A method, system, equipment, and medium for fault identification in hydropower stations based on wavelet decomposition.
By processing mechanical vibration signals from hydropower stations using wavelet decomposition and convolutional neural networks, a fault identification model is constructed. This solves the problems of existing technologies being unable to predict hydropower station faults in advance and having low identification accuracy, thus achieving early fault identification and high-precision fault detection.
Patent Information
- Application Number
- CN202310594907.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2022-10-18
- Filing Date
- 2023-05-24
- Publication Date
- 2025-12-02
- Estimated Expiration
- 2043-05-24
AI Technical Summary
In existing technologies, it is impossible to predict hydropower station faults in advance when deploying multiple sensors, and the neural network is affected by interference signals when processing sensor signals, resulting in low recognition accuracy.
Wavelet decomposition was used to collect mechanical vibration signals from hydropower stations. After filtering and wavelet transform, training sample pairs were constructed and a fault identification model was trained. Convolutional neural networks and attention modules were used to enhance feature representation, optimize the loss function and weight update, and achieve early identification of faults in hydropower stations.
It improves the accuracy and sensitivity of fault identification in hydropower stations, enabling fault prediction before sensor signals exceed thresholds, reducing the impact of interference signals, and enhancing identification accuracy.
Smart Images
Figure CN116842417B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of hydropower station fault monitoring technology, specifically to a hydropower station fault identification method, system, equipment, and medium based on wavelet decomposition. Background Technology
[0002] In the daily operation of hydropower station units, a large number of fault signals usually occur. There are two main methods to identify faults in hydropower stations. One method is to deploy multiple sensors to collect sensing signals from various parts of the hydropower station unit and directly determine whether the hydropower station is faulty by checking whether the sensing signals exceed a threshold. The other method is to use neural networks to process the sensing signals to identify whether the hydropower station is faulty.
[0003] Existing methods that use multiple sensors to determine hydropower station faults when the sensor signals exceed a threshold cannot predict faults in advance; faults can only be determined when the sensor signals exceed the threshold.
[0004] Existing methods for identifying whether a hydropower station is malfunctioning by processing sensor signals through neural networks suffer from low accuracy because they cannot eliminate interference signals from the sensor signals. Summary of the Invention
[0005] To address the aforementioned shortcomings in existing technologies, this invention provides a method, system, equipment, and medium for fault identification in hydropower stations based on wavelet decomposition, which solves the following technical problems:
[0006] 1. Existing methods that determine hydropower station faults by deploying multiple sensors and detecting when the sensor signals exceed a threshold have the problem that they cannot predict hydropower station faults in advance and can only be determined when the sensor signals exceed the threshold.
[0007] 2. Existing methods for identifying whether a hydropower station is faulty by processing sensor signals through neural networks have the problem that interference signals cannot be eliminated from the sensor signals, resulting in low identification accuracy.
[0008] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:
[0009] Firstly, this application provides a method for fault identification in hydropower stations based on wavelet decomposition, comprising the following steps:
[0010] S1. Collect mechanical vibration signals during the operation of the hydropower station;
[0011] S2. Filter the mechanical vibration signal to obtain filtered mechanical vibration data;
[0012] S3. Perform wavelet transform on the filtered mechanical vibration data to obtain amplitude data;
[0013] S4. Match the amplitude data with the corresponding faults to construct training sample pairs;
[0014] S5. Use training samples to train the hydropower station fault identification model;
[0015] S6. The trained hydropower station fault identification model is used to process the amplitude data of the real-time mechanical vibration signal of the hydropower station to obtain the fault information.
[0016] Furthermore, step S2 includes the following sub-steps:
[0017] S21. Construct a cache window and populate it with initial values;
[0018] S22. Obtain the mechanical vibration value based on the mechanical vibration signal;
[0019] S23. Input the mechanical vibration value into the current buffer window, replacing the earliest mechanical vibration value entered in the current buffer window;
[0020] S24. Obtain the filtered value based on the median in the current cache window;
[0021] S25. Repeatedly process steps S23 to S24 according to the chronological order of acquisition of the mechanical vibration signal to obtain multiple filtered values, which constitute filtered mechanical vibration data.
