Method and device for analyzing health performance influence parameters of hydroelectric generating set

Abstract

The invention belongs to the technical field of hydroelectric generation monitoring, and particularly discloses a hydroelectric generating set health performance influence parameter analysis method and device. The method comprises the following steps: acquiring a time sequence monitoring data set of a hydroelectric generating set under a healthy working condition; the time sequence monitoring data set comprises throw data at different moments and data of a plurality of unit performance parameters corresponding to the throw data at each moment; performing noise reduction processing on each piece of throw data in the time sequence monitoring data set and the corresponding data of the plurality of unit performance parameters to obtain a noise-reduced time sequence monitoring data set; recursive feature elimination analysis and cross validation are carried out according to the time sequence monitoring data set after noise reduction, and a target unit performance parameter combination used for determining the health state of the hydroelectric generating set is screened out from the multiple unit performance parameters. Through the method, the key influence parameters of the health state of the hydroelectric generating set can be accurately identified, and the health state evaluation effect of the hydroelectric generating set is improved.

Description

Technical Field

[0001] The present application belongs to the technical field of hydropower generation monitoring, and more specifically, relates to a method and device for analyzing parameters affecting the health performance of a hydropower unit. Background Art

[0002] Hydropower stations, with their unique and flexible peak-shaving and valley-filling capabilities, maintain the safe and stable operation of power systems and are becoming an indispensable component of these systems. At the same time, to address grid frequency fluctuations in the new power system, hydropower units play a vital role in frequency and phase regulation, as well as rotating standby. However, prolonged operation under unstable conditions can cause hydropower units to suffer structural fatigue, abnormal vibration, and other failures, impacting their health. The health of hydropower units directly impacts the stable operation and economic benefits of hydropower stations and the power grid. Therefore, monitoring the health of hydropower units is crucial for optimizing maintenance strategies, extending unit lifespan, and reducing failure rates.

[0003] In existing technologies, the use of swing signals has been the primary method for assessing the health status of hydropower units. However, due to the limited number of parameters influencing the health performance of hydropower units used in the assessment process and the lack of correlation analysis between these parameters and the unit's swing signals, existing methods for assessing the health status of hydropower units have significant limitations and poor assessment results. Therefore, effectively identifying the key parameters that affect the health status of hydropower units during the assessment process remains a pressing technical challenge within the industry. Summary of the Invention

[0004] In response to the defects of the existing technology, the purpose of this application is to effectively identify the important influencing parameters that affect the health status of hydropower units, aiming to solve the problem that the existing evaluation methods for the health status of hydropower units have major limitations and poor evaluation results.

[0005] To achieve the above objectives, in a first aspect, the present application provides a method for analyzing parameters affecting the health performance of a hydropower unit, comprising: Acquire a time series monitoring data set of a hydropower unit in a healthy operating condition; the time series monitoring data set includes swing data at different moments and data of multiple unit performance parameters corresponding to the swing data at each moment; performing noise reduction processing on each swing data and corresponding data of a plurality of unit performance parameters in the time series monitoring data set to obtain a noise-reduced time series monitoring data set; Recursive feature elimination analysis and cross-validation are performed on the de-noised time series monitoring data set to screen out a target unit performance parameter combination for determining the health status of the hydropower unit from the multiple unit performance parameters.

[0006] Optionally, performing recursive feature elimination analysis and cross-validation based on the noise-reduced time series monitoring data set to screen out a target unit performance parameter combination for determining the health status of the hydropower unit from the multiple unit performance parameters includes: The swing data at each moment in the noise-reduced time series monitoring data set and the corresponding data of multiple unit performance parameters are used as a set of training samples; Using each of the training samples to perform model training and cross-validation on a preset base model in the recursive feature elimination analysis, and extracting multiple optimal unit performance parameter combinations and the verification performance average value corresponding to each of the optimal unit performance parameter combinations from the model training and cross-validation process; The target unit performance parameter combination is obtained according to the optimal unit performance parameter combination with the highest verified performance average value among the multiple optimal unit performance parameter combinations.

[0007] Optionally, the preset base model is constructed based on a bidirectional long short-term memory network and an attention mechanism network, and is used to predict swing data.

[0008] Optionally, the method of using each of the training samples to perform model training and cross-validation on a preset base model in the recursive feature elimination analysis, and extracting multiple optimal unit performance parameter combinations and the average verification performance corresponding to each optimal unit performance parameter combination from the model training and cross-validation process, includes: Step S10: In the current training round, the determined current training sample is input into the preset base model for model training, and the weight coefficient corresponding to each unit performance parameter in the current training sample is extracted from the current training round; Step S20, eliminating the unit performance parameter with the lowest weight coefficient among the unit performance parameters in the current training sample, and determining the optimal unit performance parameter combination in the current training round; Step S30, performing K-fold cross validation based on the monitoring data corresponding to the optimal unit performance parameter combination to determine the average validation performance value corresponding to the optimal unit performance parameter combination; Step S40: obtaining training samples for the next training round of the current training round from each of the training samples, and determining the monitoring data corresponding to the optimal unit performance parameter combination in the training samples of the next training round as the next training samples required for model training; Taking the next training sample as the determined current training sample, repeating steps S10 to S40 to iteratively train the preset base model until the number of training rounds reaches a preset number, and then stopping the training; Extract the optimal unit performance parameter combination in each training round and its corresponding verification performance average.

