Method for determining long-term service performance evolution law of hydroelectric generating set
Through stacking integrated learning models and vibration parameter models, the problem of difficult to predict the downward trend of equipment performance of hydropower units is solved, and the accurate evaluation and prediction of the performance evolution laws of hydropower units is achieved, and the reliability and safety of equipment are improved.
Patent Information
- Application Number
- CN202410902974.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-08
- Publication Date
- 2024-11-01
AI Technical Summary
Existing hydropower unit fault diagnosis technology cannot effectively evaluate and predict the downward trend of equipment performance, especially due to the complexity of vibration parameters and the lack of sufficient fault samples, which makes it difficult to achieve state-based maintenance.
Using a stacked integrated learning model, combining support vector machine, random tree and adaptive neural network fuzzy inference system, train the vibration parameter model in the healthy state of the unit. By obtaining operating condition data, predicting vibration parameters, and calculating the stationary component of the deviation sequence. Logarithmic energy value, reconstruction performance evolution law, and evaluation of equipment performance.
Improve prediction accuracy, accurately describe the operating status of the unit, remove the volatility of the original time series, and obtain a relatively accurate performance evolution trend, which helps to timely detect equipment abnormalities and improve reliability and safety.
Smart Images

Figure CN118886533A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of equipment fault detection, and in particular to a method for determining the long-term service performance evolution law of a hydropower unit. Background Art
[0002] As the operating time increases, the performance of hydropower unit equipment gradually declines. Determining the actual working status of the unit and accurately predicting its degradation trend in order to detect anomalies and arrange maintenance in time is crucial to achieving the transition from planned maintenance to condition-based maintenance.
[0003] At present, the fault diagnosis technology of hydropower units has not yet met the actual needs. The main challenges are the complexity of the faults and the lack of sufficient fault samples. However, with the continuous improvement of the hydropower station status monitoring system, the amount of monitoring data in the healthy state is increasing. Therefore, it is particularly important to effectively use the existing monitoring data and extract valuable information.
[0004] Among them, vibration is one of the main factors affecting the normal operation and service life of hydropower units. Excessive vibration may cause mutual collision of unit components, accelerated wear, fatigue damage, and reduce unit efficiency. In extreme cases, it may even cause catastrophic accidents. Therefore, the performance degradation trend of hydropower units can be predicted by evaluating the vibration parameters of the units.
[0005] Therefore, there is an urgent need for a method to determine the long-term service performance evolution law of hydropower units to effectively evaluate and predict the performance degradation trend of hydropower units. Summary of the invention
[0006] The purpose of the present invention is to provide a method for determining the long-term service performance evolution law of a hydropower unit, which can effectively evaluate and predict the performance degradation trend of the hydropower unit.
[0007] To achieve the above object, the present invention provides the following solutions:
[0008] A method for determining the long-term service performance evolution law of a hydropower unit, the method comprising:
[0009] Acquire the operating condition data of the current hydropower unit; the operating condition data includes active power and working water head;
[0010] The operating condition data is input into the vibration parameter model of the unit in a healthy state to obtain the predicted value of the unit vibration parameter in a healthy state; the vibration parameter model of the unit in a healthy state is obtained by training the stacked ensemble learning model using a training set; the training set includes training samples and corresponding labels; the training samples are the historical operating condition data of the hydropower unit in a healthy state; the labels are the historical unit vibration parameters of the hydropower unit in a healthy state;
[0011] Determining a vibration parameter deviation value of the unit according to the predicted vibration parameter value of the unit in the healthy state and the actual measured vibration parameter value of the unit;
[0012] Acquire multiple unit vibration parameter deviation values, and obtain a unit vibration parameter deviation sequence according to the multiple unit vibration parameter deviation values;
[0013] Decomposing the vibration parameter deviation sequence of the unit to obtain a plurality of stable components of the vibration parameters of the unit;
[0014] Calculate the logarithmic energy value of the stationary component of the vibration parameters of each unit;
[0015] Determine whether the logarithmic energy of the stable component of each vibration parameter of the unit is less than a preset logarithmic energy threshold, and if not, discard the corresponding stable component of the vibration parameter of the unit;
[0016] If so, the corresponding stable component of the unit vibration parameter is reconstructed, and the reconstructed component of the unit vibration parameter is used as the prediction result of the long-term service performance evolution law of the hydropower unit, and the equipment performance of the hydropower unit is evaluated and predicted based on the prediction result.
