Abnormal vibration detection method and device for pumped storage unit

Abstract

The invention provides a pumped storage unit vibration anomaly detection method and device, and the method comprises the steps: collecting steady-state data during the healthy operation of a pumped storage unit, and determining a unit vibration threshold value through a Gaussian cloud model and the steady-state data; collecting a throw vibration value of the pumping and storage unit running at the current moment; converting the vibration threshold into a corresponding threshold cloud, and converting the throw vibration value into a corresponding observation cloud; and determining the fault risk of the pumped storage unit at the current moment through the Warsestein distance between the threshold cloud and the observation cloud. According to the method, the process of building a complex health model is not needed, the time required for model training is saved, and the uncertainty such as fuzziness and instantaneous volatility in the working state of the pumped storage unit is fully described by using the normal cloud model.

Description

Technical Field

[0001] One or more embodiments of this specification relate to the technical field of water conservancy project hydropower generation, and particularly to a method and device for detecting abnormal vibration of a pumped storage unit. Background Art

[0002] Pumped storage units play an important role in regulating energy in a high-proportion new energy power system, and as the core equipment for energy efficiency conversion in pumped storage power stations, they undertake the important task of maintaining the stable operation of the power grid. The operating conditions of pumped storage units are complex, and they need to frequently convert between pumping and generating conditions. Therefore, they need to be frequently switched on and off, reversed, and there are differences in the vibration characteristics of the units under various conditions. The vibration problem of pumped storage units has always been the focus of attention in the hydropower industry. Abnormal vibration data not only affects the safe and stable operation of pumped storage units but may also lead to serious equipment damage and even major accidents that endanger personal safety.

[0003] In the prior art, vibration monitoring, acquisition, and evaluation of pumped storage units are key technologies to ensure their stable operation. In current state evaluation methods for pumped storage units, the method based on a health benchmark model is the mainstream. This method aims to explore the high-dimensional coupling relationship between the vibration of the unit and the operating condition parameters. However, establishing a health model requires training a very complex model, and there are deficiencies in real-time performance. Moreover, a model with point estimation as the evaluation standard cannot fully reflect the uncertainties such as randomness, fuzziness, and instantaneous volatility in the operating state of pumped storage units.

[0004] Application Content

[0005] This application describes a method and device for detecting abnormal vibration of a pumped storage unit, which can solve the above technical problems.

[0006] According to the first aspect, a method for detecting abnormal vibration of a pumped storage unit is provided. The method includes:

[0007] Collect the steady-state data when the pumped storage unit operates healthily, and use the Gaussian cloud model and the steady-state data to determine the vibration thresholds of the unit. The vibration thresholds include the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold;

[0008] Collect the swing vibration values of the pumped storage unit at the current moment of operation. The swing vibration values include the upper guide x-direction swing vibration value, the upper guide y-direction swing vibration value, the lower guide x-direction swing vibration value, and the lower guide y-direction swing vibration value;

[0009] Convert the vibration thresholds into corresponding threshold clouds, and convert the swing vibration values into corresponding observation clouds;

[0010] Determine the fault risk of the pumped-storage unit at the current moment through the Wasserstein distance between the threshold cloud and the observed cloud.

[0011] Based on the above further embodiments, collect the steady-state data of the pumped-storage unit during healthy operation, and use the Gaussian cloud model and the steady-state data to determine the vibration threshold of the unit, which specifically includes:

[0012] Collect the steady-state data of the pumped-storage unit during healthy operation, and perform denoising processing on the steady-state data, where the steady-state data includes the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value;

[0013] Use the steady-state data and the Gaussian cloud model to determine the vibration threshold of the unit, and the vibration threshold includes the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold.

[0014] Based on the above further embodiments, collecting the steady-state data of the pumped-storage unit during healthy operation and performing denoising processing on the steady-state data includes:

[0015] Collect the steady-state data of the pumped-storage unit during a preset period of healthy operation;

[0016] Perform wavelet decomposition on the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value in the steady-state data respectively;

[0017] Judge whether the absolute value of each obtained wavelet coefficient is less than a preset threshold;

[0018] If so, set the wavelet coefficient to zero;

[0019] Otherwise, retain the original value of the wavelet coefficient;

[0020] Use the modified wavelet coefficients to perform signal reconstruction on the steady-state data, that is, convert back from the wavelet domain to the time domain, and the reconstructed steady-state data is the denoised steady vibration signal.

[0021] Based on the above further embodiments, using the steady-state data and the Gaussian cloud model to determine the vibration threshold of the unit includes:

[0022] Set a sliding window for the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value in the steady-state data respectively, and each sliding window contains continuous data points;

[0023] For the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value, a Gaussian distribution is determined respectively, and four Gaussian distributions are obtained for modeling, and the parameters of the Gaussian cloud model GMM are initialized;

[0024] Traverse each sliding window, use GMM to fit the data in the window, and optimize the parameters of the GMM model through the expectation maximization algorithm;

[0025] For each sliding window, according to the four Gaussian distributions of the GMM, find the Gaussian distribution with the largest weight, and determine the unit vibration threshold according to the 3σ principle.

