S1,S2,...,S m ), construct the multi-dimensional fusion feature vector of the vibration signal of the hydropower unit (T1,T2,...,T5,F1,F2,...,F5,M1,M2,...,M n ,S1,S2,...,S m ), and then establish a multi-dimensional fusion feature set of healthy samples of key measuring points of the hydropower unit shaft system.

[0038] Furthermore, the method for obtaining the health benchmark feature vector of each key measuring point is:

[0039] The Mean-Shift density clustering method is used to calculate the drift vector of the center point according to the change in data density in the region, and then the center is moved and the next iteration is started until it converges to the point with the maximum probability density. The calculation process is as follows:

[0040] Assume that there are n data points in d-dimensional Euclidean space constituting a data set {x i , i = 1, 2, ..., n}, then the kernel density estimation function at x is expressed by the kernel function K(x) and the bandwidth h:

[0041]

[0042] Express the kernel density estimation function in a radially symmetric form:

[0043] K(x)=c k,d k(||x 2 ) (8),

[0044] In the formula, c k,d is the normalization coefficient; k(x) is the profile function of K(x);

[0045] Substituting formula (8) into formula (7), we get:

[0046]

[0047] Define g(x) = -k'(x), then the kernel function corresponding to the profile function g(x) is G(x) = cg,dg(‖x‖ 2 ), c g,d is the normalization coefficient, is the probability density estimated by the kernel function G(x);

[0048] The derivative is The gradient of is:

[0049]

[0050] Assume y j (j=1,2,…) is the result of the jth iteration. In the process of mean shift, y j+1 With y j The relationship is:

[0051]

[0052] This cycle is repeated until the convergence condition is met, such as y j+1 -y jBelow a certain threshold or reaching a maximum number of iterations to find the local maximum point of the probability density function, thereby identifying the cluster centers in the data.

[0053] Furthermore, the calculation method of the performance degradation index of each key measuring point is:

[0054] Firstly, the vibration signals of each key measuring point of the shaft system are collected during the operation of the hydropower unit, and the multi-dimensional fusion feature vector R(t) of the vibration signals of each key measuring point of the shaft system collected in real time is extracted;

[0055] The extraction method of the multi-dimensional fusion feature vector R(t) is as follows: based on the real-time collected vibration signals of each key measuring point of the shaft system, the typical time domain features (T1', T2', ..., T5'), the typical frequency domain features (F1', F2', ..., F5'), and the multi-scale attention entropy features (M1', M2', ..., M n ') and unsupervised features (S1',S2',...,S m '), construct the multi-dimensional fusion feature vector R(t)=(T1',T2',...,T5',F1',F2',...,F5',M1',M2',...,M n ',S1',S2',...,S m ');

[0056] The relative Euclidean distance is used to measure the deviation between the multi-dimensional fusion feature vector R(t) of the vibration signal of each key measuring point of the shaft system collected in real time and the corresponding healthy benchmark feature vector C, which is used as the performance degradation index H(t) of each key measuring point of the shaft system. The calculation formula is as follows:

[0057]

[0058] Furthermore, the calculation method of the comprehensive degradation index of the shaft system of the hydropower unit is:

[0059] The performance degradation index of each key measuring point is weighted and summed to obtain the comprehensive degradation index HI of the shaft system of the hydropower unit. The formula is:

[0060] HI=ω1H1+ω2H2+...+ω z H z (13)

[0061] Where H1, H2, ..., H z are the performance degradation indexes of the key measuring points of the 1st to Zth axis systems, ω1, ω2, ..., ω z It represents the weight of the performance degradation index of the corresponding measuring point, which is determined based on the power system's evaluation of the importance of each key measuring point of the hydropower unit shaft system.

[0062] The present invention also proposes a comprehensive degradation assessment system for the shaft system of a hydropower unit, including a healthy sample set acquisition module, a data preprocessing module, a healthy sample multi-dimensional fusion feature set establishment module, a healthy baseline feature vector calculation module, a performance degradation index calculation module, and a comprehensive degradation assessment module.

[0063] The healthy sample set acquisition module is used to acquire healthy sample sets of vibration signals of key measuring points of the shaft system of the hydropower unit under stable operating conditions and during the healthy operation of the hydropower unit.

[0064] The data preprocessing module uses a complementary set empirical mode decomposition method and a wavelet threshold noise reduction method to preprocess the vibration signal healthy sample set to obtain a preprocessed vibration signal healthy sample set.

