A method and device for predicting degradation trend of pumped storage unit under variable operating conditions

By constructing a health data set and training machine learning model, combining entropy weight method and similarity calculation, the CNN-GRU model is used to predict the future deterioration trend of pumped storage units, and the existing technology has not considered operating conditions and component impact. The problem has achieved high-quality deterioration trend forecast and unit health management.

CN116596120BActive Publication Date: 2024-11-26CHINA THREE GORGES CORPORATION
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310420212.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-04-18
Publication Date
2024-11-26
Estimated Expiration
2043-04-18

AI Technical Summary

Technical Problem

The existing pumped storage unit deterioration trend prediction method does not fully consider the impact of operating conditions and different components on overall health performance, resulting in low monitoring data quality and lack of practical guidance.

Method used

By constructing a health data set, analyzing the impact of working condition parameters on vibration characteristics, screening the main working condition parameters and vibration monitoring feature quantities, training machine learning models, determining degradation indexes, predicting future degradation trends, combining entropy weight method and similarity calculations , predict using the CNN-GRU model.

Benefits of technology

The quality of the unit's on-site monitoring data has been improved, the healthy status has been fully characterized, and the future deterioration trend has been effectively predicted, supporting the unit's active maintenance and safe and stable operation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116596120B_ABST
    Figure CN116596120B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and device for predicting the degradation trend of a pumped storage unit under variable operating conditions. The method includes: constructing a healthy data set after cleaning the historical stable operation monitoring data of the unit; analyzing the influence of the operating condition parameters on the vibration of key components and the correlation between different vibration monitoring feature quantities to screen the healthy data set, and obtaining a healthy training set to train a preset first machine learning model to obtain a standard health model of the unit; inputting the real-time monitoring operating condition parameters into the standard health model to obtain the standard value of the corresponding multi-source vibration monitoring feature quantity, determining the weight of each vibration monitoring quantity by the entropy weight method, and determining the degradation index based on the preset similarity method for the standard value of the multi-source vibration monitoring quantity and the obtained real-time monitoring value; extracting multiple degradation indexes and arranging them in chronological order to obtain a degradation sequence to train a preset second machine learning model to obtain a degradation trend prediction model. The present invention can comprehensively characterize the health status level of the unit and realize the effective prediction of the degradation trend of the unit.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of pumped storage unit degradation trend prediction, and in particular to a method and device for predicting a pumped storage unit degradation trend under variable operating conditions. Background Art

[0002] Pumped storage units are known as the "crown jewel" in the field of hydropower technology. Their safe, stable and efficient operation is of vital importance to the safety of power stations and the stability of power grids. Due to the harsh operating environment of the units, frequent switching of operating conditions, and the influence of water-machine-electricity coupling factors, unit failures are diversified, mixed and gradually developed. A large number of unit failure cases at home and abroad have shown that once a unit fails, it will affect the life of the unit and the stability of the power grid at the least, and cause huge economic losses and serious safety accidents at the worst. Therefore, it is urgent to use advanced theoretical methods and complete technical means to strengthen the management and maintenance of the health status of pumped storage units, grasp the health status of the units in real time, predict potential failures of the units in time, and scientifically guide the status maintenance of the units, so as to avoid the occurrence of safety accidents.

[0003] The degradation trend prediction of pumped storage units analyzes the evolution characteristics of unit performance degradation, comprehensively utilizes historical monitoring data and real-time online monitoring data, predicts the evolution trend of unit performance, and timely captures early signs of failure, which is of great significance for realizing active maintenance and health management of units. The core task of degradation trend prediction is to build a reasonable and comprehensive degradation degree model and an effective prediction model. In the specific implementation process, the current research mainly has the following problems: (1) The operating environment of pumped-storage units is harsh and the operating conditions are complex and changeable. At the same time, they are affected by multiple factors such as monitoring equipment and on-site environment, which makes the online monitoring system of the unit frequently experience occasional failures such as sensor failure, acquisition equipment failure, and communication interruption, resulting in low quality of on-site monitoring data of the unit, which greatly affects the reliability of subsequent unit degradation assessment and degradation trend prediction; (2) Pumped-storage units are frequently started and stopped, and the operating conditions change complexly. When predicting the characteristic parameters, the changes in the operating conditions of the pumped-storage units are not fully considered, resulting in the lack of engineering practice guidance significance in the prediction results; (3) When constructing the degradation model, only the monitoring data of a single component of the unit is considered, ignoring the impact of different components of the unit on its overall health performance, so it cannot fully represent the actual health status of the unit. Summary of the invention

[0004] Therefore, in order to overcome the problem that traditional degradation trend prediction does not fully consider the changes in the operating conditions of the pumped-storage unit and the impact of different components of the unit on its overall health performance, the present invention provides a method and device for predicting the degradation trend of the pumped-storage unit under varying operating conditions, which fully considers the impact of changes in the operating conditions of the pumped-storage unit on the vibration characteristics and the correlation between multi-source vibration quantities, and can comprehensively and effectively characterize the health status level of the unit and effectively predict the future degradation trend of the unit.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a method for predicting a degradation trend of a pumped storage unit under variable operating conditions, comprising:

[0007] Based on the preset processing algorithm, the historical stable operation monitoring data of the pumped storage unit is cleaned and processed to construct a healthy data set, which contains the multi-source vibration monitoring feature quantities of different operating condition parameters and key components of the unit;

