Hydroelectric generating set vibration trend prediction method based on WPT-SCA-GRU
The vibration signal of the hydroelectric unit is preprocessed and feature extraction through the WPT-SCA-GRU method, which solves the vibration signal processing problem in the prior art, improves the prediction accuracy, and ensures the stable operation of the unit.
Patent Information
- Application Number
- CN202510197650.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-21
- Publication Date
- 2025-06-20
AI Technical Summary
The existing vibration prediction methods for hydroelectric units are difficult to effectively remove interference components when processing nonlinear vibration signals and extract complex features, resulting in insufficient prediction accuracy.
Using the WPT-SCA-GRU-based method, the interference components of the vibration signal are removed through wavelet packet transformation, and the optimal parameters of the GRU model are searched using the GRU model, and finally the decomposed signal is predicted using the GRU model.
Effectively remove interference components in vibration data, extract nonlinear degradation characteristics, improve the accuracy of vibration trend prediction, and ensure the stable operation of the unit.
Smart Images
Figure CN120180865A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of vibration trend prediction of hydro-generating units, in particular to a vibration trend prediction method for hydro-generating units based on WPT-SCA-GRU. Background Art
[0002] As the core equipment of a hydropower station, a hydro-generating unit mainly drives a water turbine to rotate through water energy. During the operation of the unit, it will be affected by factors such as hydraulic, mechanical, and electromagnetic, which may cause abnormal vibration of the unit. Vibration is one of the common fault indicators during the operation of a hydro-generating unit. Abnormal vibration often affects the operation safety of the equipment and may even lead to equipment damage or accidents. Therefore, a vibration trend prediction method is needed to predict the vibration of the unit in advance and maintain the stable operation of the unit.
[0003] Common vibration prediction methods for hydro-generating units mainly include prediction methods based on physical models and prediction methods based on data-driven. Prediction methods based on physical models mainly use mathematical models to describe the vibration characteristics during the operation of the unit and predict future vibration trends. Data-driven machine learning methods mainly use a large amount of historical vibration data as input, including information such as the amplitude, frequency spectrum distribution, and time series of vibration signals, to train the model and predict the future state of the system. Currently, the defects and deficiencies of the existing technologies are as follows:
[0004] Prediction methods based on physical models require in-depth understanding of the structure, material properties, working conditions, etc. of the unit. The model complexity is relatively high, and its accuracy also highly depends on the accuracy of the model. In some cases, it can provide relatively accurate and theoretically supported prediction results, but problems such as its complexity, real-time performance, and dependence on parameter accuracy limit its wide application in actual use.
[0005] Existing mainstream data-driven prediction methods are mainly divided into traditional machine learning and deep learning methods. Traditional machine learning methods need to rely on expert experience for feature selection and extraction and cannot handle high-dimensional data and data with non-linear relationships well. Deep learning methods do not require manual design and selection of features and can directly extract useful feature representations from raw data. However, in the actual working environment, a hydro-generating unit is affected by various factors, and the unit vibration signal shows non-linear characteristics, which contain certain interference components. Common methods are difficult to extract the complex features therein. How to extract and predict the next vibration situation from the existing non-linear vibration signals and complete the vibration trend prediction task of the unit is particularly important. Summary of the Invention
[0006] The object of the present invention is to overcome the above deficiencies and provide a vibration trend prediction method for hydro-generator units based on WPT-SCA-GRU, which can effectively remove the interference components in the vibration data, extract the non-linear degradation characteristics in the vibration signals of hydro-generator units, reflect the changes in the vibration trends of the units, and has a certain practical significance for maintaining the stable operation of the units.