[0065] The healthy sample multi-dimensional fusion feature set establishment module is used to analyze the pre-processed vibration signal healthy sample set, obtain typical time domain features, typical frequency domain features and multi-scale attention entropy features, and mine the unsupervised features in the pre-processed vibration signal healthy sample set through stacked denoising autoencoders. Combined with the typical time domain features, typical frequency domain features, multi-scale attention entropy features and unsupervised features, a multi-dimensional fusion feature set of healthy samples for key measuring points of the hydropower unit shaft system is established.

[0066] The health benchmark feature vector calculation module calculates the cluster center of the multidimensional fusion feature set of health samples of each key measuring point of the hydropower unit shaft system based on the Mean-Shift density clustering method, and uses the cluster center as the health benchmark feature vector of each key measuring point.

[0067] The performance degradation index calculation module is used to collect vibration signals of each key measuring point of the shaft system during the operation of the hydropower unit in real time, extract the multi-dimensional fusion feature vector of the vibration signals of each key measuring point of the shaft system collected in real time, and calculate the relative Euclidean distance between the multi-dimensional fusion feature vector of the vibration signals of each key measuring point of the shaft system collected in real time and the corresponding healthy benchmark feature vector as the performance degradation index of each key measuring point of the shaft system of the hydropower unit.

[0068] The comprehensive degradation assessment module is used to perform weighted summation on the performance degradation index of each key measuring point of the hydropower unit shaft system, calculate the comprehensive degradation index of the hydropower unit shaft system, compare the comprehensive degradation index of the hydropower unit shaft system with the health limit of the comprehensive degradation assessment index, and obtain the comprehensive degradation assessment result of the hydropower unit shaft system.

[0069] The present invention also proposes an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor implements the comprehensive degradation assessment method for the shaft system of a hydropower unit as described above by executing the computer instructions.

[0070] The present invention also proposes an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor implements the comprehensive degradation assessment method for the shaft system of a hydropower unit as described above by executing the computer instructions.

[0071] Compared with the prior art, the present invention first obtains a healthy sample set of vibration signals of each key measuring point of the shaft system of the hydropower unit during healthy operation, and pre-processes it by using a complementary set empirical mode decomposition method and a wavelet threshold noise reduction method, and mines the implicit information reflecting the state of the hydropower unit in the vibration signals of each key measuring point of the shaft system of the hydropower unit from multiple angles, constructs a multidimensional fusion feature set of healthy samples of each key measuring point of the shaft system of the hydropower unit that integrates typical time domain features, typical frequency domain features, multi-scale attention entropy features and unsupervised features, and calculates the healthy benchmark feature vector of each key measuring point based on the Mean-Shift density clustering method; by calculating the relative Euclidean distance between the multidimensional fusion feature vector of the vibration signal of each key measuring point of the shaft system collected in real time and the healthy benchmark feature vector, the performance degradation index of each key measuring point of the shaft system of the hydropower unit is calculated, and the performance degradation index of each key measuring point is weighted and summed to obtain the comprehensive degradation index of the shaft system of the hydropower unit.

[0072] The beneficial effects of the present invention are as follows: First, in view of the problem that the nonlinear non-stationary characteristics and redundant noises that commonly exist in the vibration signals of hydropower units affect accurate feature extraction, based on the signal processing capability and anti-modal aliasing advantages of the complementary set empirical mode decomposition method, combined with the wavelet threshold noise reduction method, the interference noise of the vibration signal is removed, and data support is provided for the subsequent construction of a comprehensive degradation index that effectively quantifies the operating status of the unit shaft system; Second, for the hydropower unit of a highly complex system including water, machinery and electricity, the implicit information reflecting the state of the hydropower unit is mined from multiple angles, the potential information of the vibration signal is fully mined, and each key measuring point of the hydropower unit is collected and analyzed, solving the limitation problem of relying on the vibration signal data of a single measuring point to describe the overall operating status of the shaft system; Third, the performance degradation index of each key measuring point is weighted and summed, and the degradation information of multiple key measuring points is integrated to construct a comprehensive degradation index of the hydropower unit shaft system, so as to achieve a more comprehensive description of the operating status of the hydropower unit, which is helpful for the operation and maintenance personnel to grasp the operating status of the hydropower unit in real time, and effectively ensure the safe and stable operation of the hydropower unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 The figure is a flow chart of the method of the present invention.

[0074] Figure 2 It is a data preprocessing flow chart of the present invention.