[0026] Based on the above further embodiments, the conversion of the vibration threshold to the corresponding threshold cloud includes:

[0027] According to the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold, calculate the mean value of each vibration threshold respectively where x i is the i-th vibration threshold, the first-order absolute central moment The second-order central moment

[0028] According to the distribution characteristics of the normal cloud, when the first-order absolute central moment of the first normal cloud, that is, the expected value then the entropy

[0029] Calculate the hyperentropy according to the characteristics of the normal cloud distribution variance Obtain the expected value Ex1, entropy En1, and hyperentropy He1 of each swing vibration threshold;

[0030] For the vibration threshold in each direction, define a normal threshold cloud model using the expected value Ex1, entropy En1, and hyperentropy He1;

[0031] Through the inverse cloud transformation, convert the normal threshold cloud model into the threshold cloud.

[0032] Based on the above further embodiments, the conversion of the swing vibration value to the corresponding observed cloud includes:

[0033] According to the upper guide x-direction swing vibration value, the upper guide y-direction swing vibration value, the lower guide x-direction swing vibration value, and the lower guide y-direction swing vibration value, calculate the mean value of the vibration thresholds in each direction respectively where y iis the i-th swing vibration value, n is the number of swing vibration values in each direction, and the first-order absolute central moment Second-order central moment

[0034] According to the distribution characteristics of the normal cloud, when At this time, the first-order absolute central moment of the normal cloud is the expected value Then the entropy

[0035] Calculate the hyperentropy according to the characteristics of the variance of the normal cloud distribution Obtain the expected value Ey2, entropy En2, and hyperentropy He2 of the vibration thresholds in each direction;

[0036] For the swing vibration values in each direction, define a normal swing vibration cloud model using the expected value Ex, entropy En2, and hyperentropy He2;

[0037] Through the inverse cloud transformation, convert the normal swing vibration cloud model into the observed cloud.

[0038] Based on the above further embodiments, determining the fault risk of the pumped storage unit at the current moment through the Wasserstein distance between the threshold cloud and the observed cloud includes:

[0039] Based on the concept similarity comparison method LCIM of the cloud model, compare the similarity between the threshold cloud and the observed cloud;

[0040] Let the threshold cloud C1 = (Ex1, En1, He1), and the observed cloud C2 = (Ey2, En2, He2);

[0041] Use the similarity formula as follows:

[0042]

[0043] Obtain the similarity between the threshold cloud and the observed cloud, and determine the fault risk of the pumped storage unit at the current moment according to the similarity.

[0044] According to the second aspect, a method for detecting abnormal vibration of a pumped storage unit is provided, and the method includes:

[0045] Collect the steady-state data of the pumped storage unit during healthy operation, and use the Gaussian cloud model and the steady-state data to determine the vibration threshold of the unit. The vibration threshold includes the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold;

[0046] Collect the swing and vibration values of the pumped storage unit at the current moment of operation, where the swing and vibration values include the upper guide x-direction swing and vibration value, the upper guide y-direction swing and vibration value, the lower guide x-direction swing and vibration value, and the lower guide y-direction swing and vibration value;

[0047] Convert the vibration threshold into a corresponding threshold cloud, and convert the swing and vibration values into corresponding observation clouds;

[0048] Determine the fault risk of the pumped storage unit at the current moment through the Wasserstein distance between the threshold cloud and the observation cloud.

[0049] Based on the above further embodiments, the first processing module is specifically configured to collect the steady-state data of the pumped storage unit during healthy operation, and perform denoising processing on the steady-state data, where the steady-state data includes the first upper guide x-direction swing and vibration value, the first upper guide y-direction swing and vibration value, the first lower guide x-direction swing and vibration value, and the first lower guide y-direction swing and vibration value;

[0050] Use the steady-state data and the Gaussian cloud model to determine the vibration threshold of the unit, where the vibration threshold includes the upper guide x-direction swing and vibration threshold, the upper guide y-direction swing and vibration threshold, the lower guide x-direction swing and vibration threshold, and the lower guide y-direction swing and vibration threshold.

[0051] Based on the above further embodiments, the first processing module is specifically configured to collect the steady-state data of the pumped storage unit during a preset period of healthy operation;

[0052] Perform wavelet decomposition on the first upper guide x-direction swing and vibration value, the first upper guide y-direction swing and vibration value, the first lower guide x-direction swing and vibration value, and the first lower guide y-direction swing and vibration value in the steady-state data respectively;

[0053] Judge whether the absolute value of each obtained wavelet coefficient is less than a preset threshold;

[0054] If so, set the wavelet coefficient to zero;

[0055] Otherwise, retain the original value of the wavelet coefficient;

[0056] Use the modified wavelet coefficients to perform signal reconstruction on the steady-state data, that is, convert back from the wavelet domain to the time domain, and the reconstructed steady-state data is the denoised steady vibration signal.

[0057] Based on the above further embodiments, the second processing module is specifically configured to set a sliding window for the first upper guide x-direction swing and vibration value, the first upper guide y-direction swing and vibration value, the first lower guide x-direction swing and vibration value, and the first lower guide y-direction swing and vibration value in the steady-state data respectively, and each sliding window contains continuous data points;

[0058] Determine a Gaussian distribution for the first upper guide x-directional swing vibration value, the first upper guide y-directional swing vibration value, the first lower guide x-directional swing vibration value, and the first lower guide y-directional swing vibration value respectively, obtain four Gaussian distributions for modeling, and initialize the parameters of the Gaussian cloud model GMM;

[0059] Traverse each sliding window, use GMM to fit the data within the window, and optimize the parameters of the GMM model through the expectation-maximization algorithm;

[0060] For each sliding window, find the Gaussian distribution with the largest weight according to the four Gaussian distributions of the GMM, and determine the unit vibration threshold according to the 3σ principle.