[0009] Optionally, performing noise reduction processing on various types of signal data in the time series monitoring data set to obtain a noise-reduced time series monitoring data set includes: Perform multi-level wavelet decomposition on various signal data in the time series monitoring data set to obtain wavelet coefficients corresponding to each level of wavelet decomposition; the wavelet coefficients include approximate coefficients and detail coefficients; Modifying each detail coefficient according to a threshold value corresponding to each level of detail coefficient to obtain modified detail coefficients at each level; the threshold value is dynamically calculated based on the local energy distribution and entropy value characteristics of the signal; Wavelet reconstruction is performed using the approximate coefficients at each level and the modified detail coefficients at each level to obtain a denoised time series monitoring data set.

[0010] Optionally, the multiple unit performance parameters include at least guide vane opening, active power, reactive power, speed, excitation current, excitation voltage, water head, tailwater pipe inlet pressure, tailwater pipe outlet pressure, volute inlet pressure, tailwater cone outlet pressure, pressure between runner and guide vane, outer pressure between base ring and runner lower ring, inner force between base ring and runner lower ring, outer pressure of runner on top cover, maximum temperature of stator coil, maximum temperature of stator tooth pressure plate, maximum temperature of water guide shoe, maximum temperature of upper guide bearing oil side, maximum temperature of lower guide bearing oil inlet side, maximum temperature of thrust bearing oil inlet side, maximum temperature of stator core, oil temperature of water guide bearing oil tank, temperature of first water guide bearing shoe, temperature of second water guide bearing shoe, oil temperature of lower guide bearing thrust oil tank and oil temperature of upper guide bearing oil tank.

[0011] In a second aspect, the present application provides a device for analyzing parameters affecting the health performance of a hydropower unit, comprising: An acquisition module is used to acquire a time series monitoring data set of a hydropower unit in a healthy working condition; the time series monitoring data set includes swing data at different moments and data of multiple unit performance parameters corresponding to the swing data at each moment; A noise reduction module is used to perform noise reduction processing on various signal data in the time series monitoring data set to obtain a noise-reduced time series monitoring data set; An analysis module is used to perform recursive feature elimination analysis and cross-validation based on the noise-reduced time series monitoring data set, and screen out a target unit performance parameter combination for determining the health status of the hydropower unit from the multiple unit performance parameters.

[0012] In a third aspect, the present application provides an electronic device comprising: at least one memory for storing programs; and at least one processor for executing the programs stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method described in the first aspect or any possible implementation of the first aspect.

[0013] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.

[0014] In a fifth aspect, the present application provides a computer program product, which, when executed on a processor, enables the processor to execute the method described in the first aspect or any possible implementation of the first aspect.

[0015] It can be understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0016] In general, the above technical solutions conceived by this application have the following beneficial effects compared with the existing technologies: The present application provides a method and device for analyzing parameters affecting the health performance of a hydropower unit. By introducing a feature screening method that combines recursive feature elimination with cross-validation, a time-series monitoring data set of the hydropower unit under healthy operating conditions is used for noise reduction, which includes swings at different moments and corresponding multiple unit performance parameter data. The noise-reduced time-series monitoring data set is used for recursive feature elimination analysis, and the base model in the recursive feature elimination analysis is trained to extract the feature importance of various types of unit performance parameter data. Features are then dynamically eliminated based on the feature importance. Feature evaluation is then combined with cross-validation to reduce data overfitting and retain the feature combination with the best verification score. Target unit performance parameter combinations with a high correlation with the health status of the hydropower unit are thus screened out from multiple unit performance parameters. This can accurately identify key influencing parameters of the health status of the hydropower unit and improve the effect of the health status evaluation of the hydropower unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is one of the flow charts of the method for analyzing parameters affecting the health performance of a hydropower unit provided in an embodiment of the present application; Figure 2 This is the second flow chart of the method for analyzing parameters affecting the health performance of a hydropower unit provided in an embodiment of the present application; Figure 3 This is one of the result diagrams of the method for analyzing parameters affecting the health performance of a hydropower unit provided in an embodiment of the present application; Figure 4 This is the second schematic diagram of the results of the method for analyzing parameters affecting the health performance of a hydropower unit provided in an embodiment of the present application; Figure 5 Schematic diagram of the structure of the device for analyzing parameters affecting the health performance of a hydropower unit provided in an embodiment of the present application; Figure 6It is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0019] The terms "first" and "second" in the specification and claims of this application are used to distinguish different objects, rather than to describe a specific order of objects. For example, the terms "first water-guide bearing pad temperature" and "second water-guide bearing pad temperature" are used to distinguish the water-guide bearing pad temperatures at different locations, rather than to describe a specific order of water-guide bearing pad temperatures.

[0020] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0021] The embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application.