[0007] To solve the above technical problems, the technical solution adopted by the present invention is: a vibration trend prediction method for hydro-generator units based on WPT-SCA-GRU, which includes the following steps:
[0008] Step S1, WPT decomposition: Collect the upper guide vibration signals of the hydro-generator unit monitored by the sensor, perform wavelet packet transform on the data to obtain subsequences of low-frequency signals and high-frequency signals, and perform maximum-minimum normalization on the above subsequences;
[0009] Step S2, vibration trend prediction of the hydro-generator unit: Input the subsequences finally obtained in Step S1 into the GRU model obtained by the sine-cosine optimization algorithm. After the prediction indexes of each component are de-normalized, they are weighted and superimposed to obtain the final vibration trend prediction result;
[0010] Step S3, model performance evaluation: Compare the prediction result with the actual value, and repeat the experiment using the data in different directions of the upper guide to evaluate the prediction performance of the model.
[0011] Further, the specific content of Step S1 is: Collect the upper guide swing signals of the hydro-generator unit monitored by the sensor, perform wavelet packet transform on the data, decompose the vibration signal using wavelet transform, decompose the high-frequency signal into low-frequency signals, use the multiple low-frequency subsequences obtained after decomposition as the input variables of the prediction model, and then perform maximum-minimum normalization on the above subsequences to scale the data to between [0, 1].
[0012] Furthermore, in Step S1, the specific steps of performing wavelet packet transform on the data are as follows:
[0013] S1.1, Given the orthogonal scaling function φ(t) and the wavelet function Then the two-scale difference equation of wavelet packet decomposition can be expressed as:
[0014]
[0015] Among them, h(k) is the low-pass filter, g(k) is the high-pass filter, and the two satisfy the relationship g(k) = (-1)h(1 - k);
[0016] S1.2, The wavelet packet transform coefficients can be expressed as:
[0017]
[0018] where j is the wavelet packet decomposition level, m is the wavelet coefficient number, k is the decomposition level, n is the number of nodes at the j-th level, and d j,n is the signal in the n-node frequency band at the j-th level;
[0019] S1.3. Reconstruct the signal in the n-node frequency band at the j-th level. The energy factor of each frequency band is defined as:
[0020] E[j,n] = Σ(d j,n ) 2 .
[0021] Furthermore, in step S1, the formula for maximum-minimum normalization is as follows:
[0022]
[0023] where x and are the data before and after normalization respectively, x min and x max are the minimum and maximum values of the data respectively.
[0024] Further, step S2 is specifically as follows: Establish a GRU prediction model, use the sine-cosine optimization algorithm to find the optimal model parameters suitable for the model, input the obtained subsequences into the optimal parameter model respectively to obtain the prediction indicators of each sequence, and then perform denormalization on the prediction indicators respectively and weighted superposition to obtain the final vibration trend prediction result.
[0025] Furthermore, in step S2, the specific steps of using the sine-cosine optimization algorithm to find the optimal model parameters are as follows:
[0026] Based on the mathematical model of sine and cosine functions, randomly generate an initial set according to prior information. The position update equation of the particles in the set is as follows:
[0027]
[0028] where represents the position vector of the i-th particle at the t-th iteration, is the optimal solution of the i-th particle at the t-th iteration; r1 is a linearly decreasing conversion parameter, and its expression is:
[0029]
[0030] where t represents the current iteration number, T maxrepresents the total number of iterations, a is a constant greater than zero; r2 is a random variable in the range of [0, 2π]; r3 is a random variable in the range of [-2, 2], which is used to enhance (r3 > 1) or weaken (r3 < 1) the influence of the optimal solution of the i-th particle on the position vector of the current solution; r4 controls the switching between the sine and cosine functions in the equation.
[0031] Further, in step S2, the process of inputting the obtained subsequences into the optimal parameter model respectively to obtain the prediction indexes of each sequence is as follows:
[0032]
[0033] Among them, x t is the input of the hidden layer at time t, h t is the output of the current layer at time t, r t , z t are the reset gate and update gate respectively, is the set of the input x t at time t and the output h t-1 at time t - 1, σ and tanh represent the sigmoid function and hyperbolic tangent function respectively, W r , W z , W are the corresponding weight matrices respectively.