[0075] Figure 3 It is a schematic diagram of the calculation flow of the performance degradation index of the present invention.

[0076] Figure 4 It is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0077] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0078] The comprehensive degradation assessment method of the shaft system of a hydropower unit is shown in the flow chart below: Figure 1 As shown, the specific steps are described as follows.

[0079] Firstly, a healthy sample set of vibration signals of key measuring points of the shaft system of the hydropower unit is obtained during the healthy operation of the hydropower unit under stable operating conditions.

[0080] Next, the complementary set empirical mode decomposition method and wavelet threshold denoising method are used to preprocess the healthy sample set of vibration signals to obtain the preprocessed healthy sample set of vibration signals. Figure 2 As shown in the figure, complementary ensemble empirical mode decomposition is abbreviated as CEEMD in English.

[0081] Specifically, for the healthy sample set of vibration signals of key measuring points of the shaft system of the hydropower unit during healthy operation, the complementary set empirical mode decomposition method is used to decompose the vibration signal step by step into several relatively stable intrinsic mode functions (IMFs) and a standard residual.

[0082] The mutual correlation coefficient between each IMF component and the original vibration signal is calculated by formula (1) to determine the noise component and effective component in the intrinsic mode function IMF. Formula (1) is as follows:

[0083]

[0084] Where i represents the order number of the intrinsic mode function IMF, MF i represents the i-th order eigenmode function, and y represents the original vibration signal.

[0085] For the noise component and effective component, it can be stipulated that: when R≤0.2, it is an effective component; when R>0.2, it is a noise component.

[0086] The wavelet threshold denoising method is used to denoise the noise component, and the noise component after denoising is superimposed with the effective component and in the original vibration signal to obtain the preprocessed vibration signal healthy sample set.

[0087] Then, the preprocessed vibration signal healthy sample set is analyzed to obtain typical time domain features, typical frequency domain features and multi-scale attention entropy features. The unsupervised features in the preprocessed vibration signal healthy sample set are mined by stacking denoising autoencoders. Combined with typical time domain features, typical frequency domain features, multi-scale attention entropy features and unsupervised features, a multi-dimensional fusion feature set of healthy samples of key measuring points of the hydropower unit shaft system is established.

[0088] Specifically, typical time domain features and typical frequency domain features of the vibration signal are extracted based on the amplitude domain analysis method and Fourier transform theory.

[0089] Typical time domain characteristics include: standard deviation T1, peak-to-peak value T2, kurtosis T3, skewness T4 and root mean square value T5.

[0090] Typical frequency domain features include: frequency amplitude mean F1, frequency centroid value F2, frequency standard deviation F3, frequency variation coefficient F4 and frequency slope F5.

[0091] The method for obtaining multi-scale attention entropy features is:

[0092] For the preprocessed vibration signal healthy sample set, given a time series signal X with a length of N = {x1, x2, ..., x N}, and then the coarse-grained time series is obtained by using formula (2), namely:

[0093]

[0094] Where τ is the scale factor, ranging from 1 to τ max , τ max is the maximum scale factor; Y i (τ) is a coarse-grained time series at scale τ; when τ = 1, the coarse-grained component is the original signal; when τ > 1, the original signal is divided into multiple coarse-grained sequences of length [N / τ] after coarse-grained processing;

[0095] Compute the coarse-grained time series Y at scale τ i (τ) The attention entropy value is defined as MATE(X,τ) of the time series signal X at this scale:

[0096] MATE(X,τ)=ATE(Y i (τ) )(3),

[0097] In the formula, ATE(Y i (τ) ) represents the sequence Y i (τ) The attention entropy value of

[0098] Repeat the calculation until τ = τ max , the multi-scale attention entropy of the signal is the set of MATE(X,τ), that is:

[0099] MATE={MATE(X,τ)},τ=1,2,...,τ max (4).

[0100] Unsupervised feature mining methods include:

[0101] Multiple denoising autoencoders are stacked in sequence to form a stacked denoising autoencoder. Each denoising autoencoder is trained independently in a bottom-up order. The lower denoising autoencoder uses the implicit representation of the denoising autoencoder trained in the upper layer as a pure input, and trains layer by layer until the highest hidden layer outputs the deepest high-order features of the original input signal, that is, unsupervised features. It should be noted that unsupervised features refer to signal features extracted by stacking denoising autoencoders, that is, stacking denoising autoencoder features, which belong to a category of unsupervised features.