[0017] Optionally, the stacked ensemble learning model includes a first base learner, a second base learner and a meta learner; the first base learner and the second base learner are connected in parallel and then connected in series with the meta learner.
[0018] Optionally, the first base learner is a support vector machine base learner; the second base learner is a random tree base learner; and the meta-learner is an adaptive neural network fuzzy inference system meta-learner.
[0019] Optionally, the training process of the vibration parameter model of the unit in a healthy state specifically includes:
[0020] The training sample (X tr , Y tr ) are respectively input into the first basis learner and the second basis learner, and the first unit vibration parameter prediction value and the second unit vibration parameter prediction value are output; wherein the training set (X, Y) is divided into training samples (X tr , Y tr ) and test samples (X te , Y te ); X = {x i ,i=1,2,3,…,n},x i ={active power, working head}, x i represents the historical operating condition data corresponding to the i-th sample; Y = {y i,i=1,2,3,…,n},y i represents the vibration parameter value of the unit corresponding to the i-th sample; (X tr , Y tr )={(x i ,y i ), i = 1, 2, 3, ..., nk} means selecting nk samples as training samples; (X te , Y te )={(x i ,y i ), i=1, 2, 3, ..., k}, indicating that k samples are selected as test samples;
[0021] The first unit vibration parameter prediction value and the second unit vibration parameter prediction value are used as the first fusion training sample (P tr , Y tr ), input to the meta-learner, and output the predicted value of the vibration parameter of the third unit;
[0022] The parameters of the first base learner and the second base learner are adjusted according to the error between the first unit vibration parameter prediction value and the second unit vibration parameter prediction value and the corresponding label, and the parameters of the meta learner are adjusted according to the error between the third unit vibration parameter prediction value and the corresponding label to obtain the vibration parameter model of the unit in a healthy state.
[0023] Optionally, after obtaining the vibration parameter model of the unit in a healthy state, the method further includes: testing the vibration parameter model of the unit in a healthy state using a test set, specifically including:
[0024] The test set (X te , Y te ) are respectively input into the first basis learner and the second basis learner of the vibration parameter model of the unit in a healthy state to obtain the vibration parameter prediction value of the fourth unit and the vibration parameter prediction value of the fifth unit; the test set includes test samples and corresponding labels; the test samples are the historical operating condition data of the test water-generating unit in a healthy state;
[0025] The predicted values of the vibration parameters of the fourth unit and the fifth unit are used as the second fusion training samples (P te , Y te ), input into the meta-learner of the vibration parameter model of the unit in a healthy state, and output the predicted value of the vibration parameter of the sixth unit;
[0026] The prediction effect of the vibration parameter model of the unit in a healthy state is evaluated according to the predicted value of the vibration parameter of the sixth unit.
[0027] Optionally, the method for decomposing the vibration parameter deviation sequence of the unit is empirical Fourier decomposition.
[0028] Optionally, performing empirical Fourier decomposition on the vibration parameter deviation sequence of the unit specifically includes:
[0029] The vibration parameter deviation sequence of the unit is subjected to Fourier transform to obtain the Fourier spectrum of the vibration parameter deviation sequence of the unit; the calculation formula of Fourier transform is as follows:
[0030]
[0031] Where, f(t) is the representation of the unit vibration parameter deviation sequence in the time domain; ω is the frequency; j is the imaginary unit; t is the time; is the Fourier spectrum of the unit vibration parameter deviation sequence;
[0032] Based on the Fourier spectrum of the vibration parameter deviation sequence of the unit, the boundary of each segment is determined by using an improved segmentation technique; the expression of the boundary of each segment is as follows:
[0033]
[0034] Among them, ω n is the boundary of the segment; Represents Ω n and Ω n+1 The Fourier spectrum amplitude between; N represents the total number of frequency segments obtained by segmentation; n is the number of frequency segments; 1≤n≤N;
[0035] Based on the boundary ω of each segment n , construct a filter bank; the expression of the filter bank is as follows:
[0036]
[0037] in, represents a filter bank;
[0038] Based on the filter bank, a filtered frequency domain signal is obtained; the expression of the filtered frequency domain signal is as follows:
[0039]
[0040] Perform inverse Fourier transform on the filtered frequency domain signal to obtain multiple unit vibration parameter steady components of the unit vibration parameter deviation sequence; the expression of the unit vibration parameter steady component is as follows:
[0041]
[0042] Among them, f n (t) is the stable component of the unit vibration parameter.