[0022] Figure 1 This is one of the flow charts of the method for analyzing parameters affecting the health performance of a hydropower unit provided in the embodiment of the present application, such as Figure 1 As shown, the method includes: Step S1, obtaining a time series monitoring data set of a hydropower unit in a healthy operating condition; the time series monitoring data set includes swing data at different moments and data of multiple unit performance parameters corresponding to the swing data at each moment; Step S2, performing noise reduction processing on each swing data and corresponding data of multiple unit performance parameters in the time series monitoring data set to obtain a noise-reduced time series monitoring data set; Step S3 , performing recursive feature elimination analysis and cross-validation based on the noise-reduced time series monitoring data set, and screening out a target unit performance parameter combination for determining the health status of the hydropower unit from multiple unit performance parameters.

[0023] Specifically, in the embodiments of the present application, due to the complex operating conditions and severe fluctuations of the hydropower unit, the signals collected on-site contain a large amount of noise, especially unexpected noise caused by human factors or sensor distortion. To this end, in step S1, the steady-state monitoring data of the unit under healthy operating conditions 10 minutes after startup and 5 minutes before shutdown can be extracted as the steady-state data of the hydropower unit, thereby obtaining a time-series monitoring data set of the hydropower unit under healthy operating conditions.

[0024] The swing data can be specifically the swing of the upper guide bearing and / or the lower guide bearing, for example, the swing of the upper guide bearing x The swing and the lower guide x Towards the swing, select the upper guide x Swing and downwind x The vibration values ​​of the two swings are used as indicators of healthy working conditions.

[0025] Among them, the performance parameters of the multiple units can specifically adopt the operating parameters of different parts of the hydropower unit, which can be selected according to the properties of the swing data.

[0026] Based on the contents of the above embodiments, as an optional embodiment, multiple unit performance parameters at least include guide vane opening, active power, reactive power, speed, excitation current, excitation voltage, water head, tailwater pipe inlet pressure, tailwater pipe outlet pressure, volute inlet pressure, tailwater cone outlet pressure, pressure between the runner and the guide vane, outer pressure between the base ring and the runner lower ring, inner force between the base ring and the runner lower ring, outer pressure of the runner on the top cover, maximum temperature of the stator coil, maximum temperature of the stator tooth pressure plate, maximum temperature of the water guide shoe, maximum temperature of the upper guide bearing oil side, maximum temperature of the lower guide bearing oil inlet side, maximum temperature of the thrust bearing oil inlet side, maximum temperature of the stator core, water guide bearing oil tank oil temperature, first water guide bearing shoe temperature, second water guide bearing shoe temperature, lower guide bearing thrust oil tank oil temperature and upper guide bearing oil tank oil temperature.

[0027] Specifically, in the embodiment of the present application, in step S1, the collection and the upper guide x Swing and downwind xThe unit performance parameters with different degrees of correlation with the swing include at least 27 related parameters, specifically the guide vane opening, active power, reactive power, speed, excitation current, excitation voltage, water head, tailwater pipe inlet pressure, tailwater pipe outlet pressure, volute inlet pressure, tailwater cone outlet pressure, pressure between the runner and the guide vane, outer pressure between the base ring and the lower ring of the runner, inner force between the base ring and the lower ring of the runner, outer pressure of the runner on the top cover, maximum temperature of the stator coil, maximum temperature of the stator tooth pressure plate, maximum temperature of the water guide bushing, maximum temperature of the oil side of the upper guide bearing, maximum temperature of the oil inlet side of the lower guide bearing bushing, maximum temperature of the oil inlet side of the thrust bearing bushing, maximum temperature of the stator core, oil temperature of the water guide bearing oil tank, temperature of the first water guide bearing bushing, temperature of the second water guide bearing bushing, oil temperature of the thrust oil tank of the lower guide bearing and oil temperature of the upper guide bearing oil tank.

[0028] The first water-guide bearing shoe temperature and the second water-guide bearing shoe temperature are shoe temperatures of different water-guide bearings respectively.

[0029] Optionally, in the embodiment of the present application, it is also possible to determine the swing data according to its attributes, such as the upper guide y In addition to the above 27 related parameters, the swing in the direction of y Identify and screen subsequent key parameters of other unit performance parameters related to swing.

[0030] The method of the embodiment of the present application can screen out more comprehensive and richer parameters affecting the health performance of the unit by selecting multi-source unit performance parameter data to identify key influencing parameters of the health status of the hydropower unit, thereby facilitating a more accurate assessment of the health status of the hydropower unit.

[0031] Furthermore, in an embodiment of the present application, in step S2, the unit stability data extracted in the previous step can be further subjected to noise reduction processing to remove noise components, thereby obtaining steady-state data that effectively suppresses noise and completely retains fault characteristic information, that is, obtaining a time series monitoring data set after noise reduction.

[0032] Based on the content of the above embodiment, as an optional embodiment, step S2, performing noise reduction processing on various types of signal data in the time series monitoring data set to obtain a noise-reduced time series monitoring data set, includes: Perform multi-level wavelet decomposition on various signal data in the time series monitoring data set to obtain the wavelet coefficients corresponding to each level of wavelet decomposition; the wavelet coefficients include approximate coefficients and detail coefficients; Modify each detail coefficient according to the threshold value corresponding to each level of detail coefficient to obtain the modified detail coefficient of each level; the threshold value is dynamically calculated based on the local energy distribution and entropy value characteristics of the signal; The approximate coefficients at each level and the modified detail coefficients at each level are used to perform wavelet reconstruction to obtain the denoised time series monitoring data set.

