Hydropower station water guide mechanism operation fault identification method based on deep learning

Abstract

The invention discloses a hydropower station water guide mechanism operation fault identification method based on deep learning, and the method comprises the following steps: processing vibration signal data of a water guide mechanism, and constructing a data set; dividing the data set into a training set and a test set in proportion; constructing a hydropower station water guide mechanism operation fault identification deep learning model; and taking the training set as input, training the model so as to obtain a water turbine water guide mechanism fault identification model, and verifying the effect of the model through the test set. According to the hydropower station water guide mechanism operation fault identification method based on deep learning, the time domain and frequency domain characteristics of the time history signal are fused, the time influence in the time history signal is considered at the same time, single-angle analysis is avoided, and the precision and robustness of a fault diagnosis model are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric power engineering, and in particular relates to a method for identifying operating faults of a water guiding mechanism of a hydropower station based on deep learning. Background Art

[0002] The water-guiding mechanism of a hydropower station is the core supporting component of the turbine and a part of the flow channel. It can control the flow rate. Once the impeller fails, the water flow speed and direction may change, thus affecting energy conversion. In severe cases, the water-guiding mechanism may collapse, causing damage to the entire plant and even endangering the lives of workers. Currently, there is a lack of effective diagnosis of the operating signals of the water-guiding structure. There are mainly the following problems: (1) The current diagnostic method basically converts the time-series signal into a characteristic value from the perspective of time domain, frequency domain, etc., analyzes the fault from a macro perspective, and ignores the influence of time on the signal at the micro level; (2) The influence of time on fault diagnosis is only considered from the perspective of time history, while the influence of the time domain and frequency domain characteristics of the signal on fault diagnosis is ignored. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a method for identifying operating faults of a water-conducting mechanism of a hydropower station based on deep learning, which can improve the accuracy and robustness of a fault diagnosis model.

[0004] The technical solution of the present invention is: a method for identifying operating faults of a water diversion mechanism of a hydropower station based on deep learning, comprising the following steps:

[0005] (1) Process the vibration signal data of the water diversion mechanism and construct a data set;

[0006] (2) Divide the data set into a training set and a test set in a ratio of 8:2;

[0007] (3) Constructing a deep learning model for identifying operational faults of water diversion mechanisms in hydropower stations, including the following steps:

[0008] S1. Starting from the time-course signal analysis, a one-dimensional CNN network structure is constructed, wherein the one-dimensional CNN network structure includes: a 1×3 convolution layer, an attention mechanism, a pooling layer, a normalization layer, an LSTM layer, a fully connected layer, and a DROPOUT regularization. Two layers of the 1×3 convolution layers extract the features of the time-course signal, and the other 1×3 convolution layer introduces an attention mechanism.

[0009] S2. Obtaining the frequency domain feature of the time-course signal by Fourier transform, wherein the frequency domain feature is the maximum frequency value;

[0010] S3, extracting a time domain feature from the time history signal, wherein the time domain feature is a root mean square value;

[0011] S4, using SVM classifier to classify the features of three dimensions;

[0012] (4) The training set is used as input to train the model to obtain a fault identification model for the water turbine guide mechanism, and the effect of the model is verified through the test set.

[0013] Furthermore, the processing of the vibration signal data of the water guiding mechanism is specifically to extract features through three different paths, and the three different paths include: path 1: extracting features of the time history signal through a 1×3 convolution layer, path 2: extracting frequency domain features through Fourier transform, and path 3: extracting time domain features from the time history signal.

[0014] Furthermore, the convolution operation of the 1×3 convolutional layer is performed on the time series signal, and the expression of the convolution operation is:

[0015] Y=σ(W*X+b),

[0016] Among them, * represents the convolution operation, W represents the convolution kernel, σ represents the activation function, and b represents the bias term;

[0017] The expression of the attention mechanism is:

[0018]

[0019] Among them, Q represents the query matrix, K represents the key matrix, V represents the value matrix, and d k represents the scaling factor, T represents the transpose of the matrix;

[0020] The expression of the pooling layer is:

[0021] Y pool =max(X i：i+k ),

[0022] Among them, Y pool represents the output of the pooling layer, X i：i+k Represents a subsequence or subregion of the i-th element to the i+k-th element of the input data X;

[0023] The expression of the normalization layer is:

[0024]

[0025] Among them, X norm represents the normalized input data, X represents the input data, μ represents the mean of the input data, and σ represents the standard deviation of the input data;

[0026] The expression of the LSTM layer is:

[0027] f t =σ(W f ·[h t-1 , xt ]+b f )