[0022] Furthermore, the formula for calculating the filtered value in step S24 is as follows:
[0023]
[0024] Among them, y i The filter value corresponding to the buffer window when the i-th input mechanical vibration value is given; x0 is the median of the current buffer window; a is the filter weight; x1 to x i x represents the i mechanical vibration values in the buffer window, arranged in chronological order; 0-1 x represents the mechanical vibration value on one side of the median in the current cache window. 0+1 The mechanical vibration value on the other side of the median in the current cache window; y i-1 The filter value corresponding to the buffer window when the (i-1)th input mechanical vibration value is given.
[0025] The beneficial effects of the above-mentioned further scheme are as follows: In the buffer window, each time a new mechanical vibration value is input, the buffer window will exclude the oldest mechanical vibration value to ensure that the data in the buffer window is the most recently collected data. The median in the buffer window is extracted as the basis for the filtering value. If the new mechanical vibration value is greater than the median, the value is increased based on the median. If the new mechanical vibration value is less than the median, the value is decreased based on the median. This makes the change of the filtering value follow the change of the new mechanical vibration value, thus increasing the sensitivity.
[0026] Furthermore, the hydropower station fault identification model in step S5 includes: a first convolutional layer, a second convolutional layer, a third convolutional layer, a first additive layer A1, a max pooling layer, an average pooling layer, a first attention module, a second attention module, a concat layer, and a fully connected layer;
[0027] The first, second, and third convolutional layers are used to input amplitude data for each frequency. The first input of the first additive layer A1 is connected to the output of the first convolutional layer, its second input is connected to the output of the second convolutional layer, its third input is connected to the output of the third convolutional layer, and its output is connected to the inputs of the max pooling layer and the average pooling layer, respectively. The output of the max pooling layer is connected to the input of the first attention module. The output of the average pooling layer is connected to the input of the second attention module. The first input of the concat layer is connected to the output of the first attention module, its second input is connected to the output of the second attention module, and its output is connected to the input of the fully connected layer. The output of the fully connected layer serves as the output of the hydropower station fault identification model.
[0028] The beneficial effects of the above further scheme are as follows: The first, second, and third convolutional layers are used to input amplitude data for each frequency. The first, second, and third convolutional layers can input amplitude data for three frequencies, with each convolutional layer corresponding to the amplitude data for one frequency. After extracting feature data through the first, second, and third convolutional layers, the feature data for each frequency are added together through the first addition layer A1. Max pooling is used for pooling to retain significant features, and average pooling is used for pooling to retain overall features. The first and second attention modules are then used to enhance the feature representation capability of the model. The concat layer integrates the outputs of the first and second attention modules, and classification is performed through a fully connected layer.
[0029] Furthermore, the first attention module includes: a fourth convolutional layer, a fifth convolutional layer, a first softmax layer, a second addition layer A2, and a first multiplication layer M1;
[0030] The input of the fourth convolutional layer is connected to the input of the fifth convolutional layer and the first input of the second addition layer A2, and its output is connected to the input of the first softmax layer; the first input of the first multiplication layer M1 is connected to the output of the first softmax layer, its second input is connected to the output of the fifth convolutional layer, and its output is connected to the second input of the second addition layer A2; the output of the second addition layer A2 serves as the output of the first attention module.
[0031] Furthermore, the second attention module includes: a sixth convolutional layer, a seventh convolutional layer, a second softmax layer, a third addition layer A3, and a second multiplication layer M2;
[0032] The input of the sixth convolutional layer is connected to the input of the seventh convolutional layer and the first input of the third addition layer A3, and its output is connected to the input of the second softmax layer; the first input of the second multiplication layer M2 is connected to the output of the second softmax layer, its second input is connected to the output of the seventh convolutional layer, and its output is connected to the second input of the third addition layer A3; the output of the third addition layer A3 serves as the output of the second attention module.
[0033] Furthermore, the loss function for training the hydropower station fault identification model in step S5 is:
[0034]
[0035] Where L is the loss function, y j Y is the output of the hydropower station fault identification model during the j-th training iteration. j Let b be the label corresponding to the j-th training iteration, b be the weight coefficient, J be the total number of training iterations, and T be the training factor, where T is a positive integer. <J, This involves summing the data from the current J-th training iteration and the JT-th training iteration.
