Hydroelectric generating set state prediction method and system based on high-frequency component smooth sampling
By decomposing and fitting the vibration signal of the hydropower unit using a high-frequency component smoothing sampling method, and training it with an LSTM model, the problems of insufficient timeliness and predictive ability of hydropower unit condition monitoring are solved, and more accurate condition prediction and risk identification are achieved.
Patent Information
- Application Number
- CN202511053457.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-30
- Publication Date
- 2025-11-14
AI Technical Summary
Existing hydropower unit condition monitoring methods suffer from poor timeliness, low information density, and weak predictive ability, failing to effectively capture high-frequency variation information and resulting in inaccurate judgment of condition trends.
A method based on high-frequency component smoothing sampling is adopted. The hyperparameters of TVF-EMD are optimized by fixed-value cyclic search, the vibration signal is decomposed, B-spline fitting is performed for smoothing and sparse sampling is performed, and LSTM model is combined for training and prediction.
It improves the timeliness and information density of hydropower unit condition monitoring, enhances prediction accuracy, supports real-time condition assessment and risk identification, and is suitable for modeling combinations of various monitoring variables.
Smart Images
Figure CN120950868A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of hydropower unit monitoring technology, and more specifically, to a method and system for predicting the state of hydropower units based on high-frequency component smoothing sampling. Background Technology
[0002] As the core equipment of a hydropower station, the operating status of the hydropower unit directly affects the safety, stability, and power generation efficiency of the entire hydropower station.
[0003] In existing technologies, most methods for monitoring the condition of hydropower units rely on fixed-period data collection and manual judgment, which has at least three shortcomings. First, it suffers from poor timeliness; traditional sampling strategies fail to fully extract information from critical time periods, potentially missing early signs of faults. Second, it has low information density; conventional sampling frequencies may not effectively capture high-frequency variations in the data, leading to inaccurate judgments of condition trends. Third, it has weak predictive ability; existing models are mostly based on mean trend analysis, ignoring non-stationary characteristics during operation.
[0004] Therefore, there is an urgent need for a new state prediction method that combines high-frequency characteristics of operational data and has adaptive resolution capabilities to improve the intelligence and precision of hydropower unit operation and management. Summary of the Invention
[0005] The purpose of this application is to overcome the shortcomings of existing hydropower unit condition monitoring methods, such as poor timeliness, low information density, and weak prediction ability, and to provide a hydropower unit condition prediction method and system based on high-frequency component smooth sampling.
[0006] The objective of this application is achieved through the following technical solution: Firstly, this application proposes a hydropower unit state prediction method based on high-frequency component smoothing sampling, comprising the following steps: S1: Acquire the vibration signal of the target hydroelectric generator unit; S2: Optimize hyperparameters in TVF-EMD based on the fixed-value cyclic search method; S3: The vibration signal is decomposed using the optimized TVF-EMD to obtain a set of intrinsic mode functions containing multiple high-frequency signal components; S4: Perform cubic B-spline fitting smoothing on the high-frequency signal components to obtain a smoothed signal; S5: Perform sparse sampling on the smoothed signal at fixed intervals to obtain a signal sequence; S6: Input the sub-signals into the LSTM model for training and state prediction, and sum them to obtain the final prediction result; the sub-signals include intrinsic mode functions and signal sequences.
[0007] By adopting the above technical solution, this method addresses the problem of high-frequency components in the vibration signal of hydropower units affecting prediction accuracy. First, it uses a fixed-value cyclic search to optimize key parameters in the time-varying filter empirical mode decomposition (TVF-EMD) to perform multi-scale decomposition of the vibration signal. Then, it introduces B-spline curves for smoothing the high-frequency components and performs interval sampling to enhance the effectiveness of training samples. Finally, the processed sub-signals are input into a Long Short-Term Memory (LSTM) network for training and prediction. Experiments show that this method can more effectively capture state evolution characteristics and significantly improve prediction accuracy compared to traditional methods. It effectively overcomes the shortcomings of existing hydropower unit state monitoring methods, such as poor timeliness, low information density, and weak prediction capabilities.
[0008] Preferably, step S2 includes: Initialize the bandwidth parameter n and the center frequency ; Combine the bandwidth parameter n with the center frequency Perform the update; decompose and calculate the average discrete entropy; determine if the loop termination condition has been met; output the optimal hyperparameter bandwidth parameter n and center frequency. combination.