[0028] i t =σ(W i ·[h t-1 , x t ]+b i )

[0029]

[0030] o t =σ(W o ·[h t-1 , x t ]+b o )

[0031] h t =o t tanh(C t ),

[0032] Among them, f t represents the output of the forget gate, σ represents the Sigmoid activation function, W f represents the weight matrix of the forget gate, x t represents the input of the current time step, h t-1 represents the hidden state of the previous time step, b f represents the bias term of the forget gate, i t represents the output of the input gate, W i represents the weight matrix of the input gate, b i represents the bias term of the input gate, represents the candidate memory unit, W C represents the weight matrix of the candidate memory unit, b C represents the bias term of the candidate memory unit, C t Represents the memory unit of the current time step, C t-1 represents the memory unit of the previous time step, o t represents the output of the output gate, W o represents the weight matrix of the output gate, b o represents the bias term of the output gate, h t represents the hidden state of the current time step;

[0033] The expression of the fully connected layer is:

[0034] Y fc =W fc ·X+b fc ,

[0035] Among them, Y fc represents the output of the fully connected layer, Wfc represents the weight matrix of the fully connected layer, X represents the input vector, and b fc Represents the bias vector of the fully connected layer;

[0036] The expression of DROPOUT regularization is:

[0037] h = Dropout(h, p),

[0038] Among them, h represents the input feature vector and p represents the discard rate.

[0039] Furthermore, the expression for obtaining the frequency domain characteristics of the time-course signal through Fourier transform is:

[0040]

[0041] Among them, x(t) represents the input time-course signal, t represents the time variable, f represents the frequency variable, and j represents the imaginary unit;

[0042] The expression for extracting the time domain feature from the time history signal is:

[0043]

[0044] Where RMS represents the root mean square value, N represents the total number of data points in the time history signal, and x i Represents the i-th data point of the time history signal.

[0045] Furthermore, the expression of the SVM classifier is:

[0046]

[0047] subject to you i (w·x i +b)≥1-ξ i ,ξ i ≥0,

[0048] Among them, w represents the weight vector, b represents the bias term of the decision function, ξ i represents the slack variable, i.e., the distance that the i-th data point is not on the correct side and is outside the boundary, y i represents the category label of the i-th data point, x i represents the feature vector of the i-th data point, C represents the regularization parameter, and N represents the total number of data points.

[0049] Beneficial effects of the present invention: A method for identifying operating faults of a water diversion mechanism of a hydropower station based on deep learning of the present invention constructs a one-dimensional CNN network structure by starting from the analysis of time-course signals, which consists of a 1×3 convolution layer, an attention mechanism, a pooling layer, a normalization layer, an LSTM layer, a fully connected layer, and a DROPOUT regularization. Two 1x3 convolution layers extract characteristic information of time-course signals, and at the same time, an attention mechanism is introduced to enable another convolution layer to consider local information and global information. The pooling layer and the normalization layer improve the calculation speed and reduce redundant information. The LSTM structure avoids loss in the time dimension. Information, a fully connected layer and DropOut regularization are used to reduce the influence of feature quantities on the results, while avoiding data overfitting, and finally outputting a feature value that reflects the fault; by using Fourier transform to obtain frequency domain features and directly extracting the root mean square value from the time domain, the key information in the vibration signal of the water guiding mechanism can be fully captured; in short, a method for identifying operating faults of water guiding mechanisms of hydropower stations based on deep learning of the present invention avoids single-angle analysis by integrating the time domain and frequency domain features of time history signals and considering the influence of moments in time history signals, thereby improving the accuracy and robustness of the fault diagnosis model. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention.

[0051] Figure 1 It is a flow chart of a method for identifying operating faults of a water diversion mechanism of a hydropower station based on deep learning according to the present invention. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0053] like Figure 1 As shown, a method for identifying operating faults of a water diversion mechanism of a hydropower station based on deep learning comprises the following steps:

[0054] (1) Process the vibration signal data of the water diversion mechanism and construct a data set;

[0055] (2) Divide the data set into a training set and a test set in a ratio of 8:2;

[0056] (3) Constructing a deep learning model for identifying operational faults of water diversion mechanisms in hydropower stations, including the following steps:

[0057] S1. Starting from the time-course signal analysis, a one-dimensional CNN (convolutional neural network) network structure is constructed, wherein the one-dimensional CNN network structure includes: a 1×3 convolution layer, an attention mechanism, a pooling layer, a normalization layer, an LSTM (long short-term memory network) layer, a fully connected layer, and a DROPOUT (discarding some neurons in the neural network) regularization. Two layers of the 1×3 convolution layers extract the features of the time-course signal, and another 1×3 convolution layer introduces an attention mechanism.