[0043] Optionally, the logarithmic energy value of the stationary component of the vibration parameter of each unit is calculated, specifically including:
[0044] Based on the stationary component of the vibration parameters of each unit, a plurality of data points are determined;
[0045] Set the initial energy value to 0 and calculate the cumulative energy value of multiple data points;
[0046] Based on the accumulated energy value, the logarithmic energy value of the stable component of the vibration parameter of each unit is obtained.
[0047] Optionally, the expression for calculating the accumulated energy value of multiple data points is:
[0048] Y t =Y t-1 +((f n (t)^2)-f n (t-1)*f n (t+1));
[0049] Among them, Y t represents the accumulated energy value of the tth data point; Y t-1 represents the accumulated energy value of the t-1th data point; t represents the sequence number of the data point; f n (t) represents the stable component of the unit vibration parameters.
[0050] Optionally, the calculation formula of the logarithmic energy value of the stable component of the unit vibration parameter is as follows:
[0051] h q =log(Y t );
[0052] Among them, h q It represents the logarithmic energy value of the stable component of the unit vibration parameter, and q represents the serial number of the stable component of the unit vibration parameter.
[0053] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0054] The present invention constructs a vibration parameter model of the unit in a healthy state based on a stacked ensemble learning model, which can fully mine the original data set and comprehensively utilize the learning advantages of each base learner to improve the prediction accuracy and accurately describe the unit operation status; the unit vibration parameter deviation sequence is decomposed to obtain the stable component of the unit vibration parameter, which has better regularity than the original sequence and can better reflect the characteristics of the performance degradation trend; in addition, the stable component of the unit vibration parameter is reconstructed according to the preset logarithmic energy threshold, which effectively removes the volatility of the original time series and can obtain a more accurate stable trend of the unit performance evolution. The present invention can effectively evaluate and predict the performance degradation trend of the hydropower unit, which helps to timely discover the abnormal conditions of the equipment, thereby improving the reliability and safety of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0056] Figure 1 A schematic flow chart of a method for determining the long-term service performance evolution law of a hydropower unit provided in Example 1 of the present invention;
[0057] Figure 2 This is a schematic diagram of the evolution trend of the long-term service performance of pumped storage units;
[0058] Figure 3 This is a schematic diagram of the decomposition results of the evolution trend of pumped storage units;
[0059] Figure 4 This is a schematic diagram of the results of extracting the evolution law of the unit's long-term service performance. DETAILED DESCRIPTION
[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0061] The purpose of the present invention is to provide a method for determining the long-term service performance evolution law of a hydropower unit, which can effectively evaluate and predict the performance degradation trend of the hydropower unit.
[0062] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0063] Example 1
[0064] like Figure 1 As shown, the method for determining the long-term service performance evolution law of a hydropower unit in this embodiment includes the following steps:
[0065] Step S1: obtaining the operating condition data of the current hydropower unit; the operating condition data includes active power and working water head.
[0066] Step S2: Input the operating condition data into the vibration parameter model of the unit in a healthy state to obtain the predicted value of the unit vibration parameter in a healthy state; the vibration parameter model of the unit in a healthy state is obtained by training the stacked ensemble learning model using a training set; the training set includes training samples and corresponding labels; the training samples are historical operating condition data of the hydropower unit in a healthy state; the labels are historical unit vibration parameters of the hydropower unit in a healthy state.
[0067] Specifically, by deeply analyzing the massive state monitoring data under different operating conditions at the initial stage of unit operation, the operating condition data at the initial stage of unit operation and in the healthy state of the unit are selected as training samples. In addition, by comprehensively analyzing a large amount of field data, it is found that active power and working water head are the two main factors affecting the operating state of the hydropower unit; at the same time, sensitive characteristic parameters that can reflect the operating state of the unit are selected. The present invention selects the vibration parameters of the unit as sensitive characteristic parameters, and establishes a standard health model of the unit based on stacked ensemble learning that comprehensively considers information such as active power and working water head, and obtains the vibration parameter model f(P, H, v) of the unit in a healthy state, where v is the vibration parameter of the unit, P is the active power, and H is the working water head. This modeling method can more realistically reflect the operating condition factors (active power, working water head) that affect the state of the unit, thereby effectively utilizing the existing normal massive data and accurately describing the state of the unit.
[0068] Furthermore, the stacked ensemble learning model includes a first base learner, a second base learner and a meta-learner; the first base learner and the second base learner are connected in parallel and then connected in series with the meta-learner. The first base learner is a support vector machine base learner; the second base learner is a random tree base learner; and the meta-learner is an adaptive neural network fuzzy inference system meta-learner. The stacked ensemble learning method trains multiple base learners and uses their prediction results as inputs to the meta-learner to find the optimal combination of base learners, so as to fully mine the original data set and comprehensively utilize the learning advantages of each base learner, thereby improving the prediction accuracy.