[0061] Based on the above further embodiments, the third processing module is specifically configured to calculate the mean value of each vibration threshold according to the upper guide x-directional swing vibration threshold, the upper guide y-directional swing vibration threshold, the lower guide x-directional swing vibration threshold, and the lower guide y-directional swing vibration threshold where x i is the i-th vibration threshold, the first-order absolute central moment the second-order central moment

[0062] According to the distribution characteristics of the normal cloud, when the first-order absolute central moment of the first normal cloud, that is, the expected value then the entropy

[0063] Calculate the hyperentropy according to the characteristics of the variance of the normal cloud distribution Obtain the expected value Ex1, entropy En1, and hyperentropy He1 of each swing vibration threshold;

[0064] For the vibration threshold in each direction, define the normal threshold cloud model using the expected value Ex1, entropy En1, and hyperentropy He1;

[0065] Through inverse cloud transformation, convert the normal threshold cloud model into the threshold cloud.

[0066] Based on the above further embodiments, the third processing module is specifically configured to calculate the mean value of the vibration thresholds in each direction according to the upper guide x-directional swing vibration value, the upper guide y-directional swing vibration value, the lower guide x-directional swing vibration value, and the lower guide y-directional swing vibration value where y i is the i-th swing vibration value, n is the number of swing vibration values in each direction, the first-order absolute central moment the second-order central moment

[0067] According to the distribution characteristics of the normal cloud, when When it is, the first-order absolute central moment of the normal cloud is the expected value Then the entropy

[0068] Calculate the hyper entropy according to the characteristics of the variance of the normal cloud distribution Obtain the expected value Ey2, entropy En2 and hyper entropy He2 of the vibration thresholds in each direction;

[0069] For the swing vibration value in each direction, define the normal swing vibration cloud model using the expected value Ex, entropy En2 and hyper entropy He2;

[0070] Through the inverse cloud transformation, convert the normal swing vibration cloud model into the observed cloud.

[0071] Based on the above further embodiments, the fourth processing module is specifically configured to compare the similarity between the threshold cloud and the observed cloud based on the concept similarity comparison method LCIM of the cloud model;

[0072] Let the threshold cloud C1=(Ex1, En1, He1), and the observed cloud C2=(Ey2, En2, He2);

[0073] Use the similarity formula as follows:

[0074]

[0075] Obtain the similarity between the threshold cloud and the observed cloud, and determine the fault risk of the pumped storage unit at the current moment according to the similarity.

[0076] According to a third aspect, there is provided a computer storage medium, on which a computer program is stored, and when the computer program is executed by one or more processors, it implements the vibration anomaly detection method for a pumped storage unit as described in any one of the above technical solutions.

[0077] According to a fourth aspect, there is provided an electronic device, including a memory and one or more processors, on which a computer program is stored, and when the computer program is executed by the one or more processors, it implements the vibration anomaly detection method for a pumped storage unit as described in any one of the above technical solutions.

[0078] In the above systems and methods provided in the embodiments of this specification, it effectively overcomes the process of establishing a complex health model required by the existing evaluation methods for pumped storage units, saves the time required for model training, and fully describes the fuzziness, instantaneous volatility, and other uncertainties in the working state of pumped storage units using the normal cloud model. Description of the Drawings

[0079] To more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0080] Figure 1 A schematic flow chart showing a method for detecting abnormal vibration of a pumped-storage unit provided by an embodiment of this specification;

[0081] Figure 2 A schematic diagram showing the effect of the adaptive threshold of the upper guide x-directional swing provided by an embodiment of this specification;

[0082] Figure 3 A schematic diagram showing the comparison of two cloud maps provided by an embodiment of this specification;

[0083] Figure 4 A schematic diagram showing the normal cloud entropy-containing expectation curve provided by an embodiment of this specification;

[0084] Figure 5 A schematic flow chart showing a method for detecting abnormal vibration of a pumped-storage unit provided by an embodiment of this specification;

[0085] Figure 6 A schematic diagram showing a method for detecting abnormal vibration of a pumped-storage unit provided by an embodiment of this specification;

[0086] Figure 7 A schematic diagram showing a device for detecting abnormal vibration of a pumped-storage unit provided by an embodiment of this specification. Detailed implementation manners

[0087] The following describes the solutions provided in this specification with reference to the drawings.

[0088] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will describe the technical solutions in the embodiments of the present application with reference to the drawings.

[0089] In the description of the embodiments of the present application, words such as "exemplary", "for example", or "for illustration" are used to represent examples, illustrations, or explanations. Any embodiment or design solution described as "exemplary", "for example", or "for illustration" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary", "for example", or "for illustration" is intended to present relevant concepts in a specific manner.

[0090] In the description of the embodiments of the present application, the term "and / or" is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, A and / or B may represent three situations: A exists alone, B exists alone, and A and B exist simultaneously. Additionally, unless otherwise specified, the meaning of the term "plural" refers to two or more than two.

[0091] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. The terms "include", "comprise", "have" and their variants all mean "including but not limited to", unless otherwise particularly emphasized in other ways.