[0008] Analyze the impact of operating parameters on the vibration of key components to determine the main operating parameters of the unit and the correlation between different vibration monitoring feature quantities; screen the health data set based on the correlation analysis results and the main operating parameters of the unit to obtain a health training set; train the preset first machine learning model based on the health training set to obtain a standard health model of the unit;

[0009] Obtain the real-time monitoring value of the multi-source vibration monitoring characteristic quantity of the pumped storage unit and the corresponding operating condition parameters, input the corresponding operating condition parameters into the standard health model of the unit to obtain the standard value of the multi-source vibration monitoring characteristic quantity of the corresponding unit, determine the weight of each vibration monitoring quantity in the multi-source vibration monitoring characteristic quantity of the unit by the entropy weight method, and determine the unit degradation index based on the preset similarity calculation method for the standard value and the real-time monitoring value of the multi-source vibration monitoring quantity;

[0010] Based on a preset time period, multiple unit degradation indexes are extracted and arranged in chronological order to obtain a unit historical degradation degree sequence. The unit historical degradation degree sequence is used to train the preset second machine learning model to obtain a unit degradation trend prediction model, which is used to predict the future degradation trend of the unit.

[0011] Preferably, the main operating parameters of the unit include: working water head, active power, unit speed and guide vane opening; the vibration monitoring characteristic quantities include: vibration data of the upper guide bearing, lower guide bearing, water guide bearing, upper frame, lower frame, stator and top cover.

[0012] Preferably, a set F(t) of mapping relationships between main operating parameters of the unit and multi-source vibration monitoring characteristic quantities is obtained through the unit standard health model, and F(t) is expressed as:

[0013] F(t)={f0(H(t),P(t),N(t),G(t)),f1(H(t),P(t),N(t),G(t)),…,f i (H(t),P(t),N(t),G(t))}

[0014] Where i is the number of multi-source vibration monitoring quantities, f i () represents the mapping relationship between the i-th vibration monitoring characteristic quantity at time t and the main operating parameters of the unit, H(t) represents the working water head at time t, P(t) represents the active power at time t, N(t) represents the unit speed at time t, and G(t) represents the guide vane opening at time t.

[0015] Preferably, the process of determining the unit degradation index based on a preset similarity calculation method for the standard value and the real-time monitoring value of the multi-source vibration monitoring quantity includes: characterizing the degree of deviation between the standard value and the real-time monitoring value of the multi-source vibration monitoring quantity based on the Euclidean distance, and determining the unit degradation index.

[0016] Preferably, the process of determining the weight of each vibration monitoring quantity in the multi-source vibration monitoring characteristic quantity of the unit by the entropy weight method includes:

[0017] The maximum and minimum normalization method is used to convert the multi-source vibration monitoring characteristic data V within a preset time period into ij Perform normalization:

[0018]

[0019] Wherein, i is the i-th vibration monitoring quantity in the multi-source vibration monitoring quantity, N is the number of multi-source vibration monitoring characteristic quantities, j is the monitoring time, and T is the monitoring duration;

[0020] Calculate the proportion p of the i-th vibration monitoring quantity at the j-th moment ij for:

[0021]

[0022] Calculate the entropy value E of the i-th vibration monitoring quantity i for:

[0023]

[0024] Then the weight w of the i-th vibration monitoring quantity is i for:

[0025]

[0026] Preferably, the unit degradation index PDI is calculated by the following formula:

[0027]

[0028] Among them, V i (t) represents the monitoring value of the ith vibration monitoring characteristic quantity at time t, F i (t) represents the standard value of the i-th vibration monitoring characteristic quantity at time t, N is the number of multi-source vibration monitoring characteristic quantities, and w i (t) represents the weight of the i-th vibration monitoring value at time t,

[0029] Preferably, the process of obtaining the unit degradation trend prediction model includes: inputting the unit historical degradation degree sequence into the CNN-GRU model for training, and obtaining the trained model as the unit degradation trend prediction model.

[0030] In a second aspect, an embodiment of the present invention provides a device for predicting a degradation trend of a pumped storage unit under variable operating conditions, comprising:

[0031] The unit health data set construction module is used to build a health data set after cleaning and processing the historical stable operation monitoring data of the pumped storage unit based on a preset processing algorithm. The health data set contains the unit's different operating condition parameters and multi-source vibration monitoring feature quantities of key components;

[0032] The unit standard health model construction module is used to analyze the impact of operating condition parameters on the vibration of key components to determine the unit's main operating condition parameters, and analyze the correlation between different vibration monitoring feature quantities; based on the correlation analysis results and the unit's main operating condition parameters, the health data set is screened to obtain a health training set; based on the health training set, a preset first machine learning model is trained to obtain the unit's standard health model;

[0033] A unit degradation index determination module is used to obtain the real-time monitoring value of the multi-source vibration monitoring characteristic quantity of the pumped storage unit and the corresponding operating condition parameters, input the above operating condition parameters into the unit standard health model to obtain the standard value of the corresponding unit multi-source vibration monitoring characteristic quantity, determine the weight of each vibration monitoring quantity in the unit multi-source vibration monitoring characteristic quantity by the entropy weight method, and determine the unit degradation index based on the preset similarity calculation method for the standard value and the real-time monitoring value of the multi-source vibration monitoring quantity;

[0034] The degradation trend prediction module is used to extract multiple unit degradation indicators based on a preset time period, arrange them in chronological order to obtain a unit historical degradation degree sequence, and use the unit historical degradation degree sequence to train a preset second machine learning model to obtain a unit degradation trend prediction model, which is used to predict the future degradation trend of the unit.