[0036] The beneficial effect of the above further scheme is that, through weighting coefficients, the output y is respectively... j With label Y j The distance, and the total label Y in the most recent times. j With the total output y j The differences are weighted so that the loss function simultaneously considers the current training iterations when outputting y. j With label Y j The distance and the total difference in the most recent T training sessions are used to optimize the parameters of the hydropower station fault identification model.
[0037] Furthermore, the weight update formula for training the hydropower station fault identification model in step S5 is as follows:
[0038]
[0039] Among them, w * Here, w represents the updated weights, and y represents the weights before the update. j Y is the output of the hydropower station fault identification model during the j-th training iteration. j Let L be the label corresponding to the j-th training iteration, L be the loss function, th be the ratio threshold, arctan() be the trigonometric function, and e be the natural logarithm. is the partial derivative symbol, and |·| is the absolute value.
[0040] The beneficial effects of the above further scheme are: in Y j -y j When the difference is large, through Accelerate the degree of weight change in Y j -y j When the difference is small, by Reduce the degree of weight reduction.
[0041] Secondly, this application provides a hydropower station fault identification system based on wavelet decomposition, characterized in that it includes:
[0042] The acquisition module is used to collect mechanical vibration signals during the operation of the hydropower station.
[0043] The filtering module is used to filter mechanical vibration signals to obtain filtered mechanical vibration data;
[0044] The data conversion module is used to perform wavelet transform on the filtered mechanical vibration data to obtain amplitude data;
[0045] The matching module is used to match amplitude data with corresponding faults to build training sample pairs;
[0046] The training module is used to train the hydropower station fault identification model using training samples;
[0047] The data processing module is used to process the amplitude data of real-time mechanical vibration signals of hydropower stations using the trained hydropower station fault identification model to obtain fault information.
[0048] Furthermore, the hydropower station fault identification model includes: a first convolutional layer, a second convolutional layer, a third convolutional layer, a first additive layer A1, a max pooling layer, an average pooling layer, a first attention module, a second attention module, a concat layer, and a fully connected layer;
[0049] The first, second, and third convolutional layers are used to input amplitude data for each frequency. The first input of the first additive layer A1 is connected to the output of the first convolutional layer, its second input is connected to the output of the second convolutional layer, its third input is connected to the output of the third convolutional layer, and its output is connected to the inputs of the max pooling layer and the average pooling layer, respectively. The output of the max pooling layer is connected to the input of the first attention module. The output of the average pooling layer is connected to the input of the second attention module. The first input of the concat layer is connected to the output of the first attention module, its second input is connected to the output of the second attention module, and its output is connected to the input of the fully connected layer. The output of the fully connected layer serves as the output of the hydropower station fault identification model.
[0050] The beneficial effects of the above further scheme are as follows: The first, second, and third convolutional layers are used to input amplitude data for each frequency. The first, second, and third convolutional layers can input amplitude data for three frequencies, with each convolutional layer corresponding to the amplitude data for one frequency. After extracting feature data through the first, second, and third convolutional layers, the feature data for each frequency are added together through the first addition layer A1. Max pooling is used for pooling to retain significant features, and average pooling is used for pooling to retain overall features. The first and second attention modules are then used to enhance the feature representation capability of the model. The concat layer integrates the outputs of the first and second attention modules, and classification is performed through a fully connected layer.
[0051] Furthermore, the first attention module includes: a fourth convolutional layer, a fifth convolutional layer, a first softmax layer, a second addition layer A2, and a first multiplication layer M1;
[0052] The input of the fourth convolutional layer is connected to the input of the fifth convolutional layer and the first input of the second addition layer A2, and its output is connected to the input of the first softmax layer; the first input of the first multiplication layer M1 is connected to the output of the first softmax layer, its second input is connected to the output of the fifth convolutional layer, and its output is connected to the second input of the second addition layer A2; the output of the second addition layer A2 serves as the output of the first attention module.
[0053] Furthermore, the second attention module includes: a sixth convolutional layer, a seventh convolutional layer, a second softmax layer, a third addition layer A3, and a second multiplication layer M2;
[0054] The input of the sixth convolutional layer is connected to the input of the seventh convolutional layer and the first input of the third addition layer A3, and its output is connected to the input of the second softmax layer; the first input of the second multiplication layer M2 is connected to the output of the second softmax layer, its second input is connected to the output of the seventh convolutional layer, and its output is connected to the second input of the third addition layer A3; the output of the third addition layer A3 serves as the output of the second attention module.