[0092] The pumped-storage unit is an important part of a high-proportion new energy power system, which can effectively solve the intermittency and instability problems of new energy power generation (such as wind energy and solar energy). By pumping and storing energy when power is in excess and generating electricity when power is insufficient, the energy balance of the power system is achieved. As the core equipment of a pumped-storage power station, the pumped-storage unit undertakes the important task of maintaining the grid frequency and voltage stability and is the key support for the safe operation of the grid. The pumped-storage unit needs to frequently switch between the pumping and generating conditions, and this frequent change of conditions results in a complex operating state of the unit. The unit needs to frequently start and stop and perform forward and reverse operations under different conditions, and this operating mode poses extremely high requirements on the mechanical structure and control system of the unit. Under different operating conditions (such as pumping, generating, starting, stopping, etc.), the vibration characteristics of the pumped-storage unit are significantly different. This difference makes vibration monitoring and fault diagnosis more complex. Frequent condition conversion and start-up and shutdown operations will cause changes in the vibration characteristics of the unit, increasing the uncertainty of vibration data. Abnormal vibration data may indicate mechanical faults or imbalance problems inside the unit. If not discovered and processed in time, it may lead to unstable operation of the unit and even cause shutdown accidents. Continuous abnormal vibration may cause fatigue damage to the unit components, such as bearing wear, shaft deformation, blade damage, etc., increasing the equipment maintenance cost and downtime. In extreme cases, vibration problems may cause equipment damage and even trigger major accidents such as explosions and fires, posing a serious threat to the personal safety of on-site personnel.

[0093] In the prior art, the vibration monitoring, acquisition, and evaluation of pumped-storage units are key technologies to ensure their stable operation. Currently, the mainstream method for evaluating the state of pumped-storage units is mainly based on a health benchmark model, which is achieved by exploring the high-dimensional coupling relationship between the vibration of the unit and the operating conditions parameters. However, this method has some limitations. Establishing a health benchmark model requires training a very complex model, which is insufficient in terms of real-time performance. For example, although the health state model based on the Bagging algorithm can better describe the operating characteristics of the unit, its training process and computational complexity are relatively high. The model with point estimation as the evaluation criterion cannot fully reflect the randomness, fuzziness, and instantaneous volatility in the operating state of pumped-storage units. For example, traditional vibration prediction methods have poor reliability and accuracy when dealing with non-linear and non-stationary vibration signals. The operating conditions of pumped-storage units are complex, and frequent start-stop and forward-reverse operations lead to drastic changes in vibration data, making it difficult to conduct quantitative evaluation.

[0094] In summary, the development of vibration anomaly detection technology for pumped-storage units is of great significance for ensuring the safe operation of power stations and improving power generation efficiency. However, due to the complexity of vibration signals and the limitations of existing prediction methods, developing a more rapid and quantifiable vibration data anomaly detection technology for pumped-storage units has become an urgent problem in the industry. This application aims to propose a new vibration anomaly detection method to improve the real-time performance of prediction, thereby better ensuring the safe and stable operation of pumped-storage units.

[0095] In view of this, the present invention proposes a vibration anomaly detection method for pumped-storage units based on an adaptive threshold combined with a normal cloud model. The result of the adaptive threshold is used as the maximum vibration value of the pumped-storage unit. In this paper, the significance of threshold setting is to define the alarm value in the healthy working state of the pumped-storage unit. As Figures 1 to 5 shown, the vibration anomaly detection method for pumped-storage units based on an adaptive threshold combined with a normal cloud model of the present invention specifically includes the following steps:

[0096] Step 1, extraction and denoising of steady-state data of pumped storage units, specifically including: Due to the violent fluctuation of the swing over time, directly using these data increases the difficulty of the invention of this article. The steady-state monitoring data of the unit 10 minutes after startup and 5 minutes before shutdown are extracted as the steady-state data of the pumped storage unit. And the vibration values ​​of the four swings of upper x-direction swing, upper y-direction swing, lower x-direction swing, and lower y-direction swing are selected as corresponding performance indicators. These indicators can fully reflect the vibration characteristics of the unit in different directions. The unit stability data extracted in the previous step is first decomposed by wavelet. If the absolute value of the wavelet coefficient is less than the threshold, it is set to zero; if it is greater than or equal to the threshold, the original value is retained. After threshold processing, the modified wavelet coefficient is used to reconstruct the signal, that is, converting from the wavelet domain back to the time domain. This step usually involves inverse wavelet transform. The reconstructed signal is the denoised stable vibration signal, and the first step ends here.

[0097] Step 2: Input the data of the unit in healthy working state and determine the vibration threshold of the unit through Gaussian cloud model, including:

[0098] Step 2.1, extract the steady-state vibration data of the pumped storage unit 10 minutes after startup and 5 minutes before shutdown, including the upper X-direction, upper Y-direction, lower X-direction and lower Y-direction swings, set a sliding window, and each sliding window contains 10 consecutive data points.

[0099] Step 2.2, traverse each window and use the GMM model to fit the data in each window. The construction requires 4 Gaussian distributions, and the GMM model is used to fit each window.

[0100] Step 2.3, try to find the Gaussian distribution of the data represented by each window, and find the corresponding threshold according to the 3σ principle.

[0101] like Figure 2 Schematic diagram of the adaptive threshold value of the upper guide X-direction swing. According to the Gaussian distribution of the upper guide X-direction swing value fluctuating over time as shown in the figure, the threshold value corresponding to the upper guide X-direction swing value is determined.