[0034] Further, in step S2, the formula for inverse normalization of the prediction indexes is as follows:
[0035]
[0036] Among them, and y represent the data before and after inverse normalization respectively, x min and x max are the minimum and maximum values of the data respectively.
[0037] Further, step S2 also includes the following process:
[0038] Stack the degradation indexes extracted by the model together to obtain the result, and the calculation process is as follows:
[0039] Y = y1 + y2 + … + y n
[0040] Among them, Y is the prediction result, y n is the n-th degradation index.
[0041] Further, in step S2, the GRU hidden layer dimension obtained by using the sine-cosine optimization algorithm is 11, the dropout rate is 0.3, the root mean square function is selected as the loss function during training, the Adam algorithm is used as the optimizer, the training batch size is 64, the learning rate is 0.001, and the number of loops is 300.
[0042] Beneficial effects of the present invention:
[0043] 1. The present invention introduces the WPT method, which can denoise and decompose the non-linear vibration signals of the hydropower unit to obtain relatively stable signals. At the same time, the SCA is used to search for the optimal model parameters, effectively avoiding the uncertainty influence caused by manual parameter selection. Finally, the decomposed signals are predicted by the GRU model with the optimal parameters, making full use of the time information to obtain sequence features and improving the prediction accuracy.
[0044] 2. The present invention can accurately predict the vibration trend of the hydropower unit, which is of great significance for preventing excessive amplitudes of the equipment and ensuring the safe operation of the equipment.
[0045] 3. The present invention can effectively remove the interference components in the vibration data, extract the non-linear degradation features in the vibration signals of the hydropower unit, reflect the changes in the vibration trend of the unit, and has certain practical significance for maintaining the stable operation of the unit. Description of the drawings
[0046] Figure 1 is the schematic diagram of the step flow of the present invention;
[0047] Figure 2 is the signal decomposition diagram of wavelet packet transform;
[0048] Figure 3 is the flow chart of the sine-cosine algorithm;
[0049] Figure 4 is the GRU structure diagram;
[0050] Figure 5 is the schematic diagram of the real-time prediction process of the present invention;
[0051] Figure 6 is the prediction result diagram of the present invention for the upper guide swing in the X and Y directions. Detailed implementation manners
[0052] The present invention will be further described in detail below with reference to the drawings and specific embodiments.
[0053] Embodiment 1: Aiming at the problems of non-stationary and non-linear vibration signals of hydropower units, this embodiment proposes a vibration trend prediction method for hydropower units based on WPT-SCA-GRU, which can extract time series features from the vibration signals of hydropower units by using time information, further predict the vibration change trend of the units, prevent the units from experiencing severe vibration conditions, and improve the reliability of unit operation. Combining Figures 1 to 6 , the implementation and effect verification of the present invention are elaborated as follows:
[0054] As shown in the appendix Figure 1 , this embodiment provides a vibration trend prediction method for hydropower units based on WPT-SCA-GRU, including the following steps:
[0055] Collect the vibration signals of the upper guide of the hydropower unit monitored by sensors, perform wavelet packet transform on the data to obtain subsequences of low-frequency signals and high-frequency signals, normalize the maximum and minimum values of the above subsequences, and input them into the GRU model obtained by the sine-cosine optimization algorithm. After the prediction indexes of each component are de-normalized, they are weighted and superimposed to obtain the final vibration trend prediction result. The prediction result is compared with the actual value, and at the same time, the data in different directions of the upper guide are used to repeat the experiment to evaluate the prediction performance of the model.
[0056] Specifically, acceleration sensors are installed in the X and Y directions of the upper guide bearing housing to monitor the vibration signals of the upper guide bearing. Compared with other sensors, acceleration sensors have the characteristics of high precision, small size, long life, easy installation, and good stability.