[0102] Given an input vector X, the denoising autoencoder transforms by random mapping Add random noise to the input data to get "contaminated" data q D is represented as a binomial random hidden noise; therefore, the encoded output H of the denoising autoencoder and the reconstructed decoded output Z are:

[0103]

[0104] Z = g(H) = s(W'H + b2) (6),

[0105] Where W∈R m×n , W'∈R n×m are the connection weights of the encoding network and the decoding network respectively; b1∈R m×1 , b2∈R n ×1 are the biases of the encoding network and the decoding network respectively; s(x) is the activation function, which uses the Sigmoid function, that is, s(x) = 1 / (1+e -x ).

[0106] During the denoising autoencoder training process, the model weights and biases are iteratively updated through the back-propagation algorithm to minimize the reconstruction error between the reconstructed output Y and the original noise-free input X.

[0107] Normalize various feature indicators, combine typical time domain features (T1, T2, ..., T5), typical frequency domain features (F1, F2, ..., F5), multi-scale attention entropy features (M1, M2, ..., M n ) and unsupervised features (S1,S2,...,S m ), construct the multi-dimensional fusion feature vector of the vibration signal of the hydropower unit (T1,T2,...,T5,F1,F2,...,F5,M1,M2,...,M n ,S1,S2,...,S m ), and then establish a multi-dimensional fusion feature set of healthy samples of key measuring points of the hydropower unit shaft system.

[0108] Then, based on the Mean-Shift density clustering method, the cluster center of the multidimensional fusion feature set of the health samples of each key measuring point of the hydropower unit shaft system is calculated, and the cluster center is used as the health benchmark feature vector of each key measuring point.

[0109] Specifically, the method for obtaining the health benchmark feature vector of each key measuring point is:

[0110] The Mean-Shift density clustering method is used to calculate the drift vector of the center point according to the change in data density in the region, and then the center is moved and the next iteration is started until it converges to the point with the maximum probability density. The calculation process is as follows:

[0111] Assume that there are n data points in d-dimensional Euclidean space constituting a data set {x i , i = 1, 2, ..., n}, then the kernel density estimation function at x is expressed by the kernel function K(x) and the bandwidth h:

[0112]

[0113] Express the kernel density estimation function in a radially symmetric form:

[0114] K(x)=c k,d k(||x|| 2 ) (8),

[0115] In the formula, c k,d is the normalization coefficient; k(x) is the profile function of K(x);

[0116] Substituting formula (8) into formula (7), we get:

[0117]

[0118] Define g(x) = -k'(x), then the kernel function corresponding to the profile function g(x) is G(x) = c g,d g(‖x‖ 2 ), c g,d is the normalization coefficient, is the probability density estimated by the kernel function G(x);

[0119] The derivative is The gradient of is:

[0120]

[0121] Assume y j (j=1,2,…) is the result of the jth iteration. In the process of mean shift, y j+1 With y j The relationship is:

[0122]

[0123] This cycle is repeated until the convergence condition is met, such as y j+1 -y j Below a certain threshold or reaching a maximum number of iterations to find the local maximum point of the probability density function, thereby identifying the cluster centers in the data.

[0124] Next, the vibration signals of each key measuring point of the shaft system during the operation of the hydropower unit are collected in real time, and the multi-dimensional fusion feature vectors of the vibration signals of each key measuring point of the shaft system collected in real time are extracted. The relative Euclidean distance between the multi-dimensional fusion feature vectors of the vibration signals of each key measuring point of the shaft system collected in real time and the corresponding healthy benchmark feature vectors is calculated as the performance degradation index of each key measuring point of the hydropower unit shaft system. The calculation process of the performance degradation index is as follows: Figure 3 shown.

[0125] Specifically, the calculation method of the performance degradation index of each key measurement point is:

[0126] Firstly, the vibration signals of each key measuring point of the shaft system are collected during the operation of the hydropower unit, and the multi-dimensional fusion feature vector R(t) of the vibration signals of each key measuring point of the shaft system collected in real time is extracted.

[0127] The method for extracting the multidimensional fusion feature vector R(t) is the same as the method for constructing the multidimensional fusion feature vector of the vibration signal of the hydropower unit, except that the root input data they are based on are different. In the method for extracting the multidimensional fusion feature vector R(t), the root input data are: the vibration signals of each key measuring point of the shaft system of the hydropower unit collected in real time; in the method for constructing the multidimensional fusion feature vector of the vibration signal of the hydropower unit, the root input data are: the vibration signals of each key measuring point of the shaft system of the hydropower unit during healthy operation.