[0102] In step 2, the Gaussian mixture model GMM is used to model the vibration data in each direction, and the parameters (mean, covariance matrix and mixing weight) of the GMM are estimated by the expectation maximization (EM) algorithm. According to the parameters of the GMM, the threshold of the vibration data in each direction is calculated. The threshold can be defined as the mean of the GMM plus several times the standard deviation (for example, mean ± 3σ), which is used to distinguish normal from abnormal vibration.

[0103] Step 3, after completing the previous step, perform cloud transformation on the real-time collected data and the corresponding threshold data respectively to obtain two cloud diagrams. Specifically, it includes: performing normal cloud transformation on the vibration data collected at the operating moment of the pumped-storage unit, and transforming the processed data into two cloud diagrams, namely the observation cloud and the threshold cloud. The principle of the normal cloud is as follows: The overall of a concept is characterized by three digital features, Ex, En, and He. Ex is the mathematical expectation in the spatial distribution, En represents the uncertainty measure of a qualitative concept, and He is the uncertainty measure of entropy, that is, the entropy of entropy. The detailed steps include: First, calculate the sample point mean Sample first-order absolute central moment Sample second-order central moment According to the distribution characteristics of the normal cloud, when the first-order absolute central moment of the normal cloud is Then Calculate according to the characteristics of the variance of the normal cloud distribution to obtain After calculation, three parameters are obtained, and the conversion of the two cloud diagrams is realized through inverse cloud transformation. Inverse cloud transformation can define a normal cloud model through the three parameters Ex, En, and He obtained by calculation. Inverse cloud transformation is a process of converting deterministic data into a cloud model with uncertainty. Specifically, the vibration data and the threshold are respectively converted into the observation cloud and the threshold cloud. Through inverse cloud transformation, the vibration data and the threshold can be converted from numerical form into a cloud model with uncertainty, which is convenient for subsequent comparison and analysis.

[0104] Such as Figure 3 The scatter plot shows the relationship between the X-directional swing of the upper guide and the parameter H, as well as the distribution of the observation cloud and the threshold cloud. By comparing the distribution of the observation cloud and the threshold cloud, the operating state of the unit can be evaluated. If most of the data points of the observation cloud fall within the range of the threshold cloud, it indicates that the unit is operating normally. If a large number of data points of the observation cloud exceed the range of the threshold cloud, it may indicate that there is an abnormality in the unit and further inspection and maintenance are required.

[0105] Such as Figure 4 It shows the distribution of the X-directional horizontal swing of the upper guide bearing, and the data is fitted through the normal cloud model. By comparing the actual data (blue scatter points) with the expected curve (red curve) and the normal cloud model (green curve), the distribution characteristics and uncertainty of the data can be evaluated. It helps to identify potential outliers or deviations from the normal distribution, providing a basis for the monitoring of the operating state and fault diagnosis of the equipment.

[0106] Step 4, compare the difference degree between the two cloud diagrams to determine the operating state of the unit.

[0107] Convert the threshold obtained by the Gaussian mixture model into the corresponding threshold cloud, convert the vibration data of the pumped-storage unit during the operation time into the corresponding observed cloud, and determine the difference between the two clouds by comparing the Wasserstein distance between the two clouds. By comparing the characteristics of the two cloud diagrams, the present application adopts a concept similarity comparison method based on the cloud model (LCIM) to compare the similarity of the two cloud diagrams.

[0108] The specific process is as follows:

[0109] First, let C1 = (Ex1, En1, He1) and C2 = (Ex2, En2, He2). Subsequently, calculate the similarity. The similarity formula is as follows: The greater the similarity, the closer the two clouds are, and at the same time, it means that the vibration value of the pumped-storage unit is close to the threshold, and the failure risk of the equipment increases.

[0110] As Figure 5 shown, through the above method, it effectively overcomes the process of establishing a complex health model required by the existing pumped-storage unit evaluation method, saves the time required for model training, and fully describes the fuzziness, instantaneous volatility, and other uncertainties in the working state of the pumped-storage unit using the normal cloud model.

[0111] Next, in combination with Figure 6 the vibration anomaly detection method of the pumped-storage unit of the present invention will be introduced in detail. Specifically, it includes the following steps:

[0112] 110. Collect the steady-state data of the pumped-storage unit during healthy operation, and use the Gaussian cloud model and the steady-state data to determine the vibration threshold of the unit. The vibration threshold includes the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold;

[0113] 120. Collect the swing vibration values of the pumped-storage unit during the current operation. The swing vibration values include the upper guide x-direction swing vibration value, the upper guide y-direction swing vibration value, the lower guide x-direction swing vibration value, and the lower guide y-direction swing vibration value;

[0114] 130. Convert the vibration threshold into the corresponding threshold cloud, and convert the swing vibration value into the corresponding observed cloud;

[0115] 140. Determine the failure risk of the pumped-storage unit at the current moment through the Wasserstein distance between the threshold cloud and the observed cloud.

[0116] Based on the above further embodiments, in step 110, collecting the steady-state data of the pumped-storage unit during healthy operation and using the Gaussian cloud model and the steady-state data to determine the vibration threshold of the unit specifically includes:

[0117] Collect the steady-state data of the pumped storage unit during healthy operation, and perform denoising processing on the steady-state data, where the steady-state data includes the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value;

[0118] Use the steady-state data and the Gaussian cloud model to determine the vibration threshold of the unit, and the vibration threshold includes the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold.