[0009] Preferably, the vibration signal obtained by the optimized TVF-EMD decomposition is a fuzziness minimization index. ,in The entropy that represents a true signal.
[0010] Preferably, step S4 includes: B-spline functions employ cubic non-uniform node interpolation, and their basis functions are defined as follows:
[0011] in, Represents second-order spline basis functions. Represents the node sequence.
[0012] Preferably, the fixed interval is determined based on the sampling frequency fs and the target frequency band bandwidth, satisfying the Nyquist criterion:
[0013] in, This represents the high-frequency limit of the target signal.
[0014] Preferably, step S6 includes: The LSTM model includes an input layer, two hidden state units, and an output layer. Its activation function is tanh, and its loss function is mean squared error. .
[0015] Preferably, the sub-signals are divided into low-frequency trend signals and high-frequency disturbance signals, both of which participate in the training to preserve the full picture of state changes.
[0016] Preferably, the sub-signals are divided into a training set and a test set, with the ratio of the training set to the test set being 7:3, and a rolling time window mechanism is used to improve generalization ability.
[0017] Preferably, the final prediction result can be determined by setting a threshold value. To achieve early warning and alarm of operational risks; if If so, an alarm will be triggered.
[0018] Secondly, this application proposes a hydropower unit state prediction system based on high-frequency component smoothing sampling. The hydropower unit state prediction system based on high-frequency component smoothing sampling includes a processor and a memory. The memory stores a computer program, which is loaded and executed by the processor to implement the hydropower unit state prediction method based on high-frequency component smoothing sampling as described in any of the first aspects.
[0019] The main solution and its various further alternatives described above can be freely combined to form multiple solutions, all of which are solutions that can be adopted and are claimed in this application; furthermore, the (non-conflicting alternatives) can also be freely combined with each other and with other alternatives. Those skilled in the art, after understanding the solution of this application, will realize from the prior art and common general knowledge that there are many combinations, all of which are technical solutions to be protected in this application, and will not be exhaustively listed here.
[0020] The beneficial effects of this application are as follows: 1. Accurately detect key change segments to improve model sensitivity; 2. Reduce redundant sampling points and improve computational efficiency; 3. Enhance the accuracy of trend prediction and support real-time status assessment and risk identification; 4. It is applicable to modeling combinations of various monitoring variables and has strong versatility. Attached Figure Description
[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0022] Figure 1 This is a schematic diagram of the process structure of the hydropower unit state prediction method based on high-frequency component smooth sampling in the embodiments of this application.
[0023] Figure 2 The process of optimizing hyperparameters for TVF-EMD.
[0024] Figure 3 The execution steps of the hydropower unit state prediction method based on high-frequency component smooth sampling are described below.
[0025] Figure 4 For default parameters ( The TVF-EMD algorithm decomposes the oscillating signal under the conditions of n=0.1 and n=3.
[0026] Figure 5 After parameter optimization ( The TVF-EMD algorithm (n=0.3, n=14) decomposes the oscillating signal.
[0027] Figure 6 This represents the prediction results of the training set on the LSTM model.
[0028] Figure 7 This represents the prediction results of the test set on the LSTM model. Detailed Implementation
[0029] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.
[0030] Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0031] This invention provides a method for predicting the state of hydropower units based on high-frequency component smoothing sampling.
[0032] refer to Figure 1 and Figure 2 A hydropower unit state prediction method based on high-frequency component smoothing sampling includes the following steps: S1: Obtain the vibration signal x(t) of the target hydropower unit.
[0033] It should be understood that after obtaining the vibration signal x(t), it needs to be preprocessed. The preprocessing process is existing technology and will not be described in detail here.
[0034] S2: Based on the fixed-value cyclic search method, the hyperparameters (bandwidth parameter n and center frequency) in TVF-EMD (Time-Varying Filtered Empirical Mode Decomposition) are analyzed. ) to perform optimization.
[0035] refer to Figure 3 First, initialize the bandwidth parameter n and the center frequency. To make the bandwidth parameter n=3, the center frequency =0.1; then set the bandwidth parameter n and the center frequency... The process involves updating the parameters; then decomposing and calculating the average discrete entropy; subsequently determining whether the loop termination condition has been met; finally, the optimal hyperparameter bandwidth parameter n and center frequency are output. Combination, where bandwidth parameter n=14, center frequency =0.3.