[0128] In the extraction of the multi-dimensional fusion feature vector R(t), based on the real-time collected vibration signals of the key measuring points of the hydropower unit shaft system, the typical time domain features (T1', T2', ..., T5'), typical frequency domain features (F1', F2', ..., F5'), and multi-scale attention entropy features (M1', M2', ..., M n ') and unsupervised features (S1',S2',...,S m '), construct the multi-dimensional fusion feature vector R(t)=(T1',T2',...,T5',F1',F2',...,F5',M1',M2',...,M n ',S1',S2',...,S m ');

[0129] The relative Euclidean distance is used to measure the deviation between the multi-dimensional fusion feature vector R(t) of the vibration signal of each key measuring point of the shaft system collected in real time and the corresponding healthy benchmark feature vector C, which is used as the performance degradation index H(t) of each key measuring point of the shaft system. The calculation formula is as follows:

[0130]

[0131] Finally, the performance degradation index of each key measuring point of the hydropower unit shaft system is weighted and summed to calculate the comprehensive degradation index of the hydropower unit shaft system. The comprehensive degradation index of the hydropower unit shaft system is compared with the health limit of the comprehensive degradation assessment index to obtain the comprehensive degradation assessment result of the hydropower unit shaft system.

[0132] Specifically, the calculation method of the comprehensive degradation index of the shaft system of the hydropower unit is:

[0133] The performance degradation index of each key measuring point is weighted and summed to obtain the comprehensive degradation index HI of the shaft system of the hydropower unit. The formula is:

[0134] HI=ω1H1+ω2H2+...+ω z H z (13)

[0135] Where H1, H2, ..., Hz are the performance degradation indexes of the key measuring points of the 1st to Zth axis systems, ω1, ω2, ..., ω z It represents the weight of the performance degradation index of the corresponding measuring point, which is determined based on the power system's evaluation of the importance of each key measuring point of the hydropower unit shaft system.

[0136] It should be noted that the comprehensive degradation degree of the shaft system of the hydropower unit is judged based on the degree of change of the comprehensive degradation indicators of the shaft system, and the health limit values ​​of the comprehensive degradation assessment indicators will vary according to the specific actual conditions of the hydropower unit.

[0137] The present invention also proposes a comprehensive degradation assessment system for the shaft system of a hydropower unit, the structure diagram of which is shown in FIG. Figure 4 As shown, it consists of a healthy sample set acquisition module, a data preprocessing module, a healthy sample multi-dimensional fusion feature set establishment module, a healthy benchmark feature vector calculation module, a performance degradation index calculation module, and a comprehensive degradation evaluation module.

[0138] The healthy sample set acquisition module is used to obtain the healthy sample set of vibration signals of key measuring points of the shaft system of the hydropower unit under stable operating conditions and during the healthy operation of the hydropower unit.

[0139] The data preprocessing module uses the complementary set empirical mode decomposition method and the wavelet threshold denoising method to preprocess the healthy sample set of the vibration signal to obtain the preprocessed healthy sample set of the vibration signal.

[0140] The healthy sample multi-dimensional fusion feature set establishment module is used to analyze the pre-processed vibration signal healthy sample set, obtain typical time domain features, typical frequency domain features and multi-scale attention entropy features, and mine the unsupervised features in the pre-processed vibration signal healthy sample set through stacked denoising autoencoders. Combined with typical time domain features, typical frequency domain features, multi-scale attention entropy features and unsupervised features, a multi-dimensional fusion feature set of healthy samples for key measuring points of the hydropower unit shaft system is established.

[0141] The health benchmark feature vector calculation module calculates the cluster center of the multidimensional fusion feature set of health samples of each key measuring point of the hydropower unit shaft system based on the Mean-Shift density clustering method, and uses the cluster center as the health benchmark feature vector of each key measuring point.

[0142] The performance degradation index calculation module is used to collect the vibration signals of each key measuring point of the shaft system during the operation of the hydropower unit in real time, extract the multi-dimensional fusion feature vectors of the vibration signals of each key measuring point of the shaft system collected in real time, and calculate the relative Euclidean distance between the multi-dimensional fusion feature vectors of the vibration signals of each key measuring point of the shaft system collected in real time and the corresponding healthy benchmark feature vectors as the performance degradation index of each key measuring point of the hydropower unit shaft system.