[0057] Furthermore, perform wavelet packet transform (WPT) on the monitored data. WPT can decompose complex non-linear signals, and at the same time can remove the clutter factors in the signals, making the signals tend to be stable, better understanding and analyzing the inherent properties and dynamic characteristics of the signals. The decomposed sub-signals are shown in the appendix Figure 2 , and its calculation process is as follows:
[0058] a. Given the orthogonal scaling function φ(t) and the wavelet function Then the two-scale difference equation of wavelet packet decomposition can be expressed as:
[0059]
[0060] where h(k) is the low-pass filter and g(k) is the high-pass filter, and the two satisfy the relationship g(k) = (-1)h(1 - k).
[0061] b. The wavelet packet transform coefficients can be expressed as:
[0062]
[0063] Among them, j is the number of wavelet packet decomposition layers, m is the wavelet coefficient number, and n is the number of nodes in the j-th layer.
[0064] c. Reconstruct the frequency band signals of the n nodes in the j-th layer. The energy factor of each frequency band is defined as:
[0065] E[j,n] = ∑(d j,n ) 2
[0066] Furthermore, perform maximum-minimum normalization on the decomposed sub-signals, compress the data to between [0,1], which can effectively eliminate the influence of abnormal data and enable comparison between data indicators. The steps of normalization are as follows:
[0067]
[0068] Among them, x and are the data before and after normalization respectively, x min and x max are the minimum and maximum values of the data respectively.
[0069] As shown in the appendix Figure 3 Use the sine-cosine optimization algorithm (SCA) to search for the optimal hidden layer dimension and regularization rate of GRU. The SCA algorithm can make multiple initial random solutions move towards or away from the optimal solution through the mathematical models of sine and cosine, effectively avoiding local optima and making the parameters converge to the global optimum. Its calculation process is as follows:
[0070] Based on the mathematical models of sine and cosine functions, randomly generate an initial set according to prior information. The position update equation of the particles in the set is as follows:
[0071]
[0072] Among them, represents the position vector of the i-th particle at the t-th iteration, is the optimal solution of the i-th particle at the t-th iteration. r1 is a linearly decreasing conversion parameter, and its expression is:
[0073]
[0074] Among them, t represents the current iteration number, T max represents the total number of iterations, a is a constant greater than zero; r2 is a random variable in [0, 2π]; r3 is a random variable in [-2, 2], which is used to enhance (r3 > 1) or weaken (r3 < 1) the influence of the optimal solution of the i-th particle on the position vector of the current solution; r4 controls the switching of the equation between sine and cosine functions.
[0075] Further, the decomposed sub-signals are respectively input into the GRU model with adjusted parameters. By learning the time-related sequences in the data, the model can better capture the degradation features therein. The GRU model structure is as shown in Appendix Figure 4 and is as follows:
[0076]
[0077] where x t is the input of the hidden layer at time t, h t is the output of the current layer at time t, r t , z t are the reset gate and the update gate respectively, is the set of the input x t at time t and the output h t-1 at time t - 1. σ and tanh respectively represent the sigmoid function and the hyperbolic tangent function. W r , W z , are the corresponding weight matrices respectively, and b r , b z , are the corresponding bias matrices respectively.
[0078] Further, the inverse normalization is performed on the prediction results of each component to better compare the predicted values with the actual values. The specific steps are as follows:
[0079]
[0080] where and y respectively represent the data before and after inverse normalization.
[0081] Further, the degradation indices extracted by the model are stacked together to obtain the result. The calculation process is as follows:
[0082] Y = y1 + y2 + … + y n
[0083] where Y is the prediction result, and y n is the nth degradation index.
[0084] Further, the process of real-time prediction of the vibration signal of the unit is as shown in Appendix Figure 5 . The collected data set is divided into a training set and a test set after preprocessing. The training set is trained, and the optimal model parameters are saved. The test set is imported into the trained model for vibration trend prediction. Finally, the predicted values are compared with the actual values, and the data in different directions of the upper guide are used to repeat the experiment to evaluate the prediction performance of the model.