[0058] S2. Obtaining the frequency domain feature of the time-course signal by Fourier transform, wherein the frequency domain feature is the maximum frequency value;

[0059] S3, extracting a time domain feature from the time history signal, wherein the time domain feature is a root mean square value;

[0060] S4, using SVM classifier to classify the features of three dimensions;

[0061] (4) The training set is used as input to train the model to obtain a fault identification model for the water turbine guide mechanism, and the effect of the model is verified through the test set.

[0062] Preferably, the processing of the vibration signal data of the water guiding mechanism is specifically to extract features through three different paths, and the three different paths include: path 1: extracting features of the time history signal through a 1×3 convolution layer, path 2: extracting frequency domain features through Fourier transform, and path 3: extracting time domain features from the time history signal.

[0063] Preferably, the convolution operation of the 1×3 convolutional layer is performed on the time series signal, and the expression of the convolution operation is:

[0064] Y=σ(W*X+b),

[0065] Among them, * represents the convolution operation, W represents the convolution kernel, σ represents the activation function, and b represents the bias term;

[0066] The expression of the attention mechanism is:

[0067]

[0068] Among them, Q represents the query matrix, K represents the key matrix, V represents the value matrix, and d k represents the scaling factor, T represents the transpose of the matrix;

[0069] The expression of the pooling layer is:

[0070] Y pool =max(X i：i+k ),

[0071] Among them, Y pool represents the output of the pooling layer, Xi：i+k Represents a subsequence or subregion of the i-th element to the i+k-th element of the input data X;

[0072] The expression of the normalization layer is:

[0073]

[0074] Among them, X norm represents the normalized input data, X represents the input data, μ represents the mean of the input data, and σ represents the standard deviation of the input data;

[0075] The expression of the LSTM layer is:

[0076] f t =σ(W f ·[h t-1 , x t ]+b f )

[0077] i t =σ(W i ·[h t-1 , x t ]+b i )

[0078]

[0079] o t =σ(W o ·[h t-1 , x t ]+b o )

[0080] h t =o t tanh(C t ),

[0081] Among them, f t represents the output of the forget gate, σ represents the Sigmoid activation function, W f represents the weight matrix of the forget gate, x t represents the input of the current time step, h t-1 represents the hidden state of the previous time step, b f represents the bias term of the forget gate, i t represents the output of the input gate, W i represents the weight matrix of the input gate, b i represents the bias term of the input gate, represents the candidate memory unit, W C represents the weight matrix of the candidate memory unit, b C represents the bias term of the candidate memory unit, Ct Represents the memory unit of the current time step, C t-1 represents the memory unit of the previous time step, o t represents the output of the output gate, W o represents the weight matrix of the output gate, b o represents the bias term of the output gate, h t represents the hidden state of the current time step;

[0082] The expression of the fully connected layer is:

[0083] Y fc =W fc ·X+b fc ,

[0084] Among them, Y fc represents the output of the fully connected layer, W fc represents the weight matrix of the fully connected layer, X represents the input vector, and b fc Represents the bias vector of the fully connected layer;

[0085] The expression of DROPOUT regularization is:

[0086] h = Dropout(h, p),

[0087] Among them, h represents the input feature vector and P represents the discard rate.

[0088] Preferably, the expression for obtaining the frequency domain characteristics of the time-course signal by Fourier transform is:

[0089]

[0090] Among them, x(t) represents the input time-course signal, t represents the time variable, f represents the frequency variable, and j represents the imaginary unit;

[0091] The expression for extracting the time domain feature from the time history signal is:

[0092]

[0093] Where RMS represents the root mean square value, N represents the total number of data points in the time history signal, and x i Represents the i-th data point of the time history signal.

[0094] Preferably, the expression of the SVM classifier is:

[0095]

[0096] subject to you i (w·x i +b)≥1-ξi ,ξ i ≥0,

[0097] Among them, w represents the weight vector, b represents the bias term of the decision function, and ξ i represents the slack variable, i.e., the distance that the i-th data point is not on the correct side and is outside the boundary, y i represents the category label of the i-th data point, x i represents the feature vector of the i-th data point, C represents the regularization parameter, and N represents the total number of data points.