[0035] In a third aspect, an embodiment of the present invention provides a computer device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes a method for predicting a degradation trend of a pumped-storage unit under variable operating conditions according to the first aspect of the embodiment of the present invention.

[0036] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute a method for predicting the degradation trend of a pumped-storage unit under variable operating conditions according to the first aspect of the embodiment of the present invention.

[0037] The technical solution of the present invention has the following advantages:

[0038] The present invention provides a method and device for predicting the degradation trend of a pumped storage unit under variable operating conditions. The method cleans the historical stable operation monitoring data of the unit and then constructs a healthy data set. The method analyzes the influence of the operating condition parameters on the vibration of key components and the correlation between different vibration monitoring feature quantities to screen the healthy data set, and obtains a healthy training set to train a preset first machine learning model to obtain a standard health model of the unit. The real-time monitoring operating condition parameters are input into the standard health model to obtain the standard value of the corresponding multi-source vibration monitoring feature quantity, and the weight of each vibration monitoring quantity is determined by the entropy weight method, and the degradation index is determined based on the preset similarity method for the standard value of the multi-source vibration monitoring quantity and the obtained real-time monitoring value. A plurality of degradation indexes are extracted and arranged in chronological order to obtain a degradation sequence, and a preset second machine learning model is trained to obtain a degradation trend prediction model, which improves the quality of the on-site monitoring data of the unit, fully considers the influence of the operating condition change of the pumped storage unit on the vibration characteristics and the correlation between the multi-source vibration quantities, can comprehensively and effectively characterize the health status level of the unit, and can effectively predict the future degradation trend of the unit, so as to provide technical support for ensuring the safe and stable operation of the unit, realizing the active maintenance of the unit, and avoiding the occurrence of potential failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 A schematic flow chart of a method for predicting the degradation trend of a pumped storage unit under variable operating conditions provided in an embodiment of the present invention;

[0041] Figure 2A schematic diagram of the monitoring arrangement of key components of a pumped storage unit provided in an embodiment of the present invention;

[0042] Figure 3 A specific process diagram of a method for predicting the degradation trend of a pumped storage unit under variable operating conditions provided in an embodiment of the present invention;

[0043] Figure 4 (a)- Figure 4 (c) is a schematic diagram of the analysis results of the Y-direction vibration of the lower guide bearing corresponding to different operating condition parameters provided in an embodiment of the present invention;

[0044] Figure 5 A schematic diagram of a confusion matrix of vibration monitoring values ​​provided in an embodiment of the present invention;

[0045] Figure 6 A schematic diagram of a PDI sequence of a unit provided in an embodiment of the present invention;

[0046] Figure 7 Module composition diagram of the device for predicting the degradation trend of a pumped storage unit under variable operating conditions provided in an embodiment of the present invention;

[0047] Figure 8 A composition diagram of a specific example of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only an embodiment of a part of the present invention, not all embodiments, and is not intended to limit the scope of the present invention. In addition, in the following description, the description of well-known structures and technologies is omitted to avoid unnecessary confusion of the concepts disclosed in the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.

[0049] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0050] Example 1

[0051] The embodiment of the present invention provides a method for predicting the degradation trend of a pumped storage unit under variable operating conditions, such as Figure 1 As shown, the method includes:

[0052] Step S1: Based on a preset processing algorithm, the historical stable operation monitoring data of the pumped storage unit is cleaned and processed to construct a healthy data set, which includes different operating condition parameters of the unit and multi-source vibration monitoring feature quantities of key components.

[0053] It should be noted that the historical stable operation monitoring data of the pumped storage unit is the monitoring data of the stable operation of the unit after the initial running-in period obtained from the pumped storage unit status monitoring system, which represents the operation data of the normal operation state (healthy state) of the unit. In this embodiment, the pumped storage unit includes three unit operating conditions: power generation, pumping, and no-load. It is only used as an example and is adaptively adjusted according to actual application needs.

[0054] In this embodiment, the preset processing algorithm is an unsupervised clustering algorithm, which is used to eliminate abnormal samples according to the attributes of the data itself, and obtain a clean healthy data set after cleaning the original monitoring data. Specifically, the unsupervised clustering algorithms include: K-Means clustering algorithm, hierarchical clustering algorithm, t-SNE clustering algorithm and DBSCAN clustering algorithm, which are only used as examples and are determined according to actual application needs. This improves the quality of the unit's on-site monitoring data through data cleaning, which helps to improve the reliability of subsequent unit degradation assessment and degradation trend prediction.

[0055] Step S2: Analyze the influence of operating condition parameters on the vibration of key components to determine the main operating parameters of the unit and the correlation between different vibration monitoring feature quantities; screen the health data set based on the correlation analysis results and the main operating parameters of the unit to obtain a health training set; train a preset first machine learning model based on the health training set to obtain a standard health model of the unit.

[0056] In this embodiment, the unit operating condition parameters in the health data set include: working water head, flow, active power, reactive power, unit speed and guide vane opening. By analyzing the influence of the above unit operating condition parameters on the vibration of key components, that is, the mapping relationship between them and each vibration monitoring feature quantity, the sensitive state parameters with a mapping relationship are selected as the main unit operating condition parameters. The main unit operating condition parameters after analysis in this embodiment include: working water head, active power, unit speed and guide vane opening, and the main unit operating condition parameters are used as the input of the unit standard health model.