[0033] Specifically, in an embodiment of the present application, in step S2, an adaptive wavelet threshold noise reduction algorithm may be used to perform noise reduction processing on various types of signal data in the time series monitoring data set to obtain a noise-reduced time series monitoring data set.

[0034] More specifically, we first perform multi-level wavelet decomposition on various signal data in the time-series monitoring dataset, including swing data and corresponding unit performance parameter data, to obtain the wavelet coefficients corresponding to each level of wavelet decomposition. These wavelet coefficients can be selected based on the characteristics of the hydropower unit data to decompose the signal into approximate coefficients (representing the low-frequency portion of the signal) and detail coefficients (representing the high-frequency portion of the signal) at different scales.

[0035] Furthermore, the wavelet coefficients at each scale are processed using an adaptive dynamic thresholding algorithm, primarily modifying the detail coefficients at each scale. Using the Stein Unbiased Likelihood Estimation (SURE) algorithm, the threshold corresponding to the detail coefficients at each scale is dynamically calculated using the signal's local energy distribution and entropy characteristics. Based on the calculated adaptive threshold, either a hard thresholding or a soft thresholding method is used to modify the detail coefficients: when the absolute value of the detail coefficient is less than the adaptive threshold, it is set to zero; when it is greater than or equal to the adaptive threshold, it is shrunk using the soft thresholding function. After thresholding, the modified detail coefficients at each level are obtained.

[0036] Finally, the approximate coefficients at each level and the modified detail coefficients at each level can be used for wavelet reconstruction. The signal can be converted from the wavelet domain back to the time domain through inverse wavelet transform, and finally various types of signal denoising data that can effectively suppress noise and completely retain fault characteristic information can be obtained, that is, the denoised time series monitoring data set can be obtained.

[0037] The method of the embodiment of the present application adopts an adaptive threshold mechanism to perform noise reduction processing on various signal data in the time series monitoring data set, which can automatically distinguish between random noise and weak fault characteristics, avoid the problems of over-noise reduction or feature loss caused by traditional fixed threshold methods, and improve the reliability and accuracy of hydropower unit monitoring data.

[0038] Furthermore, in an embodiment of the present application, step S3 introduces a recursive feature elimination (RFE) analysis and cross-validation (CV) method. The RFE method is a wrapper-based feature selection method. Its core concept is to use a base model for multiple rounds of training and gradually eliminate a certain number of features based on the feature weight coefficients. Unlike traditional RFE, in an embodiment of the present application, by combining the RFE method and the CV method for feature selection, it is possible to accurately identify the key influencing parameters of the health status of the hydropower unit.

[0039] More specifically, in this step, RFE and CV analysis are performed based on the above-mentioned denoised time series monitoring data set, the denoised time series monitoring data is used to train the base model in RFE, the feature importance of each type of unit performance parameter data is extracted, and the features with lower importance are dynamically and gradually eliminated. Then, feature evaluation is performed in combination with CV to reduce data overfitting, reduce the accidental error of feature selection, and retain the feature combination with the best verification score. Finally, the target unit performance parameter combination for determining the health status of the hydropower unit can be gradually screened out from multiple unit performance parameters.

[0040] It can be understood that the target unit performance parameter combination is composed of some unit performance parameters that have a high correlation with the swing of the hydropower unit among multiple unit performance parameters. It is an optimal unit performance parameter combination and can be used to more accurately evaluate the health status of the hydropower unit.

[0041] The method for analyzing parameters affecting the health performance of hydropower units in an embodiment of the present application introduces a feature screening method that combines recursive feature elimination with cross-validation, and uses a time-series monitoring data set of hydropower units under healthy conditions for noise reduction, which includes swings at different moments and corresponding multiple unit performance parameter data. The noise-reduced time-series monitoring data set is used for recursive feature elimination analysis, and the base model in the recursive feature elimination analysis is trained to extract the feature importance of various types of unit performance parameter data, and then dynamically eliminate features according to the feature importance. After that, feature evaluation is performed in combination with cross-validation to reduce data overfitting and retain the feature combination with the best verification score, thereby screening out a target unit performance parameter combination with a high correlation with the health status of the hydropower unit from multiple unit performance parameters, which can accurately identify the key influencing parameters of the health status of the hydropower unit and improve the effect of the health status evaluation of the hydropower unit.

[0042] Based on the content of the above embodiment, as an optional embodiment, step S3, performing RFE analysis and CV verification based on the noise-reduced time series monitoring data set, and screening out a target unit performance parameter combination for determining the health status of the hydropower unit from multiple unit performance parameters, including: The swing data at each moment in the noise-reduced time series monitoring data set and the corresponding data of multiple unit performance parameters are used as a set of training samples; The preset base model in the RFE analysis is trained and verified using each training sample, and multiple optimal unit performance parameter combinations and the corresponding verification performance average values ​​of each optimal unit performance parameter combination are extracted from the model training and cross-validation process; The target unit performance parameter combination is obtained according to the optimal unit performance parameter combination with the highest average performance value among multiple optimal unit performance parameter combinations.