[0055] Furthermore, the loss function of the trained hydropower station fault identification model is:
[0056]
[0057] Where L is the loss function, y j Y is the output of the hydropower station fault identification model during the j-th training iteration. j Let b be the label corresponding to the j-th training iteration, b be the weight coefficient, J be the total number of training iterations, and T be the training factor, where T is a positive integer. <J, This involves summing the data from the current J-th training iteration and the JT-th training iteration.
[0058] The beneficial effect of the above further scheme is that, through weighting coefficients, the output y is respectively... j With label Y j The distance, and the total label Y in the most recent times. j With the total output y j The differences are weighted so that the loss function simultaneously considers the current training iterations when outputting y. j With label Y j The distance and the total difference in the most recent T training sessions are used to optimize the parameters of the hydropower station fault identification model.
[0059] Furthermore, the weight update formula for the trained hydropower station fault identification model is as follows:
[0060]
[0061] Among them, w * Here, w represents the updated weights, and y represents the weights before the update. j Y is the output of the hydropower station fault identification model during the j-th training iteration. j Let L be the label corresponding to the j-th training iteration, L be the loss function, th be the ratio threshold, arctan() be the trigonometric function, and e be the natural logarithm. is the partial derivative symbol, and |·| is the absolute value.
[0062] The beneficial effects of the above further scheme are: in Y j -y jWhen the difference is large, through Accelerate the degree of weight change in Y j -y j When the difference is small, by Reduce the degree of weight reduction.
[0063] Thirdly, this application provides an electronic device, characterized in that it comprises:
[0064] Memory, used to store one or more programs;
[0065] processor;
[0066] When one or more of the above programs are executed by the above processor, the wavelet decomposition-based hydropower station fault identification method as described in any of the first aspects above is implemented.
[0067] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the wavelet decomposition-based hydropower station fault identification method as described in any of the first aspects above.
[0068] In summary, the beneficial effects of this invention are as follows:
[0069] 1. After collecting the mechanical vibration signal during the operation of a hydropower station, this invention filters the mechanical vibration signal to eliminate interference signals, and performs wavelet transform on the filtered mechanical vibration data to obtain amplitude data for each frequency, further eliminating the influence of interference signals and improving recognition accuracy.
[0070] 2. This invention constructs a hydropower station fault identification model to process the amplitude data of real-time mechanical vibration signals of hydropower stations, thereby realizing the identification of hydropower station faults and solving the problem that hydropower station faults can only be determined when the sensing signal exceeds a threshold. Attached Figure Description
[0071] Figure 1 This is a flowchart of a hydropower station fault identification method based on wavelet decomposition.
[0072] Figure 2 A schematic diagram of the structure of a hydropower station fault identification model;
[0073] Figure 3 This is a schematic block diagram of an electronic device. Detailed Implementation
[0074] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.
[0075] Example 1
[0076] like Figure 1 The image shows a method for identifying faults in a hydropower station based on wavelet decomposition, as provided in Embodiment 1 of this application, comprising the following steps:
[0077] S1. Collect mechanical vibration signals during the operation of the hydropower station;
[0078] S2. Filter the mechanical vibration signal to obtain filtered mechanical vibration data;
[0079] Step S2 includes the following sub-steps:
[0080] S21. Construct a cache window and populate it with initial values;
[0081] S22. Obtain the mechanical vibration value based on the mechanical vibration signal;
[0082] The process in step S22 is to discretize the continuous signal and input the discretized mechanical vibration values into the current buffer window in the order of their occurrence, with the first collected values being input first and the last collected values being input last.
[0083] S23. Input the mechanical vibration value into the current buffer window, replacing the earliest mechanical vibration value entered in the current buffer window;
[0084] S24. Obtain the filtered value based on the median in the current cache window;
[0085] S25. Repeatedly process steps S23 to S24 according to the chronological order of acquisition of the mechanical vibration signal to obtain multiple filtered values, which constitute filtered mechanical vibration data.