[0057] In this embodiment, the vibration monitoring feature quantity in the health data set includes: vibration data of the upper guide bearing, the lower guide bearing, the water guide bearing, the upper frame, the lower frame, the stator and the top cover, and the schematic diagram of the monitoring arrangement of the key components of the pumped storage unit, such as Figure 2Specifically, 17 vibration monitoring data of 7 key components of the pumped storage unit, including the upper guide bearing, lower guide bearing, water guide bearing, upper frame, lower frame, stator, and top cover, are selected for correlation analysis. By calculating the Pearson correlation coefficient between each vibration monitoring quantity, the vibration monitoring quantity with small correlation and representativeness is selected to form the multi-source vibration monitoring characteristic quantity of the unit, which is used as the output of the standard health model of the unit. The situation of each vibration monitoring quantity in this embodiment is shown in Table 1.

[0058] Table 1 Vibration monitoring quantities

[0059]

[0060] In this embodiment, according to a preset ratio, the health data set is screened based on the correlation analysis results and the main operating parameters of the unit to obtain a healthy training set and divide it into a training set and a test set. Specifically, the preset ratio is 3:1, which is only used as an example and is adaptively adjusted according to the actual amount of data. The training set is used to train the preset first machine learning model, and a corresponding unit standard health model is established for each vibration monitoring quantity in the unit's multi-source vibration monitoring feature quantity, and based on the unit standard health model, a set F(t) of mapping relationships between the main operating parameters and the multi-source vibration monitoring feature quantities at any time is obtained, and F(t) is expressed as:

[0061] F(t)={f0(H(t),P(t),N(t),G(t)),f1(H(t),P(t),N(t),G(t)),…,f i (H(t),P(t),N(t),G(t))}

[0062] Where i is the number of multi-source vibration monitoring quantities, f i () represents the mapping relationship between the i-th vibration monitoring characteristic quantity at time t and the main operating parameters of the unit, H(t) represents the working water head at time t, P(t) represents the active power at time t, N(t) represents the unit speed at time t, and G(t) represents the guide vane opening at time t.

[0063] It should be noted that the present embodiment presets a first machine learning model, including: a BP neural network model, a DNN network model, an SVR model and a Gaussian process regression model, which is not specifically limited here and is selected according to actual application requirements.

[0064] In this embodiment, after obtaining the trained standard health model of the unit, the validity of the standard health model is verified using a test set based on preset evaluation indicators. Specifically, the preset evaluation indicators include: correlation coefficient (R), mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE), which are only used as examples and are not limited to this. The standard health model is quantitatively evaluated to evaluate the model fitting ability, wherein the larger the R value, the smaller the MAE, MAPE and RMSE values, which means the better the model fitting effect. The above-mentioned preset evaluation indicators are calculated based on the following formula, specifically:

[0065]

[0066]

[0067]

[0068]

[0069] Among them, f(t) is the actual monitoring value, f′(t) is the standard health model fitting value, and They represent the average values ​​of the actual monitoring value and the fitted value respectively, and N is the number of samples in the test set.

[0070] Step S3: Obtain the real-time monitoring value of the multi-source vibration monitoring characteristic quantity of the pumped-storage unit and the corresponding operating condition parameters, input the corresponding operating condition parameters into the standard health model of the unit to obtain the standard value of the multi-source vibration monitoring characteristic quantity of the corresponding unit, determine the weight of each vibration monitoring quantity in the multi-source vibration monitoring characteristic quantity of the unit by the entropy weight method, and determine the unit degradation index based on the preset similarity calculation method for the standard value and real-time monitoring value of the multi-source vibration monitoring quantity.

[0071] This embodiment is combined with engineering practice. Since the overall health level of the pumped storage unit is greatly affected by damaged components, a higher weight should be given to the corresponding component monitoring quantity. In addition, the vibration monitoring quantity of the damaged component has a large degree of variation, and the corresponding information entropy is small, so the entropy weight method is used to dynamically update the weights of the vibration monitoring quantities of different components. The entropy weight method can integrate the multi-source vibration monitoring quantity information of the unit and determine the weights of each vibration monitoring quantity in the multi-source vibration monitoring characteristic quantity of the unit. The calculation process of determining the weight by the entropy weight method includes:

[0072] Step S31: Use the maximum and minimum normalization method to normalize the multi-source vibration monitoring feature data V within a preset time period ij Perform normalization:

[0073]

[0074] Among them, i is the i-th vibration monitoring quantity in the multi-source vibration monitoring quantity, N is the number of multi-source vibration monitoring characteristic quantities, j is the monitoring time, and T is the monitoring duration.

[0075] Step S32: Calculate the proportion p of the i-th vibration monitoring value at the j-th moment ij for:

[0076]

[0077] Step S33: Calculate the entropy value E of the i-th vibration monitoring value i for:

[0078]

[0079] Step S34: The weight w of the i-th vibration monitoring value is i for:

[0080]

[0081] In this embodiment, the unit degradation index is determined based on the Euclidean distance characterizing the deviation between the standard value and the real-time monitoring value of the multi-source vibration monitoring quantity. The unit degradation index PDI is calculated by the following formula:

[0082]

[0083] Among them, V i (t) represents the monitoring value of the ith vibration monitoring characteristic quantity at time t, F i (t) represents the standard value of the i-th vibration monitoring characteristic quantity at time t, N is the number of multi-source vibration monitoring characteristic quantities, and w i (t) represents the weight of the i-th vibration monitoring value at time t, It should be noted that, in addition to the Euclidean distance, the preset similarity calculation method can also use other calculation methods, such as Mahalanobis distance and Manhattan distance, by simply bringing the standard value and real-time monitoring value of the multi-source vibration monitoring quantity into the corresponding formula for solution. The preset similarity calculation method of this embodiment uses the Euclidean distance only as an example, and can be adaptively adjusted according to actual application requirements.