[0043] Specifically, in the embodiment of the present application, in step S3, after obtaining the noise-reduced time series monitoring data set, the swing data at each moment in the noise-reduced time series monitoring data set and the corresponding data of multiple unit performance parameters can be used as a set of training samples. x The data of the 27 unit performance parameters and the corresponding swing are used as a set of training samples. x The data of axial swing and its corresponding 27 unit performance parameters are used as a set of training samples.

[0044] Furthermore, in an embodiment of the present application, each training sample is used to perform model training and CV verification on the preset base model in the RFE analysis, and the weight coefficient of each unit performance parameter is extracted from the model training and cross-validation process to determine the degree of correlation between each unit performance parameter and the swing data. By gradually eliminating a certain number of parameters with low correlation and low importance, the optimal unit performance parameter combination and its corresponding verification performance average value in each model training process are determined, thereby obtaining multiple optimal unit performance parameter combinations and the verification performance average value corresponding to each optimal unit performance parameter combination.

[0045] Based on the content of the above embodiment, as an optional embodiment, the preset base model is constructed based on a bidirectional long short-term memory (BiLSTM) network and an attention mechanism (Attention) network, and is used to predict swing data.

[0046] Specifically, the BiLSTM network can capture the advantages of the forward and backward bidirectional temporal dependencies of monitoring data, and the Attention network can assign weights to hidden states at different time steps, highlight key time points, and generate context vectors through weighted summation to reflect the comprehensive contribution of each feature to the health status.

[0047] The method of the embodiment of the present application, by selecting the BiLSTM-Attention network model as the preset base model, can effectively solve the timing dependency problem that the traditional RFE method based on the support vector machine (SVM) as the base model cannot handle. It can not only model the timing relationship, but also quantify the importance of features through attention weights, providing a dynamic elimination basis for RFE, which is conducive to more accurate identification of key influencing parameters of the health status of hydropower units.

[0048] Based on the content of the above embodiment, as an optional embodiment, each training sample is used to perform model training and CV verification on the preset base model in the RFE analysis, and multiple optimal unit performance parameter combinations and the verification performance average corresponding to each optimal unit performance parameter combination are extracted from the model training and CV verification process, including: Step S10: In the current training round, the determined current training sample is input into the preset base model for model training, and the weight coefficient corresponding to each unit performance parameter in the current training sample is extracted from the current training round; Step S20, eliminating the unit performance parameter with the lowest weight coefficient among the unit performance parameters in the current training sample, and determining the optimal unit performance parameter combination in the current training round; Step S30, performing K-fold cross validation based on the monitoring data corresponding to the optimal unit performance parameter combination to determine the average verification performance value corresponding to the optimal unit performance parameter combination; Step S40: obtaining training samples for the next training round of the current training round from each training sample, and determining the monitoring data corresponding to the optimal unit performance parameter combination in the training samples of the next training round as the next training sample required for model training; The next training sample is used as the determined current training sample, and steps S10 to S40 are repeated to iteratively train the preset base model until the number of training rounds reaches the preset number, and the training is stopped; Extract the optimal unit performance parameter combination in each training round and its corresponding verification performance average.

[0049] Specifically, in an embodiment of the present application, the specific steps of using each training sample to perform model training and CV verification on the preset base model in the RFE analysis, and extracting multiple optimal unit performance parameter combinations and the verification performance average value corresponding to each optimal unit performance parameter combination from the model training and CV verification process include: According to step S10, in the current training round, the determined current training samples can be input into the preset base model for model training, and the weight coefficients corresponding to the performance parameters of each unit in the current training samples can be extracted from the current training round.

[0050] More specifically, when the current training round is the first, the first training sample among all training samples input into the model is used as the current training sample and fed into a preset base model, such as a BiLSTM-Attention network model, for model training. During the model's prediction of swing data, the Attention network outputs weight coefficients corresponding to all unit performance parameters. These weight coefficients reflect the contribution of different unit performance parameters to the predicted swing data.

[0051] According to step S20, the unit performance parameter with the lowest weight coefficient among the unit performance parameters in the first training sample is eliminated, and the remaining unit performance parameters are retained, so that the optimal unit performance parameter combination in the first training round can be determined.

[0052] According to step S30, a K-fold cross-validation is performed on the preset base model based on the monitoring data corresponding to the optimal unit performance parameter combination in the first training round. The performance values ​​recorded during the K-fold validation process are averaged to determine the average verification performance value corresponding to the optimal unit performance parameter combination.

[0053] In step S40, a training sample for the second training round is obtained from each training sample, and the monitoring data in this training sample corresponding to the optimal unit performance parameter combination in the first training round is determined as the next training sample required for model training. In other words, this next training sample is used to perform a second training on the preset base model.