[0098] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying operating faults of water diversion mechanisms in a hydropower station based on deep learning, characterized in that: The following steps are involved: (1) Process the vibration signal data of the water diversion mechanism and construct a data set; (2) Divide the data set into a training set and a test set in a ratio of 8:2; (3) Constructing a deep learning model for identifying operational faults of water diversion mechanisms in hydropower stations, including the following steps: S1. Starting from the time-course signal analysis, a one-dimensional CNN network structure is constructed, wherein the one-dimensional CNN network structure includes: a 1×3 convolution layer, an attention mechanism, a pooling layer, a normalization layer, an LSTM layer, a fully connected layer, and a DROPOUT regularization. Two layers of the 1×3 convolution layers extract the features of the time-course signal, and the other 1×3 convolution layer introduces the attention mechanism. S2. Obtaining the frequency domain feature of the time-course signal by Fourier transform, wherein the frequency domain feature is the maximum frequency value; S3, extracting a time domain feature from the time history signal, wherein the time domain feature is a root mean square value; S4, using SVM classifier to classify the features of three dimensions; (4) The training set is used as input to train the model to obtain a fault identification model for the water turbine guide mechanism, and the effect of the model is verified through the test set.

2. According to the method for identifying operating faults of water diversion mechanisms of a hydropower station based on deep learning in claim 1, it is characterized in that: The processing of the vibration signal data of the water guiding mechanism is specifically to extract features through three different paths, and the three different paths include: path 1: extracting features of the time history signal through a 1×3 convolution layer, path 2: extracting frequency domain features through Fourier transform, and path 3: extracting time domain features from the time history signal.

3. The method for identifying operating faults of water diversion mechanisms of a hydropower station based on deep learning according to claim 1 is characterized in that: The convolution operation of the 1×3 convolutional layer is performed on the time series signal, and the expression of the convolution operation is: Y=σ(W*X+b), Among them, * represents the convolution operation, W represents the convolution kernel, σ represents the activation function, and b represents the bias term; The expression of the attention mechanism is: Among them, Q represents the query matrix, K represents the key matrix, V represents the value matrix, and d k represents the scaling factor, and T represents the transpose of the matrix; The expression of the pooling layer is: AND pllo =max(X i:i+k ), Among them, Y pool represents the output of the pooling layer, X i:i+k Represents a subsequence or subregion of the i-th element to the i+k-th element of the input data X; The expression of the normalization layer is: Among them, X norm represents the normalized input data, X represents the input data, μ represents the mean of the input data, and σ represents the standard deviation of the input data; The expression of the LSTM layer is: f t =σ(W f ·[h t-1 ,x t ]+b f ) =σ(W i ·[h t-1 ,x t ]+b i ) the t =σ(W o ·[h t-1 ,x t ]+b o ) h t =o t ·tanh(C t ), Among them, f t represents the output of the forget gate, σ represents the Sigmoid activation function, W f represents the weight matrix of the forget gate, x t represents the input of the current time step, h t-1 represents the hidden state of the previous time step, b f represents the bias term of the forget gate, i t represents the output of the input gate, W i represents the weight matrix of the input gate, b i represents the bias term of the input gate, represents the candidate memory unit, W C represents the weight matrix of the candidate memory unit, b C represents the bias term of the candidate memory unit, C t Represents the memory unit of the current time step, C t-1 represents the memory unit of the previous time step, o t represents the output of the output gate, W o represents the weight matrix of the output gate, b o represents the bias term of the output gate, h t represents the hidden state of the current time step; The expression of the fully connected layer is: Y fc =W fc ·X+b fc , Among them, Y fc represents the output of the fully connected layer, W fc represents the weight matrix of the fully connected layer, X represents the input vector, and b fc Represents the bias vector of the fully connected layer; The expression of DROPOUT regularization is: h = Dropout(h, p), Among them, h represents the input feature vector and p represents the discard rate.

4. The method for identifying operating faults of water diversion mechanisms of a hydropower station based on deep learning according to claim 1 is characterized in that: The expression for obtaining the frequency domain characteristics of the time-course signal by Fourier transform is: Among them, x(t) represents the input time-course signal, t represents the time variable, f represents the frequency variable, and j represents the imaginary unit; The expression for extracting the time domain feature from the time history signal is: Where RMS represents the root mean square value, N represents the total number of data points in the time history signal, and x i Represents the i-th data point of the time history signal.

5. The method for identifying operating faults of water diversion mechanisms of a hydropower station based on deep learning according to claim 1 is characterized in that: The expression of the SVM classifier is: subject to y i (w·x i +b)≥1-ξ i ,ξ i ≥0, Among them, w represents the weight vector, b represents the bias term of the decision function, ξ i represents the slack variable, i.e., the distance that the i-th data point is not on the correct side and is outside the boundary, y i represents the category label of the i-th data point, x i represents the feature vector of the i-th data point, C represents the regularization parameter, and N represents the total number of data points.

Images