[0036] The discrete entropy values of the decomposition results before and after hyperparameter optimization are shown in Table 1 below: Table 1
[0037] As shown in Table 1, after optimizing the hyperparameters in TVF-EMD, the average discrete entropy of the signal components decreased, which confirms that hyperparameter optimization can effectively reduce the complexity of the decomposed signal components and improve the energy concentration and demixing accuracy of the signal components.
[0038] Initialize the bandwidth parameter n=3 and the center frequency When = 0.1, the TVF-EMD decomposition results of the oscillation signal are as follows: Figure 4 As shown. The optimal output hyperparameter bandwidth parameter n=14 and center frequency... After the value is 0.3, the TVF-EMD decomposition results of the oscillation signal are as follows: Figure 5 As shown.
[0039] S3: The vibration signal x(t) is decomposed using the optimized TVF-EMD to obtain a set of multiple high-frequency signal components. eigenmode functions .
[0040] The optimized TVF-EMD decomposition of the vibration signal x(t) is the ambiguity minimization index. ,in Entropy represents the true signal and is used to measure the purity of a component.
[0041] S4: For high-frequency signal components Perform cubic B-spline fitting and smoothing to obtain the smoothed signal. .
[0042] B-spline functions employ cubic non-uniform node interpolation, and their basis functions are defined as follows: ; in, Represents second-order spline basis functions. Represents the node sequence.
[0043] S5: For smoothed signals Sparse sampling at fixed intervals Δt yields a signal sequence. .
[0044] Fixed interval Determined based on the sampling frequency fs and the target frequency band bandwidth, satisfying the Nyquist criterion: ; in, This represents the high-frequency limit of the target signal.
[0045] S6: Input the sub-signals into the Long Short-Term Memory (LSTM) network model for training and state prediction, and sum them to obtain the final prediction result; the sub-signals include intrinsic mode functions. With signal sequence .
[0046] The LSTM model consists of an input layer, two hidden state units, and an output layer. Its activation function is tanh, and its loss function is mean squared error. ; Sub-signals are divided into low-frequency trend signals. With high-frequency disturbance signals Both are used in training to preserve the full picture of state changes. The sub-signals are divided into training and test sets for training and state prediction, with a ratio of 7:3, and a rolling time window mechanism is used to improve generalization ability.
[0047] The prediction results of the training set and the test set on the LSTM model are as follows: Figure 6 and Figure 7 As shown. By Figure 6 and Figure 7 It can be seen that when the data without smooth sampling is input into the LSTM model, the LSTM model performs slightly better on the training set than on the data with smooth sampling. However, the performance on the test set is exactly the opposite. This indicates that the data without smooth sampling leads to overfitting of the LSTM model. This proves that after smooth sampling, the data can meet the needs of the LSTM model to learn the general laws of the data and can effectively avoid the occurrence of overfitting of the LSTM model.
[0048] Table 2 below shows the MMAE, MMAPE, and RRMSE of the LSTM model before and after smoothing sampling of high-frequency signal components on the training and test sets: Table 2
[0049] Compared with single prediction methods, before smoothing sampling, the prediction results of the training set are better than those of the test set. The MMAE of the test set is 44.2% lower, MMAPE is 59.7% lower, and RRMSE is 39.8% lower than that of the training set, indicating that the model has overfitted. After smoothing, the prediction indicators of the training set and the test set are similar, indicating that the LSTM model learns the data well.
[0050] Comparing different prediction methods, the test results of the test set after smoothing are better than those without smoothing. Specifically, MMAE is reduced by 34.3%, MMAPE by 34.3%, and RRMSE by 37.5%, proving that the proposed high-frequency signal component smoothing sampling can effectively improve the accuracy of predicting the state trend of hydropower units and enriches the method of predicting the state trend of hydropower units through oscillation data.
[0051] Furthermore, this hydropower unit state prediction method based on high-frequency component smoothing sampling is applicable to state trend prediction scenarios for various types of hydropower units, including through-flow, mixed-flow, axial-flow, and pumped-storage types.
[0052] Furthermore, the final prediction result can be determined by setting a threshold value. This enables early warning and alarm functions for operational risks. Specifically, if... If so, an alarm will be triggered.