[0143] The comprehensive degradation assessment module is used to perform weighted summation of the performance degradation index of each key measuring point of the hydropower unit shaft system, calculate the comprehensive degradation index of the hydropower unit shaft system, compare the comprehensive degradation index of the hydropower unit shaft system with the health limit of the comprehensive degradation assessment index, and obtain the comprehensive degradation assessment result of the hydropower unit shaft system.

[0144] The implementation method of each module in the system is consistent with the content described in the above method and will not be repeated here.

[0145] The present invention also proposes an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to implement the comprehensive degradation assessment method and comprehensive degradation assessment system of the shaft system of a hydropower unit as described above.

[0146] The present invention also proposes an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to implement the comprehensive degradation assessment method and comprehensive degradation assessment system of the shaft system of a hydropower unit as described above.

[0147] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the application can adopt the form of complete hardware embodiment, complete software embodiment, or the embodiment in combination with software and hardware. Moreover, the application can adopt the form of the computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiment of the present application can be implemented in various computer languages, for example, object-oriented programming language Java, C++, Python and literal scripting language JavaScript, etc.

[0148] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0149] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0150] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0151] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0152] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A comprehensive degradation assessment method for the shaft system of a hydropower unit, characterized in that: include: Obtain a healthy sample set of vibration signals of key measuring points of the shaft system of the hydropower unit under stable operating conditions and during the healthy operation of the hydropower unit; The complementary set empirical mode decomposition method and wavelet threshold denoising method are used to preprocess the healthy sample set of vibration signals to obtain the preprocessed healthy sample set of vibration signals. The preprocessed vibration signal healthy sample set is analyzed to obtain typical time domain features, typical frequency domain features and multi-scale attention entropy features. The unsupervised features in the preprocessed vibration signal healthy sample set are mined through stacked denoising autoencoders. The multi-dimensional fusion feature set of healthy samples of key measuring points of the hydropower unit shaft system is established by combining typical time domain features, typical frequency domain features, multi-scale attention entropy features and unsupervised features. Based on the Mean-Shift density clustering method, the cluster center of the multi-dimensional fusion feature set of the health samples of each key measuring point of the hydropower unit shaft system is calculated, and the cluster center is used as the health benchmark feature vector of each key measuring point. The vibration signals of each key measuring point of the shaft system during the operation of the hydropower unit are collected in real time, and the multi-dimensional fusion feature vectors of the vibration signals of each key measuring point of the shaft system collected in real time are extracted. The relative Euclidean distance between the multi-dimensional fusion feature vectors of the vibration signals of each key measuring point of the shaft system collected in real time and the corresponding healthy reference feature vectors is calculated as the performance degradation index of each key measuring point of the shaft system of the hydropower unit; The performance degradation index of each key measuring point of the hydropower unit shaft system is weighted and summed to calculate the comprehensive degradation index of the hydropower unit shaft system. The comprehensive degradation index of the hydropower unit shaft system is compared with the health limit of the comprehensive degradation assessment index to obtain the comprehensive degradation assessment result of the hydropower unit shaft system.

2. The comprehensive degradation assessment method for the shaft system of a hydropower unit according to claim 1 is characterized in that: The preprocessing method of the vibration signal healthy sample set includes: For the healthy sample set of vibration signals of key measuring points of the shaft system of the hydropower unit during healthy operation, the complementary set empirical mode decomposition method is used to decompose the vibration signal step by step into several relatively stable intrinsic mode functions IMF and a standard residual. The mutual correlation coefficient between each IMF component and the original vibration signal is calculated by formula (1) to determine the noise component and effective component in the intrinsic mode function IMF. Formula (1) is as follows: Where i represents the order number of the intrinsic mode function IMF, and IMF i represents the i-th order eigenmode function, and y represents the original vibration signal; The wavelet threshold denoising method is used to denoise the noise component, and the noise component after denoising is superimposed with the effective component and in the original vibration signal to obtain the preprocessed vibration signal healthy sample set.