[0086] The formula for calculating the filter value in step S24 is as follows:
[0087]
[0088] Among them, y i The filter value corresponding to the buffer window when the i-th input mechanical vibration value is given; x0 is the median of the current buffer window; a is the filter weight; x1 to x i x represents the i mechanical vibration values in the buffer window, arranged in chronological order; 0-1x represents the mechanical vibration value on one side of the median in the current cache window. 0+1 The mechanical vibration value on the other side of the median in the current cache window; y i-1 The filter value corresponding to the buffer window when the (i-1)th input mechanical vibration value is given.
[0089] In the buffer window, for each new mechanical vibration value input, the buffer window will exclude the oldest mechanical vibration value to ensure that the data in the buffer window is the most recently collected data. The median in the buffer window is extracted as the basis for the filter value. If the new mechanical vibration value is greater than the median, the value is increased based on the median. If the new mechanical vibration value is less than the median, the value is decreased based on the median. This makes the change of the filter value follow the change of the new mechanical vibration value, increasing the sensitivity.
[0090] S3. Perform wavelet transform on the filtered mechanical vibration data to obtain amplitude data;
[0091] S4. Match the amplitude data with the corresponding faults to construct training sample pairs;
[0092] S5. Use training samples to train the hydropower station fault identification model;
[0093] like Figure 2 As shown, the hydropower station fault identification model in step S5 includes: a first convolutional layer, a second convolutional layer, a third convolutional layer, a first additive layer A1, a max pooling layer, an average pooling layer, a first attention module, a second attention module, a concat layer, and a fully connected layer.
[0094] The first, second, and third convolutional layers are used to input amplitude data for each frequency. The first input of the first additive layer A1 is connected to the output of the first convolutional layer, its second input is connected to the output of the second convolutional layer, its third input is connected to the output of the third convolutional layer, and its output is connected to the inputs of the max pooling layer and the average pooling layer, respectively. The output of the max pooling layer is connected to the input of the first attention module. The output of the average pooling layer is connected to the input of the second attention module. The first input of the concat layer is connected to the output of the first attention module, its second input is connected to the output of the second attention module, and its output is connected to the input of the fully connected layer. The output of the fully connected layer serves as the output of the hydropower station fault identification model.
[0095] The first, second, and third convolutional layers are used to input amplitude data for each frequency. Each convolutional layer can input amplitude data for three frequencies, with each layer corresponding to the amplitude data for one frequency. After extracting feature data through the first, second, and third convolutional layers, the feature data for each frequency is added together through the first addition layer A1. Max pooling is used to perform pooling to retain salient features, and average pooling is used to perform pooling to retain overall features. The first and second attention modules are then used to enhance the feature representation capability of the model. The concat layer integrates the outputs of the first and second attention modules, and classification is performed through a fully connected layer.
[0096] The first attention module includes: a fourth convolutional layer, a fifth convolutional layer, a first softmax layer, a second addition layer A2, and a first multiplication layer M1;
[0097] The input of the fourth convolutional layer is connected to the input of the fifth convolutional layer and the first input of the second addition layer A2, and its output is connected to the input of the first softmax layer; the first input of the first multiplication layer M1 is connected to the output of the first softmax layer, its second input is connected to the output of the fifth convolutional layer, and its output is connected to the second input of the second addition layer A2; the output of the second addition layer A2 serves as the output of the first attention module.
[0098] The second attention module includes: a sixth convolutional layer, a seventh convolutional layer, a second softmax layer, a third addition layer A3, and a second multiplication layer M2;
[0099] The input of the sixth convolutional layer is connected to the input of the seventh convolutional layer and the first input of the third addition layer A3, and its output is connected to the input of the second softmax layer; the first input of the second multiplication layer M2 is connected to the output of the second softmax layer, its second input is connected to the output of the seventh convolutional layer, and its output is connected to the second input of the third addition layer A3; the output of the third addition layer A3 serves as the output of the second attention module.
[0100] The loss function for training the hydropower station fault identification model in step S5 is:
[0101]
[0102] Where L is the loss function, y j Y is the output of the hydropower station fault identification model during the j-th training iteration. j Let b be the label corresponding to the j-th training iteration, b be the weight coefficient, J be the total number of training iterations, and T be the training factor, where T is a positive integer. <J, This involves summing the data from the current J-th training iteration and the JT-th training iteration.