[0053] Based on the same inventive concept, embodiments of the present invention also provide a hydropower unit state prediction system based on high-frequency component smooth sampling.
[0054] A hydropower unit state prediction system based on high-frequency component smooth sampling includes a processor and a memory. The memory stores a computer program, which is loaded and executed by the processor to implement the aforementioned hydropower unit state prediction method based on high-frequency component smooth sampling.
[0055] The various variations and specific examples of the hydropower unit state prediction method based on high-frequency component smoothing sampling provided in the above embodiments are also applicable to the hydropower unit state prediction system based on high-frequency component smoothing sampling in this embodiment. Through the foregoing detailed description of the hydropower unit state prediction method based on high-frequency component smoothing sampling, those skilled in the art can clearly understand the implementation method of the hydropower unit state prediction system based on high-frequency component smoothing sampling in this embodiment. For the sake of brevity, it will not be described in detail here.
[0056] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A method for predicting the state of a hydropower unit based on smooth sampling of high-frequency components, characterized in that, Includes the following steps: S1: Acquire the vibration signal of the target hydroelectric generator unit; S2: Optimize hyperparameters in TVF-EMD based on the fixed-value cyclic search method; S3: The vibration signal is decomposed using the optimized TVF-EMD to obtain a set of intrinsic mode functions containing multiple high-frequency signal components; S4: Perform cubic B-spline fitting smoothing on the high-frequency signal components to obtain a smoothed signal; S5: Perform sparse sampling on the smoothed signal at fixed intervals to obtain a signal sequence; S6: Input the sub-signals into the LSTM model for training and state prediction, and sum them to obtain the final prediction result; the sub-signals include intrinsic mode functions and signal sequences.
2. The hydropower unit state prediction method based on high-frequency component smoothing sampling as described in claim 1, characterized in that, Step S2 includes: Initialize the bandwidth parameter n and the center frequency ; Combine the bandwidth parameter n with the center frequency Perform the update; decompose and calculate the average discrete entropy; determine if the loop termination condition has been met; output the optimal hyperparameter bandwidth parameter n and center frequency. combination.
3. The hydropower unit state prediction method based on high-frequency component smoothing sampling as described in claim 1, characterized in that, The optimized TVF-EMD decomposition of the vibration signal is the ambiguity minimization index. ,in The entropy represents the true signal.
4. The hydropower unit state prediction method based on high-frequency component smoothing sampling as described in claim 1, characterized in that, Step S4 includes: B-spline functions employ cubic non-uniform node interpolation, and their basis functions are defined as follows: in, Represents second-order spline basis functions. Represents the node sequence.
5. The hydropower unit state prediction method based on high-frequency component smoothing sampling as described in claim 1, characterized in that, Fixed interval Determined based on the sampling frequency fs and the target frequency band bandwidth, satisfying the Nyquist criterion: in, This represents the high-frequency limit of the target signal.
6. The hydropower unit state prediction method based on high-frequency component smoothing sampling as described in claim 1, characterized in that, Step S6 includes: The LSTM model includes an input layer, two hidden state units, and an output layer. Its activation function is tanh, and its loss function is mean squared error. 。 7. The hydropower unit state prediction method based on high-frequency component smoothing sampling as described in claim 6, characterized in that, The sub-signals are divided into low-frequency trend signals and high-frequency disturbance signals, both of which participate in the training to preserve the full picture of state changes.
8. The hydropower unit state prediction method based on high-frequency component smoothing sampling as described in claim 7, characterized in that, The sub-signals are divided into a training set and a test set, with a ratio of 7:3, and a rolling time window mechanism is used to improve generalization ability.
9. The hydropower unit state prediction method based on high-frequency component smoothing sampling as described in claim 1, characterized in that, The final prediction result is obtained by setting a threshold value. To achieve early warning and alarm of operational risks; if If so, an alarm will be triggered.
10. A hydropower unit state prediction system based on high-frequency component smoothing sampling, characterized in that, The hydropower unit state prediction system based on high-frequency component smoothing sampling includes a processor and a memory, wherein the memory stores a computer program, and the computer program is loaded and executed by the processor to implement the hydropower unit state prediction method based on high-frequency component smoothing sampling as described in any one of claims 1-9.