[0084] Step S4: extract multiple unit degradation indicators based on a preset time period, arrange them in chronological order to obtain a unit historical degradation degree sequence, and use the unit historical degradation degree sequence to train a preset second machine learning model to obtain a unit degradation trend prediction model for predicting the future degradation trend of the unit.

[0085] In this embodiment, the historical degradation degree sequence of the unit is input into the CNN-GRU model for training, and the trained model is obtained as the unit degradation trend prediction model. It should be noted that the preset second machine learning model includes: SVR model, NARX model, LSTM model, CNN model and CNN-GRU model, which is only used as an example and is not limited to this, and is determined according to actual application.

[0086] In a specific embodiment, the operating status monitoring data of the 2 vibration units of a pumped storage power station from May 31, 2008 to December 15, 2011 are obtained through an online monitoring system, including: operating condition parameter data such as working water head, active power, unit speed and guide vane opening, and vibration monitoring data of key components such as upper guide bearing, upper frame, stator, lower guide bearing, lower frame, water guide bearing and top cover. The preset processing algorithm uses the DBSCAN clustering algorithm to clean the above operating status monitoring data, and then trains the BP neural network model and the CNN-GRU model to realize the prediction of the deterioration trend of the pumped storage unit under variable operating conditions, such as Figure 3 As shown, the prediction methods include:

[0087] 1. Construct the unit health data set. The monitoring data of the unit in a healthy state from May 31, 2008 to December 24, 2009 were selected to construct a health data set containing 4,000 monitoring samples. The health data set covers the three operating conditions of pumped storage units: power generation, pumping and no-load. The unit working head varies between 315m and 345m, the unit speed is mainly concentrated at 330r / min, 380r / min and 450r / min, the active power is mainly concentrated at 150MW, 200MW and 250MW, and the guide vane opening varies between 0% and 80%.

[0088] 2. Construct a standard health model for the unit. Based on the constructed health data set, analyze the impact of the corresponding operating parameters on the vibration monitoring quantity of each component, and determine the main operating parameters of the unit. The main operating parameters of the unit include: working head, active power, unit speed and guide vane opening, which are used as inputs of the standard health model of the unit. Figure 4 Taking the Y-axis vibration monitoring of the water guide bearing as an example, the changes in the vibration amplitude of the parameters under different operating conditions are analyzed by three-dimensional visualization. As can be seen from the figure, the active power, working water head, unit speed and guide vane opening of the unit operating condition parameters have a great influence on the changes in the Y-axis vibration of the water guide, and there is a certain mapping relationship. Just as an example, other main operating condition parameters are determined by the same analysis using the above three-dimensional visualization.

[0089] Furthermore, correlation analysis was performed on different vibration monitoring characteristic quantities. A total of 17 vibration monitoring quantities of 7 key components of the unit, including the upper guide bearing, lower guide bearing, water guide bearing, upper frame, lower frame, stator, and top cover, were selected for correlation analysis. Figure 5 The correlation confusion matrix of 17 vibration monitoring quantities is displayed, analyzed and the Pearson correlation coefficient is calculated. The results show that the upper guide bearing X-direction vibration and Y-direction vibration have significant correlation, with a correlation coefficient of 0.96; the lower guide bearing X-direction vibration and Y-direction vibration have significant correlation, with a correlation coefficient of 0.99; the water guide bearing X-direction vibration and Y-direction vibration have significant correlation, with a correlation coefficient of 0.99; the upper frame X-direction vibration and Y-direction vibration have significant correlation, with a correlation coefficient of 0.93; the lower frame X-direction vibration and Y-direction vibration have significant correlation, with a correlation coefficient of 0.93. Based on the correlation analysis results, the Y-direction vibration of the six key components of the upper guide bearing, lower guide bearing, water guide bearing, upper frame, lower frame and top cover is determined as the output of the standard health model of the unit according to the correlation coefficient greater than 0.9.

[0090] Standard health models are established for the Y-axis vibration of the six key components, namely the upper guide bearing, lower guide bearing, water guide bearing, upper frame, lower frame and top cover, and the MAE, MAPE and RMSE fitting error evaluation indicators and correlation coefficient R of the model are calculated. The results are shown in Table 2. As can be seen from the table, the R values ​​of the six standard health models are all greater than 0.96, indicating the effectiveness of the standard health model constructed based on the BP neural network, which provides model support and data basis for the subsequent calculation of degradation indexes.

[0091] Table 2 Results of standard health model evaluation indicators

[0092]

[0093] 3. Determine the degradation index. Obtain the operating condition parameter data and vibration monitoring data of the unit from June 22, 2010 to December 15, 2011, a total of 542 days, through the online monitoring system. Input the operating condition parameters into the above six standard health models respectively, obtain the standard values ​​of the six vibration monitoring quantities, substitute the multi-source vibration standard values ​​and measured values ​​into the formula for determining the unit degradation index PDI based on Euclidean distance, obtain the degradation index PDI of the corresponding unit, and arrange them in chronological order to obtain the PDI sequence, such as Figure 6As shown. As can be seen from the figure, the dotted line in the figure is an auxiliary line, which intuitively represents the overall trend of PDI (upward trend), that is, from June 22, 2010 to October 20, 2011, the PDI of the unit rose steadily with a small amplitude, indicating that the overall operation state of the unit was stable and the performance was good; after October 20, 2011, the PDI of the unit rose by a large margin, indicating that the performance of the unit degraded rapidly, and it was necessary to strengthen vigilance and promptly investigate the abnormal causes to avoid the occurrence of unit failures. In summary, the degradation index of this embodiment can well characterize the characteristics of the gradual deterioration of the unit performance over time.