[0054] If it is determined that the current training round has not reached the preset number of times, the preset base model is trained again according to step S10, using the next training sample as the determined current training sample. The weight coefficients corresponding to the various unit performance parameters in the current training sample are further extracted from the second training round, and steps S20 to S40 are repeated to complete the second training round of the preset base model. At this point, the optimal unit performance parameter combination for the second training round is determined by eliminating the parameter with the lowest weight coefficient from the optimal unit performance parameter combination in the first training round.

[0055] Similarly, through iterative training of the preset base model, the parameters with the lowest weight coefficients in the optimal unit performance parameter combination in each training round are gradually eliminated, and K-fold CV verification is performed until the preset number of training rounds is reached, at which point training is terminated. At this point, the optimal unit performance parameter combination in each training round and its corresponding verification performance average can be extracted.

[0056] The method of the embodiment of the present application iteratively trains the preset base model in the RFE, gradually eliminates the parameters with the lowest weight coefficient in the optimal unit performance parameter combination in each training round, and combines K-fold CV verification to ensure the generalization of the optimal unit performance parameter combination and retain the feature combination with the best verification score.

[0057] Furthermore, in an embodiment of the present application, the optimal unit performance parameter combination with the highest verified performance average value among multiple optimal unit performance parameter combinations can be directly used as the target unit performance parameter combination. Then, the target unit performance parameter combination can be used to analyze the weight coefficient of each parameter for the health status of the hydropower unit, and the importance of each parameter to the status assessment of the hydropower unit can be obtained. The higher the weight coefficient, the more important it is. On the contrary, the lower the weight coefficient, the lower the correlation and the less important it is, thereby preparing for the subsequent unit status assessment work.

[0058] The method of the embodiment of the present application uses the swing data of each moment in the denoised time series monitoring data set and the data of multiple unit performance parameters corresponding thereto as a group of training samples, adopts BiLSTM-Attention as the base model, and uses each training sample to perform model training and CV verification on the preset base model in the RFE analysis. In combination with RFE processing and CV verification, feature selection is performed on the performance parameters of each unit, and features with lower importance are gradually eliminated to obtain the optimal feature subset. The key parameters that have a greater impact on the health status of the hydropower unit are selected, thereby providing a more reliable data basis for subsequent unit health status assessment.

[0059] Figure 2 This is the second flow chart of the method for analyzing parameters affecting the health performance of a hydropower unit provided in the embodiment of the present application, such as Figure 2 As shown, in the embodiment of the present application, step 1 (Step 1) extracts steady-state data of the hydropower unit. Due to the complex operating conditions and severe fluctuations of the hydropower unit, the signals collected on-site contain a large amount of noise, especially unexpected noise caused by human factors or sensor distortion. Therefore, in the embodiment of the present application, steady-state monitoring data for 10 minutes after the unit is started and 5 minutes before shutdown is extracted as the steady-state data of the hydropower unit. In other words, a time-series monitoring data set of the hydropower unit in a healthy operating condition is obtained, including swing data at different times and data on multiple unit performance parameters corresponding to the swing data at each time.

[0060] More specifically, we can obtain the upper guide x Swing and downwind x The vibration values ​​of the two swings are used as health indicators. Meanwhile, 27 parameters, including guide vane opening, active power, reactive power, speed, excitation current, excitation voltage, head, and draft tube inlet pressure, are selected as unit performance parameters to be analyzed. This concludes Step 1.

[0061] In step 2, noise reduction is performed. In this embodiment, adaptive wavelet threshold noise reduction is performed on the unit steady-state data extracted in the previous step. Based on the characteristics of the hydropower unit data, the various signal data in the unit steady-state data can be decomposed into approximate coefficients (representing the low-frequency portion of the signal) and detail coefficients (representing the high-frequency portion of the signal) at different scales. An adaptive threshold is calculated for the detail coefficients at each scale. Based on the calculated adaptive threshold, either a hard threshold or a soft threshold method is selected to process the detail coefficients. Finally, an inverse wavelet transform is performed on the thresholded approximate coefficients and detail coefficients to reconstruct the de-noised time series monitoring dataset. This concludes step 2.

[0062] Step 3 is divided into three steps: Step 3.1, Step 3.2, and Step 3.3. Following the aforementioned processing method, in Step 3.1, the denoised time series monitoring dataset is input into the base model constructed using the BiLSTM-Attention network model using the RFE method for model training. In Step 3.2, the RFE method is used to assess feature importance and remove features with lower weights. Specifically, during model training, the weight coefficients of each unit performance parameter for swing data prediction are extracted to determine the feature importance of each unit performance parameter. In each round of model training, unit performance parameters with lower weight coefficients are removed to determine the optimal unit performance parameter combination for each round of model training. In Step 3.3, feature selection is performed using K-fold CV validation. The average validation performance of the optimal unit performance parameter combination in each round of model training is calculated to more accurately identify the key influencing parameters of the hydropower unit health status. This concludes Step 3.

[0063] In step 4, after the model training is completed, the optimal unit performance parameter combination with the highest average verification performance is selected as the target unit performance parameter combination, and the weight coefficients of each unit performance parameter extracted in each round of model training can be saved and visualized through a bar chart for users to view. Figure 3 and Figure 4 As shown in the figure, the influencing parameters with higher correlation are selected according to the weight coefficient of each unit performance parameter to prepare for the subsequent health status assessment of the hydropower unit.