[0069] Furthermore, the training process of the vibration parameter model of the unit in a healthy state specifically includes:
[0070] Step S21: inputting the training samples into the first basis learner and the second basis learner respectively, and outputting the predicted value of the vibration parameter of the first unit and the predicted value of the vibration parameter of the second unit.
[0071] Step S22: using the first unit vibration parameter prediction value and the second unit vibration parameter prediction value as fusion training samples, inputting them into the meta-learner, and outputting the third unit vibration parameter prediction value.
[0072] Step S23: Adjust the parameters of the first base learner and the second base learner according to the error between the first unit vibration parameter prediction value and the second unit vibration parameter prediction value and the corresponding label, and adjust the parameters of the meta learner according to the error between the third unit vibration parameter prediction value and the corresponding label to obtain the vibration parameter model of the unit in a healthy state.
[0073] Specifically, when obtaining training samples, the training set (X, Y) is divided into training samples (X tr , Y tr ) and test samples (X te , Y te ); X = {x i , i = 1, 2, 3, ..., n}, where x i ={active power, working head}, x i represents the historical operating condition data corresponding to the i-th sample; Y = {y i ,i=1,2,3,…,n}, where, y i represents the response value corresponding to the i-th sample, i.e., the vibration parameter value of the unit; (X tr , Y tr )={(x i ,y i ), i = 1, 2, 3, ..., nk}, indicating that nk samples are selected as training samples; (X te , Y te )={(x i ,y i ), i=1, 2, 3, ..., k}, indicating that k samples are selected as test samples.
[0074] During the training process, when the first base learner and the second base learner are trained, the prediction results of each base learner for the entire original training set are obtained, that is, the prediction value of the vibration parameter of the first unit and the prediction value of the vibration parameter of the second unit: P tr = {P tr-支持向量机 , P tr-随机树Then, the prediction results of each base learner and the corresponding original response value are constructed into the first fusion training sample (P tr , Y tr ), and used as a training sample of the meta-learner, which is input into the meta-learner to obtain the predicted value of the vibration parameter of the third unit.
[0075] During the test, each trained base learner is used to obtain the prediction value P corresponding to the original test sample. te , namely the vibration parameter prediction value of the fourth unit and the vibration parameter prediction value of the fifth unit, and construct the second fusion training sample (P te , Y te ), input into the meta-learner of the vibration parameter model of the unit in a healthy state, and output the predicted value of the vibration parameter of the sixth unit for testing the trained meta-learner, where P te = {P te-支持向量机 , P te-随机树}.
[0076] Step S3: determining the deviation value of the unit vibration parameter according to the predicted value of the unit vibration parameter in the healthy state and the measured value of the unit vibration parameter.
[0077] Step S4: Obtain multiple unit vibration parameter deviation values, and obtain the unit vibration parameter deviation sequence according to the multiple unit vibration parameter deviation values. The unit vibration parameter deviation sequence represents the unit degradation trend. Since the obtained unit vibration parameter deviation sequence has strong volatility and non-stationarity, how to obtain the performance degradation trend of the unit in the subsequent period of time is a key issue in the on-site operation and maintenance of the unit. Therefore, it is necessary to effectively process it in order to obtain a true and reliable stable trend of the unit performance evolution.
[0078] Step S5: Decomposing the vibration parameter deviation sequence of the unit to obtain a plurality of stable components of the vibration parameters of the unit.
[0079] Furthermore, in step S5, the method for decomposing the vibration parameter deviation sequence of the unit is empirical Fourier decomposition. Specifically, the steps of performing empirical Fourier decomposition on the vibration parameter deviation sequence of the unit are as follows:
[0080] Step S51: Perform Fourier transform on the vibration parameter deviation sequence of the unit to obtain the Fourier spectrum of the vibration parameter deviation sequence of the unit. The specific formula is as follows:
[0081]
[0082] Where: f(t) is the representation of the unit vibration parameter deviation sequence in the time domain; ω is the frequency, j is the imaginary unit; t represents time; Fourier spectrum representing the deviation sequence of unit vibration parameters.
[0083] Step S52: Based on the Fourier spectrum of the vibration parameter deviation sequence of the unit, the boundary ω of each segment is determined by using an improved segmentation technique. n .