3. The comprehensive degradation assessment method for the shaft system of a hydropower unit according to claim 2 is characterized in that: The typical time domain features and typical frequency domain features are obtained by extracting vibration signals based on the amplitude domain analysis method and Fourier transform theory; The typical time domain features include: standard deviation T1, peak-to-peak value T2, kurtosis T3, skewness T4 and root mean square value T5; The typical frequency domain features include: frequency amplitude mean F1, frequency centroid value F2, frequency standard deviation F3, frequency variation coefficient F4 and frequency slope F5; The method for obtaining the multi-scale attention entropy feature is: For the preprocessed vibration signal healthy sample set, given a time series signal X with a length of N = {x1, x2, ..., x N }, and then the coarse-grained time series is obtained by using formula (2), namely: Where τ is the scale factor, ranging from 1 to τ max , τ max is the maximum scale factor; Y i (τ) is a coarse-grained time series at scale τ; when τ = 1, the coarse-grained component is the original signal; when τ > 1, the original signal is divided into multiple coarse-grained sequences of length [N / τ] after coarse-grained processing; Compute the coarse-grained time series Y at scale τ i (τ) The attention entropy value is defined as MATE(X,τ) of the time series signal X at this scale: MATE(X,τ)=ATE(Y i (τ) )(3), In the formula, ATE(Y i (τ) ) represents the sequence Y i (τ) The attention entropy value of Repeat the calculation until τ = τ max , the multi-scale attention entropy of the signal is the set of MATE(X,τ), that is: MATE={MATE(X,τ)},τ=1,2,...,τ max (4); The unsupervised feature mining method includes: Multiple denoising autoencoders are stacked in sequence to form a stacked denoising autoencoder. Each denoising autoencoder is trained independently in a bottom-up order. The lower denoising autoencoder uses the implicit representation of the denoising autoencoder trained in the upper layer as a pure input, and is trained layer by layer until the highest hidden layer outputs the deepest high-order features of the original input signal, that is, unsupervised features. Given an input vector X, the denoising autoencoder transforms by random mapping Add random noise to the input data to get "contaminated" data q D is represented as a binomial random hidden noise; therefore, the encoded output H of the denoising autoencoder and the reconstructed decoded output Z are: Z = g(H) = s(W'H + b2) (6), Where W∈R m×n , W'∈R n×m are the connection weights of the encoding network and the decoding network respectively; b1∈R m×1 , b2∈R n×1 are the biases of the encoding network and the decoding network respectively; s(x) is the activation function, which uses the Sigmoid function, that is, s(x) = 1 / (1+e -x ); During the denoising autoencoder training process, the model weights and biases are iteratively updated through the back-propagation algorithm to minimize the reconstruction error between the reconstructed output Y and the original noise-free input X; Normalize various feature indicators, combine typical time domain features (T1, T2, ..., T5), typical frequency domain features (F1, F2, ..., F5), multi-scale attention entropy features (M1, M2, ..., M n ) and unsupervised features (S1,S2,...,S m ), construct the multi-dimensional fusion feature vector of the vibration signal of the hydropower unit (T1,T2,...,T5,F1,F2,...,F5,M1,M2,...,M n ,S1,S2,...,S m ), and then establish a multi-dimensional fusion feature set of healthy samples of key measuring points of the hydropower unit shaft system.

4. The comprehensive degradation assessment method for the shaft system of a hydropower unit according to claim 3 is characterized in that: The method for obtaining the health benchmark feature vector of each key measuring point is: The Mean-Shift density clustering method is used to calculate the drift vector of the center point according to the change in data density in the region, and then the center is moved and the next iteration is started until it converges to the point with the maximum probability density. The calculation process is as follows: Assume that there are n data points in d-dimensional Euclidean space constituting a data set {x i , i = 1, 2, ..., n}, then the kernel density estimation function at x is expressed by the kernel function K(x) and the bandwidth h: Express the kernel density estimation function in a radially symmetric form: K(x)=c k,d k(||x|| 2 ) (8), In the formula, c k,d is the normalization coefficient; k(x) is the profile function of K(x); Substituting formula (8) into formula (7), we get: Define g(x) = -k'(x), then the kernel function corresponding to the profile function g(x) is G(x) = c g,d g(‖x‖ 2 ), c g,d is the normalization coefficient, is the probability density estimated by the kernel function G(x); The derivative is The gradient of is: Assume y j (j=1,2,…) is the result of the jth iteration. In the process of mean shift, y j+1 With y j The relationship is: This cycle is repeated until the convergence condition is met, such as y j+1 -y j Below a certain threshold or reaching a maximum number of iterations to find the local maximum point of the probability density function, thereby identifying the cluster centers in the data.