[0064] in, Figure 3 For the upper guide x Schematic diagram of the results of key influencing parameter analysis of 27 unit performance parameters corresponding to the swing. Figure 4 For the lower guide x Schematic diagram of the results of key influencing parameter analysis of 27 unit performance parameters corresponding to swing.

[0065] The following describes the device for analyzing parameters affecting the health performance of a hydropower unit provided in this application. The device for analyzing parameters affecting the health performance of a hydropower unit described below and the method for analyzing parameters affecting the health performance of a hydropower unit described above can be referenced to each other.

[0066] Figure 5 Schematic diagram of the structure of the device for analyzing parameters affecting the health performance of a hydropower unit provided in an embodiment of the present application. Figure 5 As shown, the device includes: An acquisition module 10 is used to acquire a time series monitoring data set of a hydropower unit in a healthy operating condition; the time series monitoring data set includes swing data at different moments and data of multiple unit performance parameters corresponding to the swing data at each moment; The noise reduction module 20 is used to perform noise reduction processing on various signal data in the time series monitoring data set to obtain a noise-reduced time series monitoring data set; The analysis module 30 is used to perform recursive feature elimination analysis and cross-validation based on the noise-reduced time series monitoring data set, and screen out a target unit performance parameter combination for determining the health status of the hydropower unit from multiple unit performance parameters.

[0067] It is understandable that the detailed functional implementation of each of the above units / modules can be found in the introduction of the aforementioned method embodiment, and will not be repeated here.

[0068] It should be understood that the above-mentioned device is used to execute the method in the above-mentioned embodiment. The implementation principle and technical effect of the corresponding program module in the device are similar to those described in the above-mentioned method. The working process of the device can refer to the corresponding process in the above-mentioned method and will not be repeated here.

[0069] The device for analyzing parameters affecting the health performance of a hydropower unit in an embodiment of the present application introduces a feature screening method that combines recursive feature elimination with cross-validation, and uses a time-series monitoring data set of a hydropower unit under healthy conditions for noise reduction, including swings at different moments and corresponding multiple unit performance parameter data. The noise-reduced time-series monitoring data set is used for recursive feature elimination analysis, and the base model in the recursive feature elimination analysis is trained to extract the feature importance of each type of unit performance parameter data, and then dynamically eliminate features according to the feature importance. Feature evaluation is then performed in combination with cross-validation to reduce data overfitting and retain the feature combination with the best verification score, thereby screening out a target unit performance parameter combination with a high correlation with the health status of the hydropower unit from multiple unit performance parameters. This can accurately identify key influencing parameters of the health status of the hydropower unit and improve the effect of the health status evaluation of the hydropower unit.

[0070] Based on the method in the above embodiment, the embodiment of the present application provides an electronic device, such as Figure 6 As shown, the electronic device may include: a processor (Processor) 610, a communication interface (Communications Interface) 620, a memory (Memory) 630 and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other via the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the method in the above embodiment.

[0071] In addition, the logic instructions in the aforementioned memory 630 can be implemented in the form of a software functional unit and, when sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application.

[0072] Based on the method in the above embodiment, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method in the above embodiment.

[0073] Based on the method in the above embodiment, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method in the above embodiment.

[0074] It is understood that the processor in the embodiments of the present application may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0075] The method steps in the embodiments of the present application can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and storage medium can be located in an ASIC.

[0076] The above embodiments can be implemented in whole or in part using software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions. When loaded and executed on a computer, the computer program instructions fully or partially produce the processes or functions described in the embodiments of this application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted via the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disk, hard disk, tape), optical media (e.g., DVD), or semiconductor media (e.g., solid-state drive (SSD)).

[0077] It will be understood that the various numerical numbers involved in the embodiments of the present application are merely distinctions for the convenience of description and are not intended to limit the scope of the embodiments of the present application.

[0078] It should be understood that expressions such as "include" and "may include" used in this application indicate the existence of the disclosed functions, operations, or constituent elements, and do not limit one or more additional functions, operations, and constituent elements. In this application, terms such as "include" and / or "have" may be interpreted as indicating specific characteristics, numbers, operations, constituent elements, components, or combinations thereof, but may not be interpreted as excluding the existence or possibility of adding one or more other characteristics, numbers, operations, constituent elements, components, or combinations thereof.

[0079] In addition, in this application, the expression "and / or" includes any and all combinations of the associated listed words. For example, the expression "A and / or B" may include A, may include B, or may include both A and B.

[0080] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for analyzing parameters affecting the health performance of a hydropower unit, characterized in that: include: Acquire a time series monitoring data set of a hydropower unit in a healthy operating condition; the time series monitoring data set includes swing data at different moments and data of multiple unit performance parameters corresponding to the swing data at each moment; performing noise reduction processing on each swing data and corresponding data of a plurality of unit performance parameters in the time series monitoring data set to obtain a noise-reduced time series monitoring data set; Recursive feature elimination analysis and cross-validation are performed on the de-noised time series monitoring data set to screen out a target unit performance parameter combination for determining the health status of the hydropower unit from the multiple unit performance parameters.