[0084] In the improved segmentation technique, the frequency range of [0, π] is divided into N frequency segments, all the maximum points in the frequency range are selected and sorted in descending order, and the frequencies corresponding to the first N maximum values in the sorted sequence are represented by [Ω1,Ω2,…Ω N ] indicates. In addition, Ω0=0 and Ω N+1 =π. The boundary of each segment is determined by the following equation (2).
[0085]
[0086] in, Represents Ω n and Ω n+1 The Fourier spectrum amplitude between .
[0087] Step S53: Based on the ω obtained in step S52 n , construct an ideal filter bank Its expression is as follows:
[0088]
[0089] Where, 1≤n≤N, ω n The value of is determined by formula (2).
[0090] Step S54: Based on the result obtained in step S53 Get the filtered frequency domain signal As shown in formula (4):
[0091]
[0092] Step S55: The Perform inverse Fourier transform to obtain multiple decomposition components f of the original signal in the time domain n (t), namely, multiple unit vibration parameter stationary components of the unit vibration parameter deviation sequence.
[0093]
[0094] Step S6: Calculate the logarithmic energy value of the stationary component of the vibration parameter of each unit.
[0095] Specifically, step S6 includes the following steps:
[0096] Step S61: Determine a plurality of data points based on the stationary component of the vibration parameters of each unit.
[0097] Step S62: Set the initial energy value to 0, and calculate the cumulative energy value of multiple data points. The initial energy value Y0 = 0; for t = 2 to N-1 (N is the number of data points), the expression for calculating the cumulative energy value of multiple data points is:
[0098] Y t =Y t-1 +((f n (t)^2)-f n (t-1)*f n (t+1)) (6)
[0099] Among them, Y t represents the accumulated energy value of the tth data point; Y t-1 represents the accumulated energy value of the t-1th data point; t represents the sequence number of the data point; f n (t) represents the stable component of the unit vibration parameter. After cyclic iteration, the accumulated energy value at t = N-1 is obtained.
[0100] Step S63: Based on the accumulated energy value, the logarithmic energy value of the stable component of the vibration parameter of each unit is obtained. The calculation formula is as follows:
[0101] h q =log(Y t ) (7)
[0102] Among them, h q represents the logarithmic energy value of the stable component of the unit vibration parameter, and q represents the qth stable component of the unit vibration parameter in step S61.
[0103] Step S7: Determine whether the logarithmic energy of the stable component of each unit vibration parameter is less than a preset logarithmic energy threshold.
[0104] Step S8: If not, the corresponding stable component of the unit vibration parameter is discarded.
[0105] Step S9: If yes, reconstruct the corresponding stable component of the unit vibration parameter, and use the reconstructed component of the unit vibration parameter as the prediction result of the evolution law of the long-term service performance of the hydropower unit, and evaluate and predict the equipment performance of the hydropower unit based on the prediction result.
[0106] Technical effects of the present invention:
[0107] 1) The present invention constructs a vibration parameter model of the unit in a healthy state based on a stacked ensemble learning model, which can fully mine the original data set and comprehensively utilize the learning advantages of each base learner to improve the prediction accuracy and accurately describe the unit operation status.
[0108] 2) The vibration parameter deviation sequence of the unit is decomposed to obtain the stable component of the vibration parameter of the unit, which has better regularity than the original sequence and can better reflect the characteristics of the performance degradation trend.
[0109] 3) The stable components of the unit vibration parameters are reconstructed according to the preset logarithmic energy threshold, which effectively removes the volatility of the original time series and obtains a more accurate stable trend of the unit performance evolution, thereby effectively evaluating and predicting the performance degradation trend of the hydropower unit, helping to timely discover abnormal conditions of the equipment, thereby improving the reliability and safety of the equipment.
[0110] Example 2
[0111] The above-mentioned embodiment 1 is further described below in conjunction with specific embodiment 2.
[0112] The actual state monitoring data of a hydropower unit in a pumped storage power station from September 22, 2008 to December 15, 2011 were used as samples for research to verify the effectiveness of the present invention. The X-axis horizontal vibration data of the upper frame of the unit was selected as the research object. Through analysis, it can be seen that due to the complex structure of the unit and the frequent switching of working conditions, the time series of the vibration parameters of the unit is very complex, making it difficult to accurately analyze its actual operating status.