[0119] Based on the above further embodiments, in step 110, collecting the steady-state data of the pumped storage unit during healthy operation and performing denoising processing on the steady-state data includes:

[0120] Collect the steady-state data of the pumped storage unit during a preset period of healthy operation;

[0121] Perform wavelet decomposition on the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value in the steady-state data respectively;

[0122] Judge whether the absolute value of each obtained wavelet coefficient is less than a preset threshold;

[0123] If so, set the wavelet coefficient to zero;

[0124] Otherwise, keep the original value of the wavelet coefficient;

[0125] Use the modified wavelet coefficients to perform signal reconstruction on the steady-state data, that is, convert back from the wavelet domain to the time domain, and the reconstructed steady-state data is the denoised steady vibration signal.

[0126] Based on the above further embodiments, step 120 includes:

[0127] Set a sliding window for the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value in the steady-state data respectively, and each sliding window contains continuous data points;

[0128] Determine a Gaussian distribution for the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value respectively, and perform modeling with four Gaussian distributions to initialize the parameters of the Gaussian cloud model GMM;

[0129] Traverse each sliding window, use GMM to fit the data in the window, and optimize the parameters of the GMM model through the expectation maximization algorithm;

[0130] For each sliding window, according to the four Gaussian distributions of the GMM, find the Gaussian distribution with the largest weight, and determine the vibration threshold of the unit according to the 3σ principle.

[0131] Based on the above further embodiments, in step 130, converting the vibration threshold into a corresponding threshold cloud includes:

[0132] Calculate the mean value of each vibration threshold respectively according to the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold. where x i is the i-th vibration threshold, the first-order absolute central moment the second-order central moment

[0133] According to the distribution characteristics of the normal cloud, when the first-order absolute central moment of the first normal cloud is the expected value then the entropy

[0134] Calculate the hyperentropy according to the characteristics of the variance of the normal cloud distribution. Obtain the expected value Ex1, entropy En1, and hyperentropy He1 of each swing vibration threshold.

[0135] For the vibration threshold in each direction, define a normal threshold cloud model using the expected value Ex1, entropy En1, and hyperentropy He1.

[0136] Through inverse cloud transformation, convert the normal threshold cloud model into the threshold cloud.

[0137] Based on the above further embodiments, in step 130, converting the swing vibration value into a corresponding observed cloud includes:

[0138] Calculate the mean value of the vibration threshold in each direction respectively according to the upper guide x-direction swing vibration value, the upper guide y-direction swing vibration value, the lower guide x-direction swing vibration value, and the lower guide y-direction swing vibration value. where y i is the i-th swing vibration value, n is the number of swing vibration values in each direction, the first-order absolute central moment the second-order central moment

[0139] According to the distribution characteristics of the normal cloud, when the first-order absolute central moment of the normal cloud is the expected value then the entropy

[0140] Calculate the hyperentropy according to the characteristics of the variance of the normal cloud distribution Obtain the expected values Ey2, entropy En2, and hyperentropy He2 of the vibration thresholds in each direction;

[0141] For the swing vibration values in each direction, define a normal swing vibration cloud model using the expected value Ex, entropy En2, and hyperentropy He2;

[0142] Through inverse cloud transformation, convert the normal swing vibration cloud model into the observed cloud.

[0143] Based on the above further embodiments, step 140 includes:

[0144] Based on the cloud model concept similarity comparison method LCIM, compare the similarity between the threshold cloud and the observed cloud;

[0145] Let the threshold cloud C1 = (Ex1, En1, He1), and the observed cloud C2 = (Ey2, En2, He2);

[0146] Use the similarity formula as follows:

[0147]

[0148] Obtain the similarity between the threshold cloud and the observed cloud, and determine the fault risk of the pumped storage unit at the current moment according to the similarity.

[0149] In the above method provided by the embodiments of this specification, it effectively overcomes the process of establishing a complex health model in the existing evaluation method for pumped storage units, saves the time required for model training, and fully characterizes the fuzziness, instantaneous volatility, and other uncertainties in the working state of pumped storage units using the normal cloud model.

[0150] Next, in combination with Figure 7 A detailed introduction to the vibration anomaly detection device for the pumped storage unit of the present invention will be given. Specifically, it includes:

[0151] The first processing module is used to collect the steady-state data during the healthy operation of the pumped storage unit, and use the Gaussian cloud model and the steady-state data to determine the vibration threshold of the unit. The vibration threshold includes the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold;

[0152] The second processing module is used to collect the swing vibration values during the current operation of the pumped storage unit. The swing vibration values include the upper guide x-direction swing vibration value, the upper guide y-direction swing vibration value, the lower guide x-direction swing vibration value, and the lower guide y-direction swing vibration value;

[0153] The third processing module is used to convert the vibration threshold into a corresponding threshold cloud and convert the swing vibration value into a corresponding observation cloud;

[0154] The fourth processing module is used to determine the fault risk of the pumped-storage unit at the current moment through the Wasserstein distance between the threshold cloud and the observation cloud.

[0155] Based on the above further embodiments, the first processing module is specifically configured to collect the steady-state data during the healthy operation of the pumped-storage unit and perform denoising processing on the steady-state data, where the steady-state data includes the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value;

[0156] Using the steady-state data and the Gaussian cloud model, determine the vibration threshold of the unit, where the vibration threshold includes the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold.