[0103] By using weighting coefficients, the output y is respectively... j With label Y j The distance, and the total label Y in the most recent times. j With the total output y j The differences are weighted so that the loss function simultaneously considers the current training iterations when outputting y. j With label Y j The distance and the total difference in the most recent T training sessions are used to optimize the parameters of the hydropower station fault identification model.
[0104] The weight update formula for training the hydropower station fault identification model in step S5 is as follows:
[0105]
[0106] Among them, w * Here, w represents the updated weights, and y represents the weights before the update. j Y is the output of the hydropower station fault identification model during the j-th training iteration. j Let L be the label corresponding to the j-th training iteration, L be the loss function, th be the ratio threshold, arctan() be the trigonometric function, and e be the natural logarithm. is the partial derivative symbol, and |·| is the absolute value.
[0107] In Y j -y j When the difference is large, through Accelerate the degree of weight change in Y j -y j When the difference is small, by Reduce the degree of weight reduction.
[0108] S6. The trained hydropower station fault identification model is used to process the amplitude data of the real-time mechanical vibration signal of the hydropower station to obtain the fault information.
[0109] Example 2
[0110] This application provides a hydropower station fault identification system based on wavelet decomposition, characterized in that it includes:
[0111] The acquisition module is used to collect mechanical vibration signals during the operation of the hydropower station.
[0112] The filtering module is used to filter mechanical vibration signals to obtain filtered mechanical vibration data;
[0113] The data conversion module is used to perform wavelet transform on the filtered mechanical vibration data to obtain amplitude data;
[0114] The matching module is used to match amplitude data with corresponding faults to build training sample pairs;
[0115] The training module is used to train the hydropower station fault identification model using training samples;
[0116] The data processing module is used to process the amplitude data of real-time mechanical vibration signals of hydropower stations using the trained hydropower station fault identification model to obtain fault information.
[0117] Furthermore, the hydropower station fault identification model includes: a first convolutional layer, a second convolutional layer, a third convolutional layer, a first additive layer A1, a max pooling layer, an average pooling layer, a first attention module, a second attention module, a concat layer, and a fully connected layer;
[0118] The first, second, and third convolutional layers are used to input amplitude data for each frequency. The first input of the first additive layer A1 is connected to the output of the first convolutional layer, its second input is connected to the output of the second convolutional layer, its third input is connected to the output of the third convolutional layer, and its output is connected to the inputs of the max pooling layer and the average pooling layer, respectively. The output of the max pooling layer is connected to the input of the first attention module. The output of the average pooling layer is connected to the input of the second attention module. The first input of the concat layer is connected to the output of the first attention module, its second input is connected to the output of the second attention module, and its output is connected to the input of the fully connected layer. The output of the fully connected layer serves as the output of the hydropower station fault identification model.
[0119] The beneficial effects of the above further scheme are as follows: The first, second, and third convolutional layers are used to input amplitude data for each frequency. The first, second, and third convolutional layers can input amplitude data for three frequencies, with each convolutional layer corresponding to the amplitude data for one frequency. After extracting feature data through the first, second, and third convolutional layers, the feature data for each frequency are added together through the first addition layer A1. Max pooling is used for pooling to retain significant features, and average pooling is used for pooling to retain overall features. The first and second attention modules are then used to enhance the feature representation capability of the model. The concat layer integrates the outputs of the first and second attention modules, and classification is performed through a fully connected layer.
[0120] Furthermore, the first attention module includes: a fourth convolutional layer, a fifth convolutional layer, a first softmax layer, a second addition layer A2, and a first multiplication layer M1;
[0121] The input of the fourth convolutional layer is connected to the input of the fifth convolutional layer and the first input of the second addition layer A2, and its output is connected to the input of the first softmax layer; the first input of the first multiplication layer M1 is connected to the output of the first softmax layer, its second input is connected to the output of the fifth convolutional layer, and its output is connected to the second input of the second addition layer A2; the output of the second addition layer A2 serves as the output of the first attention module.