[0085] Example verification and analysis:
[0086] The effectiveness of the present invention is verified by experimental analysis of the vibration data in the X and Y directions of the upper guide bearing of Unit 5 of a domestic hydropower station. Excluding the data with zero vibration during the shutdown process, taking the upper guide swing of the upper guide bearing during the steady-state operation of the unit as the research object, one data point is collected every 1 hour, a total of 4400 sample points are collected, the first 4000 sample points are selected for training, and the last 400 sample points are used for testing. When training, the mean square root function (MSE) is selected as the loss function, the Adam algorithm is used as the optimizer, the training batch size is 64, the learning rate is 0.001, and the number of cycles is 300. The decomposition layer of wavelet packet transform is selected as two layers. In the SCA algorithm, the initial population is set to 20, the maximum number of iterations is 10. In the GRU model, the input layer size is 24, the number of layers is 2, the GRU hidden layer obtained by using the SCA algorithm is 11, and the dropout rate is 0.3. At the same time, the mean absolute error (MAE) and the root mean square error (RMSE) are used as the indicators to evaluate the performance of the model. The smaller the values of MAE and RMSE, the better the model prediction effect. The calculation formulas are as follows:
[0087]
[0088] where n is the total number of samples, y i represents the actual remaining life percentage of the i-th sample, represents the predicted value of the remaining life of the i-th sample.
[0089] The predicted results of the vibration trend of the hydropower unit are as Figure 6 shown. It can be seen from the figure that the numerical values of the vibration trends in the X and Y directions of the upper guide of the unit obtained by model prediction are basically consistent with the actual values, indicating that the method of the present invention can effectively predict the vibration trend of the unit. In addition, three groups of models are selected for comparison to evaluate the model performance. After the Y-direction swing data are used for experiments respectively, the prediction results are shown in Table 1. It can be seen from the table that the prediction accuracy of the present invention is better than that of the comparative models, proving that the method of the present invention has certain advantages in the prediction of the vibration trend of hydropower units.
[0090] Table 1 Comparison of prediction results
[0091]
[0092] The above embodiments are only the preferred technical solutions of the present invention and should not be regarded as limitations on the present invention. The protection scope of the present invention should be the technical solutions recorded in the claims, including the equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, the equivalent replacement improvements within this scope are also within the protection scope of the present invention.
Claims
1. A method for predicting vibration trend of hydropower units based on WPT-SCA-GRU, characterized by: It includes the following steps: Step S1, WPT decomposition: collect the upper guide vibration signal of the hydropower unit obtained by sensor monitoring, perform wavelet packet transform on the data, obtain subsequences of low-frequency signals and high-frequency signals, and perform maximum and minimum value normalization on the above subsequences; Step S2, prediction of vibration trend of hydropower unit: input the subsequence finally obtained in step S1 into the GRU model obtained by the sine-cosine optimization algorithm, and after the obtained component prediction indicators are denormalized, weighted superposition is performed to obtain the final vibration trend prediction result; Step S3, model performance evaluation: Compare the predicted results with the actual values, and repeat the experiment using data from different directions to evaluate the model prediction performance.
2. According to a WPT-SCA-GRU-based hydropower unit vibration trend prediction method according to claim 1, it is characterized in that: The step S1 specifically includes: collecting the upper guide swing signal of the hydropower unit obtained by sensor monitoring, performing wavelet packet transform on the data, decomposing the vibration signal by using wavelet transform, decomposing the high-frequency signal into low-frequency signal, using the multiple low-frequency subsequences obtained after decomposition as the input of the prediction model, and then normalizing the maximum and minimum values of the above subsequences, and scaling the data to between [0,1].