[0157] Based on the above further embodiments, the first processing module is specifically configured to collect the steady-state data of the pumped-storage unit during a preset period when it is operating healthily;

[0158] Perform wavelet decomposition on the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value in the steady-state data respectively;

[0159] Judge whether the absolute value of each wavelet coefficient obtained is less than a preset threshold;

[0160] If so, set the wavelet coefficient to zero;

[0161] Otherwise, keep the original value of the wavelet coefficient;

[0162] Use the modified wavelet coefficients to perform signal reconstruction on the steady-state data, that is, convert back from the wavelet domain to the time domain, and the reconstructed steady-state data is the denoised stationary vibration signal.

[0163] Based on the above further embodiments, the second processing module is specifically configured to set a sliding window for the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value in the steady-state data respectively, and each sliding window contains continuous data points;

[0164] Determine a Gaussian distribution for the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value respectively, obtain four Gaussian distributions for modeling, and initialize the parameters of the Gaussian cloud model GMM;

[0165] Traverse each sliding window, use GMM to fit the data within the window, and optimize the parameters of the GMM model through the expectation maximization algorithm;

[0166] For each sliding window, according to the four Gaussian distributions of the GMM, find the Gaussian distribution with the largest weight, and determine the vibration threshold of the unit according to the 3σ principle.

[0167] Based on the above further embodiments, the third processing module is specifically configured to calculate the mean value of each vibration threshold according to the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold where x i is the i-th vibration threshold, the first-order absolute central moment the second-order central moment

[0168] According to the distribution characteristics of the normal cloud, when the first-order absolute central moment of the first normal cloud, i.e., the expected value then the entropy

[0169] Calculate the hyperentropy according to the characteristics of the variance of the normal cloud distribution Obtain the expected value Ex1, entropy En1, and hyperentropy He1 of each swing vibration threshold;

[0170] For the vibration threshold in each direction, define the normal threshold cloud model using the expected value Ex1, entropy En1, and hyperentropy He1;

[0171] Through inverse cloud transformation, convert the normal threshold cloud model into the threshold cloud.

[0172] Based on the above further embodiments, the third processing module is specifically configured to calculate the mean value of the vibration threshold in each direction according to the upper guide x-direction swing vibration value, the upper guide y-direction swing vibration value, the lower guide x-direction swing vibration value, and the lower guide y-direction swing vibration value where y i is the i-th swing vibration value, n is the number of swing vibration values in each direction, the first-order absolute central moment the second-order central moment

[0173] According to the distribution characteristics of the normal cloud, when the first-order absolute central moment of the normal cloud, i.e., the expected value then the entropy

[0174] Calculate the hyperentropy according to the characteristics of the variance of the normal cloud distribution Obtain the expected values Ey2, entropy En2, and hyper-entropy He2 of the vibration thresholds in all directions;

[0175] For the swing vibration values in each direction, define a normal swing vibration cloud model using the expected value Ex, entropy En2, and hyper-entropy He2;

[0176] Through inverse cloud transformation, convert the normal swing vibration cloud model into the observed cloud.

[0177] Based on the above further embodiments, the fourth processing module is specifically configured to compare the similarity between the threshold cloud and the observed cloud based on the concept similarity comparison method LCIM of the cloud model;

[0178] Let the threshold cloud C1 = (Ex1, En1, He1), and the observed cloud C2 = (Ey2, En2, He2);

[0179] Use the similarity formula as follows:

[0180]

[0181] Obtain the similarity between the threshold cloud and the observed cloud, and determine the fault risk of the pumped storage unit at the current moment according to the similarity.

[0182] In the above method provided by the embodiments of this specification, it effectively overcomes the process of establishing a complex health model in the existing evaluation method of pumped storage units, saves the time required for model training, and fully characterizes the fuzziness, instantaneous volatility, and other uncertainties in the working state of pumped storage units using the normal cloud model.

[0183] The present invention also provides a computer storage medium, on which a computer program is stored. When the computer program is executed by one or more processors, it implements the vibration anomaly detection method for pumped storage units as described in any one of the above technical solutions.

[0184] In addition, an electronic device is also provided, including a memory and one or more processors. A computer program is stored on the memory, and when the computer program is executed by the one or more processors, it implements the vibration anomaly detection method for pumped storage units as described in any one of the above technical solutions.

[0185] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of this application. It should be understood that the above are only specific embodiments of this application and are not used to limit the protection scope of this application. Any modifications, equivalent replacements, improvements, etc. made on the basis of the technical solutions of this application should be included within the protection scope of this application.

Claims

1. A method for detecting abnormal vibration of a pumped-storage unit, characterized in that, The method includes: Collecting the steady-state data of the pumped-storage unit during healthy operation, and using the Gaussian cloud model and the steady-state data to determine the vibration thresholds of the unit, where the vibration thresholds include the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold; Collecting the swing vibration values of the pumped-storage unit during operation at the current moment, where the swing vibration values include the upper guide x-direction swing vibration value, the upper guide y-direction swing vibration value, the lower guide x-direction swing vibration value, and the lower guide y-direction swing vibration value; Converting the vibration thresholds into corresponding threshold clouds, and converting the swing vibration values into corresponding observation clouds; Determining the fault risk of the pumped-storage unit at the current moment through the Wasserstein distance between the threshold cloud and the observation cloud.

2. The method according to claim 1, characterized in that The step of collecting the steady-state data of the pumped-storage unit during healthy operation and using the Gaussian cloud model and the steady-state data to determine the vibration thresholds of the unit specifically includes: Collecting the steady-state data of the pumped-storage unit during healthy operation, and performing denoising processing on the steady-state data, where the steady-state data includes the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value; Using the steady-state data and the Gaussian cloud model to determine the vibration thresholds of the unit, where the vibration thresholds include the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold.