[0122] Furthermore, the second attention module includes: a sixth convolutional layer, a seventh convolutional layer, a second softmax layer, a third addition layer A3, and a second multiplication layer M2;
[0123] The input of the sixth convolutional layer is connected to the input of the seventh convolutional layer and the first input of the third addition layer A3, and its output is connected to the input of the second softmax layer; the first input of the second multiplication layer M2 is connected to the output of the second softmax layer, its second input is connected to the output of the seventh convolutional layer, and its output is connected to the second input of the third addition layer A3; the output of the third addition layer A3 serves as the output of the second attention module.
[0124] Furthermore, the loss function of the trained hydropower station fault identification model is:
[0125]
[0126] Where L is the loss function, y j Y is the output of the hydropower station fault identification model during the j-th training iteration. j Let b be the label corresponding to the j-th training iteration, b be the weight coefficient, J be the total number of training iterations, and T be the training factor, where T is a positive integer. <J, This involves summing the data from the current J-th training iteration and the JT-th training iteration.
[0127] The beneficial effect of the above further scheme is that, through weighting coefficients, the output y is respectively... j With label Y j The distance, and the total label Y in the most recent times. j With the total output y j The differences are weighted so that the loss function simultaneously considers the current training iterations when outputting y. j With label Y j The distance and the total difference in the most recent T training sessions are used to optimize the parameters of the hydropower station fault identification model.
[0128] Furthermore, the weight update formula for the trained hydropower station fault identification model is as follows:
[0129]
[0130] Among them, w * Here, w represents the updated weights, and y represents the weights before the update. j Y is the output of the hydropower station fault identification model during the j-th training iteration. j Let L be the label corresponding to the j-th training iteration, L be the loss function, th be the ratio threshold, arctan() be the trigonometric function, and e be the natural logarithm. is the partial derivative symbol, and |·| is the absolute value.
[0131] The beneficial effects of the above further scheme are: in Y j -y j When the difference is large, through Accelerate the degree of weight change in Y j -y j When the difference is small, by Reduce the degree of weight reduction.
[0132] Example 3
[0133] Please see Figure 3 , Figure 3 This is a schematic structural block diagram of an electronic device provided in Embodiment 3 of this application.
[0134] An electronic device includes a memory 101, a processor 102, and a communication interface 103. The memory 101, processor 102, and communication interface 103 are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The memory 101 can be used to store software programs and modules. The processor 102 executes the software programs and modules stored in the memory 101 to perform various functional applications and data processing. The communication interface 103 can be used for signaling or data communication with other node devices.
[0135] The memory 101 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.
[0136] The processor 102 can be an integrated circuit chip with signal processing capabilities. The processor 102 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0137] In summary, this invention collects mechanical vibration signals from hydropower stations during operation, filters these signals to eliminate interference, and performs wavelet transform on the filtered vibration data to obtain amplitude data for each frequency, further eliminating the influence of interference and improving recognition accuracy. Furthermore, this invention constructs a hydropower station fault identification model to process the amplitude data of real-time mechanical vibration signals from hydropower stations, enabling fault identification and solving the problem that faults can only be determined when the sensor signal exceeds a threshold.
Claims
1. A method for fault identification in hydropower stations based on wavelet decomposition, characterized in that, Includes the following steps: S1. Collect mechanical vibration signals during the operation of the hydropower station; S2. Filter the mechanical vibration signal to obtain filtered mechanical vibration data; S3. Perform wavelet transform on the filtered mechanical vibration data to obtain amplitude data; S4. Match the amplitude data with the corresponding faults to construct training sample pairs; S5. Training samples are used to train the hydropower station fault identification model, which includes: a first convolutional layer, a second convolutional layer, a third convolutional layer, a first additive layer A1, a max pooling layer, an average pooling layer, a first attention module, a second attention module, a concat layer, and a fully connected layer. The first, second, and third convolutional layers are used to input amplitude data for each frequency. The first input of the first addition layer A1 is connected to the output of the first convolutional layer, its second input is connected to the output of the second convolutional layer, its third input is connected to the output of the third convolutional layer, and its output is connected to the input of the max pooling layer and the average pooling layer, respectively. The output of the max pooling layer is connected to the input of the first attention module. The output of the average pooling layer is connected to the input of the second attention module. The first input of the concat layer is connected to the output of the first attention module, its second input is connected to the output of the second attention module, and its output is connected to the input of the fully connected layer. The