3. A method for predicting vibration trend of a hydropower unit based on WPT-SCA-GRU according to claim 2, characterized in that: In step S1, the specific steps of performing wavelet packet transform on the data are as follows: S1.1, given the orthogonal scaling function φ(t) and the wavelet function Then the two-scale difference equation of wavelet packet decomposition can be expressed as: Among them, h(k) is a low-pass filter, g(k) is a high-pass filter, and the two satisfy the relationship g(k)=(-1)h(1-k); S1.2, wavelet packet transform coefficients can be expressed as: Among them, j is the number of wavelet packet decomposition layers, m is the wavelet coefficient number, k is the number of decomposition layers, n is the number of nodes in the jth layer, and d j,n is the frequency band signal of n nodes in the jth layer; S1.3, reconstruct the frequency band signal of the n-node in the j-th layer, and the energy factor of each frequency band is defined as: E[j,n]=∑(d j,n ) 2 。 4. The method for predicting vibration trend of a hydropower unit based on WPT-SCA-GRU according to claim 2 is characterized in that: In step S1, the formula for normalizing the maximum and minimum values is as follows: Among them, x and are the data before and after normalization, respectively, min and x max are the minimum and maximum values of the data respectively.
5. The method for predicting vibration trend of a hydropower unit based on WPT-SCA-GRU according to claim 1 is characterized in that: The step S2 is specifically as follows: establishing a GRU prediction model, using the sine-cosine optimization algorithm to find the optimal model parameters of the adaptation model, inputting the subsequences obtained above into the optimal parameter model respectively, obtaining the prediction index of each sequence, and then denormalizing the prediction index respectively, and weighted superposition to obtain the final vibration trend prediction result.
6. The method for predicting vibration trend of a hydropower unit based on WPT-SCA-GRU according to claim 5 is characterized in that: In step S2, the specific steps of using the sine-cosine optimization algorithm to find the optimal model parameters of the adaptation model are as follows: Based on the mathematical model of sine and cosine functions, the initial set is randomly generated according to prior information, and the position update equation of the particles in the set is as follows: in, represents the position vector of the ith particle at the tth iteration, is the optimal solution of the i-th particle at the t-th iteration; r1 is a linearly decreasing conversion parameter, and its expression is: Among them, t represents the current iteration number, T max represents the total number of iterations, a is a constant greater than zero; r2 is a random variable in [0,2π]; r3 is a random variable in [-2,2], to enhance (r3>1) or weaken (r3<1) the influence of the optimal solution of the ith particle on the position vector of the current solution; r4 controls the equation to switch between sine and cosine functions.
7. The method for predicting vibration trend of a hydropower unit based on WPT-SCA-GRU according to claim 5 is characterized in that: In step S2, the subsequences obtained above are respectively input into the optimal parameter model to obtain the prediction index of each sequence as follows: Among them, x t is the input of the hidden layer at time t, h t is the output of the current layer at time t, r t 、z t They are reset gate and update gate respectively. Enter x for time t t Output h at time t-1 t-1 The set of σ and tanh represent the sigmoid function and the hyperbolic tangent function respectively, W r , W z , W are the corresponding weight matrices respectively.
8. The method for predicting vibration trend of a hydropower unit based on WPT-SCA-GRU according to claim 5 is characterized in that: In step S2, the formulas for denormalizing the prediction indicators are as follows: in, and y represent the data before and after denormalization, respectively. min and x max are the minimum and maximum values of the data respectively.
9. The method for predicting vibration trend of a hydropower unit based on WPT-SCA-GRU according to claim 5 is characterized in that: Step S2 also includes the following process: The degradation indicators extracted by the model are stacked together to obtain the result. The calculation process is as follows: Y=y1+y2+…+y n Among them, Y is the prediction result, y n is the nth degradation index.
10. A method for predicting vibration trend of a hydropower unit based on WPT-SCA-GRU according to claim 5, characterized in that: In step S2, the hidden layer dimension of the GRU obtained by the sine-cosine optimization algorithm is 11, the dropout rate is 0.3, the root mean square function is selected as the loss function during training, the Adam algorithm is used as the optimizer, the training batch size is 64, the learning rate is 0.001, and the number of cycles is 300.