[0094] 4. Degradation trend prediction. After obtaining the PDI sequence of the unit, a CNN-GRU prediction model is built. The structure of the CNN-GRU prediction model includes: input layer, convolution layer, GRU layer, 2 fully connected layers and output layer. Among them, the drop rates of the two fully connected layers are set to 0.5 and 0.2 respectively, and the structural parameters of other layers are shown in Table 3. In this embodiment, the initial value of the learning rate of the CNN-GRU prediction model is set to 0.001, the batch size is set to 32, the loss function uses the mean square error, the optimization algorithm uses the Adam algorithm, and the maximum number of iterations is set to 300. After building the model and setting the hyperparameters, the Bayesian algorithm is used to train and optimize the learning rate and batch size. The model training and use process is specifically as follows: the historical PDI sequence data of the unit is processed into the input and output data matrix of the prediction model, the prediction time window is set to L, the data points from the i-th to the i+L-1th are taken as L data points to form an input data, and the L+1th data point is taken to form an output data, and the input and output matrices are constructed. After the CNN-GRU prediction model is trained, it is used for PDI prediction at future moments. In this embodiment, the prediction time window is set to 10, the historical PDI sequence data of the unit is processed into the input and output data matrix of the prediction model, the data set is divided according to a ratio of 4:1, and the training set data is sequentially input into the built CNN-GRU prediction model in chronological order to train the model. Substitute the test set data into the trained CNN-GRU prediction model to obtain the corresponding prediction results.

[0095] Table 3 CNN-GRU model structure parameters

[0096]

[0097]

[0098] In order to further prove the effectiveness of the CNN-GRU prediction model, common prediction models such as SVR, NARX, LSTM, and CNN were selected for comparative analysis with the CNN-GRU prediction model. Under the same conditions, the CNN-GRU prediction model and four comparative models were used to repeat the test 10 times, and the quantitative evaluation indicators such as RMSE, MAE, MAPE, and R of the model on the test set were calculated, and the average value of the 10 tests was taken for comparative analysis. The results are shown in Table 4. As can be seen from the table, the average values ​​of RMSE, MAE, and MAPE of the CNN-GRU prediction model of the embodiment of the present invention are 2.25, 1.81, and 0.17, respectively, which are the lowest among all comparative models; the correlation coefficient R value is 0.91, the highest among all comparative models, which is 10% higher than the classic NARX model, which fully demonstrates the effectiveness of the CNN-GRU prediction model and can effectively predict the future degradation trend of the unit.

[0099] Table 4 Evaluation index results of different prediction models

[0100]

[0101] The method for predicting the degradation trend of a pumped-storage unit under varying operating conditions provided in an embodiment of the present invention improves the quality of on-site monitoring data of the unit through data cleaning; at the same time, it fully considers the influence of operating condition changes of the pumped-storage unit on the vibration characteristics and the correlation between multi-source vibration quantities, and can comprehensively and effectively characterize the health status level of the unit; it can effectively predict the future degradation trend of the unit, and provide technical support for ensuring the safe and stable operation of the unit, realizing active maintenance of the unit, and avoiding the occurrence of potential faults.

[0102] Example 2

[0103] The embodiment of the present invention provides a device for predicting the degradation trend of a pumped storage unit under variable operating conditions, such as Figure 7 As shown, including:

[0104] The unit health data set construction module is used to construct a health data set after cleaning and processing the historical stable operation monitoring data of the pumped storage unit based on a preset processing algorithm. The health data set contains the multi-source vibration monitoring feature quantities of different operating condition parameters of the unit and key components; this module executes the method described in step S1 in Example 1, which will not be repeated here.

[0105] The unit standard health model construction module is used to analyze the impact of operating condition parameters on the vibration of key components to determine the main operating parameters of the unit, and analyze the correlation between different vibration monitoring feature quantities; based on the correlation analysis results and the main operating parameters of the unit, the health data set is screened to obtain a health training set; based on the health training set, a preset first machine learning model is trained to obtain a standard health model of the unit; this module executes the method described in step S2 of Example 1, which will not be repeated here.

[0106] A unit degradation index determination module is used to obtain the real-time monitoring value of the multi-source vibration monitoring characteristic quantity of the pumped-storage unit and the corresponding operating condition parameters, input the above operating condition parameters into the unit standard health model to obtain the standard value of the corresponding unit multi-source vibration monitoring characteristic quantity, determine the weight of each vibration monitoring quantity in the unit multi-source vibration monitoring characteristic quantity by the entropy weight method, and determine the unit degradation index based on the standard value and real-time monitoring value of the multi-source vibration monitoring quantity based on a preset similarity calculation method; this module executes the method described in step S3 in Example 1, which will not be repeated here.

[0107] A degradation trend prediction module is used to extract multiple unit degradation indicators based on a preset time period, arrange them in chronological order to obtain a unit historical degradation degree sequence, and use the unit historical degradation degree sequence to train a preset second machine learning model to obtain a unit degradation trend prediction model for predicting the future degradation trend of the unit; this module executes the method described in step S4 in Example 1, which will not be repeated here.