5. The comprehensive degradation assessment method for the shaft system of a hydropower unit according to claim 4 is characterized in that: The calculation method of the performance degradation index of each key measurement point is: Firstly, the vibration signals of each key measuring point of the shaft system are collected during the operation of the hydropower unit, and the multi-dimensional fusion feature vector R(t) of the vibration signals of each key measuring point of the shaft system collected in real time is extracted; The extraction method of the multi-dimensional fusion feature vector R(t) is as follows: based on the real-time collected vibration signals of each key measuring point of the shaft system, the typical time domain features (T1', T2', ..., T5'), the typical frequency domain features (F1', F2', ..., F5'), and the multi-scale attention entropy features (M1', M2', ..., M n ') and unsupervised features (S1',S2',...,S m '), construct the multi-dimensional fusion feature vector R(t)=(T1',T2',...,T5',F1',F2',...,F5',M1',M2',...,M n ',S1',S2',...,S m '); The relative Euclidean distance is used to measure the deviation between the multi-dimensional fusion feature vector R(t) of the vibration signal of each key measuring point of the shaft system collected in real time and the corresponding healthy benchmark feature vector C, which is used as the performance degradation index H(t) of each key measuring point of the shaft system. The calculation formula is as follows:

6. The comprehensive degradation assessment method for the shaft system of a hydropower unit according to claim 5 is characterized in that: The calculation method of the comprehensive degradation index of the shaft system of the hydropower unit is: The performance degradation index of each key measuring point is weighted and summed to obtain the comprehensive degradation index HI of the shaft system of the hydropower unit. The formula is: HI=ω1H1+ω2H2+...+ω z H z (13), Where H1, H2, ..., H z are the performance degradation indexes of the key measuring points of the 1st to Zth axis systems, ω1, ω2, ..., ω z It represents the weight of the performance degradation index of the corresponding measuring point, which is determined based on the power system's evaluation of the importance of each key measuring point of the hydropower unit shaft system.

7. A comprehensive degradation assessment system for a shaft system of a hydropower unit implementing the method described in any one of claims 1 to 6, characterized in that: It includes a healthy sample set acquisition module, a data preprocessing module, a healthy sample multi-dimensional fusion feature set establishment module, a healthy baseline feature vector calculation module, a performance degradation index calculation module, and a comprehensive degradation evaluation module; The healthy sample set acquisition module is used to acquire a healthy sample set of vibration signals of key measuring points of the shaft system of the hydropower unit under stable operating conditions and during the healthy operation of the hydropower unit; The data preprocessing module uses a complementary set empirical mode decomposition method and a wavelet threshold noise reduction method to preprocess the vibration signal healthy sample set to obtain a preprocessed vibration signal healthy sample set; The healthy sample multi-dimensional fusion feature set establishment module is used to analyze the pre-processed vibration signal healthy sample set, obtain typical time domain features, typical frequency domain features and multi-scale attention entropy features, and mine the unsupervised features in the pre-processed vibration signal healthy sample set through stacked denoising autoencoders. Combined with the typical time domain features, typical frequency domain features, multi-scale attention entropy features and unsupervised features, a healthy sample multi-dimensional fusion feature set of each key measuring point of the hydropower unit shaft system is established; The health benchmark feature vector calculation module calculates the cluster center of the health sample multidimensional fusion feature set of each key measuring point of the hydropower unit shaft system based on the Mean-Shift density clustering method, and uses the cluster center as the health benchmark feature vector of each key measuring point; The performance degradation index calculation module is used to collect the vibration signals of each key measuring point of the shaft system in real time during the operation of the hydropower unit, extract the multi-dimensional fusion feature vector of the vibration signals of each key measuring point of the shaft system collected in real time, and calculate the relative Euclidean distance between the multi-dimensional fusion feature vector of the vibration signals of each key measuring point of the shaft system collected in real time and the corresponding healthy reference feature vector as the performance degradation index of each key measuring point of the shaft system of the hydropower unit; The comprehensive degradation assessment module is used to perform weighted summation on the performance degradation index of each key measuring point of the hydropower unit shaft system, calculate the comprehensive degradation index of the hydropower unit shaft system, compare the comprehensive degradation index of the hydropower unit shaft system with the health limit of the comprehensive degradation assessment index, and obtain the comprehensive degradation assessment result of the hydropower unit shaft system.

8. An electronic device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor implements the comprehensive degradation assessment method for the shaft system of a hydropower unit as described in any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the comprehensive degradation assessment method for the shaft system of a hydropower unit as described in any one of claims 1 to 6 is implemented.

Images