2. The method for analyzing parameters affecting the health performance of a hydropower unit according to claim 1, characterized in that: The recursive feature elimination analysis and cross-validation are performed based on the noise-reduced time series monitoring data set to select a target unit performance parameter combination for determining the health status of the hydropower unit from the multiple unit performance parameters, including: The swing data at each moment in the noise-reduced time series monitoring data set and the corresponding data of multiple unit performance parameters are used as a set of training samples; Using each of the training samples to perform model training and cross-validation on a preset base model in the recursive feature elimination analysis, and extracting multiple optimal unit performance parameter combinations and the verification performance average value corresponding to each of the optimal unit performance parameter combinations from the model training and cross-validation process; The target unit performance parameter combination is obtained according to the optimal unit performance parameter combination with the highest verified performance average value among the multiple optimal unit performance parameter combinations.

3. The method for analyzing parameters affecting the health performance of a hydropower unit according to claim 2, characterized in that: The preset base model is constructed based on a bidirectional long short-term memory network and an attention mechanism network, and is used to predict swing data.

4. The method for analyzing parameters affecting the health performance of a hydropower unit according to claim 2, characterized in that: The method of using each of the training samples to perform model training and cross-validation on a preset base model in the recursive feature elimination analysis, and extracting multiple optimal unit performance parameter combinations and the verification performance average value corresponding to each optimal unit performance parameter combination from the model training and cross-validation process, includes: Step S10: In the current training round, the determined current training sample is input into the preset base model for model training, and the weight coefficient corresponding to each unit performance parameter in the current training sample is extracted from the current training round; Step S20, eliminating the unit performance parameter with the lowest weight coefficient among the unit performance parameters in the current training sample, and determining the optimal unit performance parameter combination in the current training round; Step S30, performing K-fold cross validation based on the monitoring data corresponding to the optimal unit performance parameter combination to determine the average validation performance value corresponding to the optimal unit performance parameter combination; Step S40: obtaining training samples for the next training round of the current training round from each of the training samples, and determining the monitoring data corresponding to the optimal unit performance parameter combination in the training samples of the next training round as the next training samples required for model training; Taking the next training sample as the determined current training sample, repeating steps S10 to S40 to iteratively train the preset base model until the number of training rounds reaches a preset number, and then stopping the training; Extract the optimal unit performance parameter combination in each training round and its corresponding verification performance average.

5. The method for analyzing parameters affecting the health performance of a hydropower unit according to any one of claims 1 to 4, characterized in that: The performing noise reduction processing on various signal data in the time series monitoring data set to obtain a noise-reduced time series monitoring data set includes: Perform multi-level wavelet decomposition on various signal data in the time series monitoring data set to obtain wavelet coefficients corresponding to each level of wavelet decomposition; the wavelet coefficients include approximate coefficients and detail coefficients; Modifying each detail coefficient according to a threshold value corresponding to each level of detail coefficient to obtain modified detail coefficients at each level; the threshold value is dynamically calculated based on the local energy distribution and entropy value characteristics of the signal; Wavelet reconstruction is performed using the approximate coefficients at each level and the modified detail coefficients at each level to obtain a denoised time series monitoring data set.

6. The method for analyzing parameters affecting the health performance of a hydropower unit according to any one of claims 1 to 4, characterized in that: The multiple unit performance parameters at least include guide vane opening, active power, reactive power, speed, excitation current, excitation voltage, water head, draft tube inlet pressure, draft tube outlet pressure, volute inlet pressure, draft cone outlet pressure, pressure between runner and guide vane, outer pressure between base ring and runner lower ring, inner force between base ring and runner lower ring, outer pressure of runner on top cover, maximum temperature of stator coil, maximum temperature of stator tooth pressure plate, maximum temperature of water guide shoe, maximum temperature of upper guide bearing oil side, maximum temperature of lower guide bearing oil inlet side, maximum temperature of thrust bearing oil inlet side, maximum temperature of stator core, oil temperature of water guide bearing oil tank, temperature of first water guide bearing shoe, temperature of second water guide bearing shoe, oil temperature of lower guide bearing thrust oil tank and oil temperature of upper guide bearing oil tank.

7. A device for analyzing parameters affecting the health performance of a hydropower unit, characterized in that: include: An acquisition module is used to acquire a time series monitoring data set of a hydropower unit in a healthy working condition; the time series monitoring data set includes swing data at different moments and data of multiple unit performance parameters corresponding to the swing data at each moment; A noise reduction module is used to perform noise reduction processing on various signal data in the time series monitoring data set to obtain a noise-reduced time series monitoring data set; An analysis module is used to perform recursive feature elimination analysis and cross-validation based on the noise-reduced time series monitoring data set, and screen out a target unit performance parameter combination for determining the health status of the hydropower unit from the multiple unit performance parameters.

8. An electronic device, characterized in that: include: at least one memory for storing a computer program; At least one processor is used to execute the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed on a processor, the processor is caused to execute the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that When the computer program product is run on a processor, the processor is enabled to perform the method according to any one of claims 1 to 6.

Images