[0108] The device for predicting the degradation trend of a pumped-storage unit under varying operating conditions provided by an embodiment of the present invention can improve the quality of on-site monitoring data of the unit; fully consider the influence of operating condition changes of the pumped-storage unit on the vibration characteristics and the correlation between multi-source vibration quantities, and can more comprehensively and effectively characterize the health status level of the unit; at the same time, by integrating the strong feature extraction capability of CNN and the strong learning and memory capability of GRU, a CNN-GRU degradation trend prediction model is built and trained, which can effectively predict the future degradation trend of the unit.

[0109] Example 3

[0110] An embodiment of the present invention provides a computer device, such as Figure 8As shown, it includes: at least one processor 801, at least one communication interface 803, a memory 804 and at least one communication bus 802. Among them, the communication bus 802 is used to realize the connection and communication between these components, the communication interface 803 may include a display screen and a keyboard, and the optional communication interface 803 may also include a standard wired interface and a wireless interface. The memory 804 may be a high-speed volatile random access memory, or a non-volatile memory, or at least one storage device located away from the aforementioned processor 801. Among them, the processor 801 can execute the variable operating condition degradation trend prediction method of the pumped storage unit in Example 1. A set of program codes are stored in the memory 804, and the processor 801 calls the program code stored in the memory 804 to execute the variable operating condition degradation trend prediction method of the pumped storage unit in Example 1.

[0111] The communication bus 802 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The communication bus 802 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 The fact that only one line is used in the diagram does not mean that there is only one bus or only one type of bus.

[0112] Among them, the memory 804 may include a volatile memory (Volatile Memory), such as a random access memory (Random Access Memory, referred to as RAM); the memory may also include a non-volatile memory (Non-volatile Memory), such as a flash memory (Flash Memory), a hard disk drive (Hard Disk Drive, referred to as HDD) or a solid-state drive (SSD); the memory 804 may also include a combination of the above types of memory.

[0113] The processor 801 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and a NP.

[0114] The processor 801 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD may be a complex programmable logic device (CPLD), a field programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0115] Optionally, the memory 804 is also used to store program instructions. The processor 801 can call the program instructions to implement the method for predicting the degradation trend of the pumped storage unit under variable operating conditions in the first embodiment of the present invention.

[0116] The embodiment of the present invention also provides a computer-readable storage medium, on which computer-executable instructions are stored, and the computer-executable instructions can execute the variable operating condition degradation trend prediction method of the pumped storage unit in embodiment 1. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk (HDD), or a solid state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.

[0117] Obviously, the above embodiments are only examples for clear explanation, and are not intended to limit the implementation methods. For ordinary technicians in the relevant field, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from this are still within the protection scope of the invention.

Claims

1. A method for predicting the degradation trend of a pumped storage unit under variable operating conditions, characterized in that: include: Based on a preset processing algorithm, a health data set is constructed after cleaning and processing the historical stable operation monitoring data of the pumped storage unit, wherein the health data set includes multi-source vibration monitoring feature quantities of different operating condition parameters of the unit and key components, wherein the key components at least include an upper guide bearing, a lower guide bearing, a water guide bearing, an upper frame, a lower frame, a stator and a top cover of the pumped storage unit; The influence of the operating condition parameters on the vibration of key components is analyzed to determine the main operating parameters of the unit and the correlation between different vibration monitoring feature quantities, wherein the main operating parameters of the unit include: working water head, active power, unit speed and guide vane opening; the vibration monitoring feature quantities include: vibration data of the upper guide bearing, lower guide bearing, water guide bearing, upper frame, lower frame, stator and top cover; based on the correlation analysis results and the main operating parameters of the unit, the health data set is screened to obtain a health training set; based on the health training set, a preset first machine learning model is trained to obtain a standard health model of the unit; through the standard health model of the unit, a set F(t) of mapping relationships between the main operating parameters of the unit and multi-source vibration monitoring feature quantities is obtained, and F(t) is expressed as: F(t)={f0(H(t),P(t),N(t),G(t)),f1(H(t),P(t),N(t),G(t)),…,f i (H(t),P(t),N(t),G(t))} Where i is the number of multi-source vibration monitoring quantities, f i () represents the mapping relationship between the i-th vibration monitoring characteristic quantity and the main operating parameters of the unit at time t, H(t) represents the working water head at time t, P(t) represents the active power at time t, N(t) represents the unit speed at time t, and G(t) represents the guide vane opening at time t; The real-time monitoring value and corresponding operating condition parameters of the multi-source vibration monitoring characteristic quantity of the pumped storage unit are obtained, and the corresponding operating condition parameters are input into the standard health model of the unit to obtain the standard value of the multi-source vibration monitoring characteristic quantity of the corresponding unit, and the weight of each vibration monitoring quantity in the multi-source vibration monitoring characteristic quantity of the unit is determined by the entropy weight method, and the standard value and the real-time monitoring value of the multi-source vibration monitoring quantity are determined based on a preset similarity calculation method. The process of determining the weight of each vibration monitoring quantity in the multi-source vibration monitoring characteristic quantity of the unit by the entropy weight method includes: The maximum and minimum normalization method is used to convert the multi-source vibration monitoring characteristic data V within a preset time period into ij Perform normalization: Wherein, i is the i-th vibration monitoring quantity in the multi-source vibration monitoring quantity, N is the number of multi-source vibration monitoring characteristic quantities, j is the monitoring time, and T is the monitoring duration; Calculate the proportion p of the i-th vibration monitoring quantity at the j-th moment ij for: Calculate the entropy value E of the i-th vibration monitoring quantity i for: Then the weight w of the i-th vibration monitoring quantity is i for: The process of determining the unit degradation index based on the preset similarity calculation method for the standard value and the real-time monitoring value of the multi-source vibration monitoring quantity includes: characterizing the deviation degree between the standard value and the real-time monitoring value of the multi-source vibration monitoring quantity based on the Euclidean distance, and determining the unit degradation index, wherein the unit degradation index PDI is calculated by the following formula: Among them, V i (t) represents the monitoring value of the ith vibration monitoring characteristic quantity at time t, F i (t) represents the standard value of the i-th vibration monitoring characteristic quantity at time t, N is the number of multi-source vibration monitoring characteristic quantities, and w i (t) represents the weight of the i-th vibration monitoring value at time t, Based on a preset time period, multiple unit degradation indexes are extracted and arranged in chronological order to obtain a unit historical degradation degree sequence. The unit historical degradation degree sequence is used to train the preset second machine learning model to obtain a unit degradation trend prediction model, which is used to predict the future degradation trend of the unit.