output of the fully connected layer serves as the output of the hydropower station fault identification model. The first attention module includes: a fourth convolutional layer, a fifth convolutional layer, a first softmax layer, a second addition layer A2, and a first multiplication layer M1. The input of the fourth convolutional layer is connected to the input of the fifth convolutional layer and the first input of the second addition layer A2, and its output is connected to the input of the first softmax layer; the first input of the first multiplication layer M1 is connected to the output of the first softmax layer, its second input is connected to the output of the fifth convolutional layer, and its output is connected to the second input of the second addition layer A2; the output of the second addition layer A2 serves as the output of the first attention module; the second attention module includes: a sixth convolutional layer, a seventh convolutional layer, a second softmax layer, a third addition layer A3, and a second multiplication layer M2; The input of the sixth convolutional layer is connected to the input of the seventh convolutional layer and the first input of the third addition layer A3, and its output is connected to the input of the second softmax layer; the first input of the second multiplication layer M2 is connected to the output of the second softmax layer, its second input is connected to the output of the seventh convolutional layer, and its output is connected to the second input of the third addition layer A3; the output of the third addition layer A3 serves as the output of the second attention module. The loss function for training the hydropower station fault identification model is: in, For loss function, For the first The output of the hydropower station fault identification model during the second training session. For the first The labels corresponding to each training iteration These are the weighting coefficients of the loss function. This represents the current total number of training iterations. As training factors, It is a positive integer. , To make the current number Next and first The data from the number of training iterations between each iteration are summed. The weight update formula for training the hydropower station fault identification model is as follows: in, For the updated model weights, The model weights before the update. For the first The output of the hydropower station fault identification model during the second training session. For the first The labels corresponding to each training iteration For loss function, The ratio threshold, For trigonometric functions, It is the natural logarithm. The sign of the partial derivative. It is the absolute value; S6. The trained hydropower station fault identification model is used to process the amplitude data of the real-time mechanical vibration signal of the hydropower station to obtain the fault information.
2. The hydropower station fault identification method based on wavelet decomposition according to claim 1, characterized in that, Step S2 includes the following sub-steps: S21. Construct a cache window and populate it with initial values; S22. Obtain the mechanical vibration value based on the mechanical vibration signal; S23. Input the mechanical vibration value into the current buffer window, replacing the earliest mechanical vibration value entered in the current buffer window; S24. Obtain the filtered value based on the median in the current cache window; S25. Repeatedly process steps S23-S24 to obtain multiple filtered values from the mechanical vibration signal in chronological order of acquisition time, thus forming filtered mechanical vibration data.
3. The hydropower station fault identification method based on wavelet decomposition according to claim 2, characterized in that, The formula for calculating the filter value in step S24 is as follows: in, For the first The filter value corresponding to the buffer window when the mechanical vibration value is input for the first time; The median in the current cache window; These are the filter weights; to Arranged in chronological order in the cache window One mechanical vibration value; This represents the mechanical vibration value to one side of the median in the current cache window. The mechanical vibration value on the other side of the median in the current cache window; For the first The filter value corresponding to the buffer window when the mechanical vibration value is input for the first time.
4. A hydropower station fault identification system employing the wavelet decomposition method according to any one of claims 1-3, characterized in that, include: The acquisition module is used to collect mechanical vibration signals during the operation of the hydropower station. The filtering module is used to filter mechanical vibration signals to obtain filtered mechanical vibration data; The data conversion module is used to perform wavelet transform on the filtered mechanical vibration data to obtain amplitude data; The matching module is used to match amplitude data with corresponding faults to build training sample pairs; The training module is used to train the hydropower station fault identification model using training samples; The data processing module is used to process the amplitude data of real-time mechanical vibration signals of hydropower stations using the trained hydropower station fault identification model to obtain fault information.
5. An electronic device, characterized in that, include: Memory, used to store one or more programs; processor; When the processor executes the one or more programs, it implements the wavelet decomposition-based hydropower station fault identification method as described in any one of claims 1-3.
6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the wavelet decomposition-based hydropower station fault identification method as described in any one of claims 1-3.
Citation Information
Patent Citations
Electromechanical equipment fault diagnosis method and system
CN110174281A
Bearing fault diagnosis method based on improved convolutional capsule network
CN113255882A