3. The method according to claim 2, wherein The step of collecting the steady-state data of the pumped-storage unit during healthy operation and performing denoising processing on the steady-state data includes: Collecting the steady-state data of the pumped-storage unit during a preset period of healthy operation; Performing wavelet decomposition on the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value in the steady-state data respectively; Judging whether the absolute value of each obtained wavelet coefficient is less than a preset threshold; If so, setting the wavelet coefficient to zero; Otherwise, retaining the original value of the wavelet coefficient; Using the modified wavelet coefficients to perform signal reconstruction on the steady-state data, that is, converting back from the wavelet domain to the time domain, and the reconstructed steady-state data is the denoised steady vibration signal.

4. The method according to claim 2, wherein The step of using the steady-state data and the Gaussian cloud model to determine the vibration thresholds of the unit includes: Setting a sliding window for the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value in the steady-state data respectively, and each sliding window contains continuous data points; Determining a Gaussian distribution for the first upper guide x-direction swing vibration value, the first upper guide y-direction swing vibration value, the first lower guide x-direction swing vibration value, and the first lower guide y-direction swing vibration value respectively, obtaining four Gaussian distributions for modeling, and initializing the parameters of the Gaussian cloud model GMM; Traversing each sliding window, using GMM to fit the data in the window, and optimizing the parameters of the GMM model through the expectation maximization algorithm. For each sliding window, according to the four Gaussian distributions of the GMM, find the Gaussian distribution with the largest weight, and determine the vibration threshold of the unit according to the 3σ principle.

5. The method according to claim 1, characterized in that The conversion of the vibration threshold into the corresponding threshold cloud includes: Calculate the mean value of each vibration threshold according to the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold respectively where x i is the i-th vibration threshold, the first-order absolute central moment the second-order central moment According to the distribution characteristics of the normal cloud, when the first-order absolute central moment of the first normal cloud, i.e., the expected value then the entropy Calculate the hyperentropy based on the characteristics of the variance of the normal cloud distribution Obtain the expected values Ex1, entropy En1, and hyperentropy He1 of the vibration thresholds of each swing For the vibration threshold in each direction, define a normal threshold cloud model using the expected value Ex1, entropy En1, and hyperentropy He1; Through inverse cloud transformation, convert the normal threshold cloud model into the threshold cloud.

6. The method according to claim 5, wherein The conversion of the swing vibration value into the corresponding observed cloud includes: Calculate the mean value of the vibration thresholds in each direction according to the upper guide x-direction swing vibration value, the upper guide y-direction swing vibration value, the lower guide x-direction swing vibration value, and the lower guide y-direction swing vibration value respectively. where y i is the i-th swing vibration value, n is the number of swing vibration values in each direction, the first-order absolute central moment the second-order central moment According to the distribution characteristics of the normal cloud, when the first-order absolute central moment of the normal cloud, that is, the expected value then the entropy Calculate the hyperentropy according to the characteristics of the variance of the normal cloud distribution Obtain the expected values Ey2, entropy En2, and hyperentropy He2 of the vibration thresholds in each direction; For the swing vibration value in each direction, define a normal swing vibration cloud model using the expected value Ex, entropy En2, and hyperentropy He2; Through inverse cloud transformation, convert the normal swing vibration cloud model into the observed cloud.

7. The method according to claim 6, characterized in that, The determination of the fault risk of the pumped-storage unit at the current moment through the Wasserstein distance between the threshold cloud and the observed cloud includes: Based on the cloud model concept similarity comparison method LCIM, compare the similarity between the threshold cloud and the observed cloud; Let the threshold cloud C1 = (Ex1, En1, He1), and the observed cloud C2 = (Ey2, En2, He2); Use the similarity formula as follows: Obtain the similarity between the threshold cloud and the observed cloud, and determine the fault risk of the pumped-storage unit at the current moment according to the similarity.

8. A vibration anomaly detection device for a pumped storage unit, characterized in that The device includes: A first processing module, configured to collect the steady-state data when the pumped-storage unit is operating healthily, and use the Gaussian cloud model and the steady-state data to determine the vibration threshold of the unit, where the vibration threshold includes the upper guide x-direction swing vibration threshold, the upper guide y-direction swing vibration threshold, the lower guide x-direction swing vibration threshold, and the lower guide y-direction swing vibration threshold; A second processing module, configured to collect the swing vibration values of the pumped-storage unit during operation at the current moment, where the swing vibration values include the upper guide x-direction swing vibration value, the upper guide y-direction swing vibration value, the lower guide x-direction swing vibration value, and the lower guide y-direction swing vibration value; A third processing module, configured to convert the vibration threshold into the corresponding threshold cloud and convert the swing vibration value into the corresponding observed cloud; A fourth processing module, configured to determine the fault risk of the pumped-storage unit at the current moment through the Wasserstein distance between the threshold cloud and the observed cloud.

9. A computer storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by one or more processors, it implements the vibration anomaly detection method for the pumped-storage unit according to any one of claims 1 to 7.

10. An electronic device, characterized in that, It includes a memory and one or more processors, and a computer program is stored on the memory. When the computer program is executed by the one or more processors, it implements the vibration anomaly detection method for the pumped-storage unit according to any one of claims 1 to 7.

Images