2. The method for predicting the degradation trend of a pumped storage unit under variable operating conditions according to claim 1, characterized in that: The process of obtaining the unit degradation trend prediction model includes: inputting the unit historical degradation degree sequence into the CNN-GRU model for training, and obtaining the trained model as the unit degradation trend prediction model.

3. A device for predicting the degradation trend of a pumped storage unit under variable operating conditions, characterized in that: include: A unit health data set construction module, used to construct a health data set after cleaning and processing the historical stable operation monitoring data of the pumped storage unit based on a preset processing algorithm, wherein the health data set includes multi-source vibration monitoring feature quantities of different operating condition parameters of the unit and key components, wherein the key components at least include the upper guide bearing, lower guide bearing, water guide bearing, upper frame, lower frame, stator and top cover of the pumped storage unit; The unit standard health model construction module is used to analyze the influence of the operating condition parameters on the vibration of key components to determine the main operating condition parameters of the unit, and analyze the correlation between different vibration monitoring feature quantities, wherein the main operating condition parameters of the unit include: working water head, active power, unit speed and guide vane opening; the vibration monitoring feature quantity includes: vibration data of the upper guide bearing, lower guide bearing, water guide bearing, upper frame, lower frame, stator and top cover; based on the correlation analysis results and the main operating condition parameters of the unit, the health data set is screened to obtain a health training set; based on the health training set, a preset first machine learning model is trained to obtain a unit standard health model; through the unit standard health model, a set F(t) of mapping relationships between the main operating condition parameters of the unit and multi-source vibration monitoring feature quantities is obtained, and F(t) is expressed as: F(t)={f0(H(t),P(t),N(t),G(t)),f1(H(t),P(t),N(t),G(t)),…,f i (H(t),P(t),N(t),G(t))} Where i is the number of multi-source vibration monitoring quantities, f i () represents the mapping relationship between the i-th vibration monitoring characteristic quantity and the main operating parameters of the unit at time t, H(t) represents the working water head at time t, P(t) represents the active power at time t, N(t) represents the unit speed at time t, and G(t) represents the guide vane opening at time t; The unit degradation index determination module is used to obtain the real-time monitoring value and corresponding operating condition parameters of the multi-source vibration monitoring characteristic quantity of the pumped storage unit, input the above operating condition parameters into the unit standard health model to obtain the standard value of the corresponding unit multi-source vibration monitoring characteristic quantity, determine the weight of each vibration monitoring quantity in the unit multi-source vibration monitoring characteristic quantity by the entropy weight method, and determine the unit degradation index based on the standard value and the real-time monitoring value of the multi-source vibration monitoring quantity based on a preset similarity calculation method, wherein the process of determining the weight of each vibration monitoring quantity in the unit multi-source vibration monitoring characteristic quantity by the entropy weight method includes: The maximum and minimum normalization method is used to convert the multi-source vibration monitoring characteristic data V within a preset time period into ij Perform normalization: Wherein, i is the i-th vibration monitoring quantity in the multi-source vibration monitoring quantity, N is the number of multi-source vibration monitoring characteristic quantities, j is the monitoring time, and T is the monitoring duration; Calculate the proportion p of the i-th vibration monitoring quantity at the j-th moment ij for: Calculate the entropy value E of the i-th vibration monitoring quantity i for: Then the weight w of the i-th vibration monitoring quantity is i for: The process of determining the unit degradation index based on the preset similarity calculation method for the standard value and the real-time monitoring value of the multi-source vibration monitoring quantity includes: characterizing the deviation degree between the standard value and the real-time monitoring value of the multi-source vibration monitoring quantity based on the Euclidean distance, and determining the unit degradation index, wherein the unit degradation index PDI is calculated by the following formula: Among them, V i (t) represents the monitoring value of the ith vibration monitoring characteristic quantity at time t, F i (t) represents the standard value of the i-th vibration monitoring characteristic quantity at time t, N is the number of multi-source vibration monitoring characteristic quantities, and w i (t) represents the weight of the i-th vibration monitoring value at time t, The degradation trend prediction module is used to extract multiple unit degradation indicators based on a preset time period, arrange them in chronological order to obtain a unit historical degradation degree sequence, and use the unit historical degradation degree sequence to train a preset second machine learning model to obtain a unit degradation trend prediction model, which is used to predict the future degradation trend of the unit.

4. A computer device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the method for predicting the degradation trend of a pumped-storage unit under variable operating conditions as described in any one of claims 1-2.

5. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the method for predicting the degradation trend of a pumped-storage unit under variable operating conditions as described in any one of claims 1-2.