[0113] Through analysis, it is found that the pumping / generating conditions, active power, and working water head of the unit have an important influence on the vibration parameters of the unit. Due to the continuous conversion of the working water head and operating conditions of the unit, the vibration changes of the unit are very complex, and the real state of the unit cannot be directly obtained from the vibration data. In order to obtain the real operating state of the pumped-storage unit in real time, it is necessary to establish a healthy state model for it. Considering the important influence of the power and working water head of the pumped-storage unit on the vibration characteristics of the unit, and the good fitting performance of stacked ensemble learning for scattered data, this embodiment is based on the stacked ensemble learning method to establish a three-dimensional surface model of unit vibration parameters-active power-working water head, and obtain the mapping relationship v=f(P,H) between the active power (P), working water head (H) and the vibration parameters (v) of the unit under the healthy state.
[0114] This embodiment uses the data of the initial operation of the unit, which is in good operating condition and has no faults, to establish the vibration parameter model of the unit in a healthy state. Among the 1000 sets of data from September 22, 2008 to October 22, 2009, 800 sets of healthy standard data that can cover the possible working head and active power variation range of the unit are extracted to establish the vibration parameter model of the unit in a healthy state, and the remaining 200 sets of data are used as test samples for model verification. It can be seen from the analysis that the predicted value of the horizontal vibration parameter of the upper frame of the unit in the X direction based on the stacked ensemble learning model is almost completely consistent with the measured value, and the calculated average relative error is 2.70%.
[0115] Substitute the real-time online data of active power, working water head, etc. in the unit status monitoring after 2 years of operation (from May 27, 2011 to December 12, 2011) into the vibration parameter model v(t) = f(P(t), H(t)) of the unit in a healthy state, calculate the predicted value v(t) of the vibration parameter under the current working condition, and compare it with the measured value to obtain the schematic diagram of the evolution trend of the long-term service performance of the pumped storage unit, as shown in the figure. Figure 2 shown.
[0116] from Figure 2 It can be seen that after two years of operation, the vibration parameters of the hydropower unit began to degrade, and the deviation between the measured values of the unit vibration parameters and the healthy values (the predicted values of the vibration parameters under the healthy state of the unit) became larger and larger. Figure 2 It can be further seen that the vibration deviation continues to fluctuate and increase, and from November 20, 2011, it has increased significantly and has exceeded 30%.
[0117] from Figure 2 It can also be seen that the unit degradation trend is highly volatile and non-stationary. How to obtain the performance degradation trend of the unit over a period of time is a key issue in the on-site operation and maintenance of the unit. Therefore, it is necessary to effectively process it in order to obtain a real and reliable stable trend of the unit performance evolution. The empirical Fourier method is used to Figure 2 The unit state parameter degradation time series shown in the figure is decomposed to obtain 10 stationary components of unit vibration parameters, as shown in Figure 3 As shown, it can be seen that the decomposed stable components of the unit vibration parameters have better regularity than the original sequence, and can better reflect the characteristics of the performance degradation trend. According to the change law of each stable component after decomposition and the preset threshold, they are reconstructed to obtain a more accurate stable trend of the unit performance evolution.
[0118] The logarithmic energy of the stationary component of the vibration parameters of each unit is calculated, and the results are shown in Table 1. As can be seen from Table 1, from u1 to u 10The approximate entropy of has a tendency to increase gradually, and the logarithmic energy of u1 is significantly smaller than the other stable components. According to the principle that the preset logarithmic energy threshold is less than 1.2, u1 is used as the final unit vibration parameter reconstruction component to complete the extraction of the evolution law of the long-term copy performance of the hydropower unit, such as Figure 4 As shown. Figure 4 It can be seen that the reconstructed performance degradation trend effectively removes the volatility of the original time series and can well reflect the long-term trend of unit performance degradation changes.
[0119] Table 1 Logarithmic energy values of the stable components of vibration parameters of each unit
[0120] <![CDATA[u1]]> <![CDATA[u2]]> <![CDATA[u3]]> <![CDATA[u4]]> <![CDATA[u5]]> <![CDATA[u6]]> <![CDATA[u7]]> <![CDATA[u8]]> <![CDATA[u9]]> <![CDATA[u 10 ]]> 0.998 1.260 3.007 3.039 3.305 3.084 3.402 3.330 2.505 7.613
[0121] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0122] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of the present invention. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
Claims
1. A method for determining the long-term service performance evolution law of a hydropower unit, characterized in that: The method comprises: Acquire the operating condition data of the current hydropower unit; the operating condition data includes active power and working water head; The operating condition data is input into the vibration parameter model of the unit in a healthy state to obtain the predicted value of the unit vibration parameter in a healthy state; the vibration parameter model of the unit in a healthy state is obtained by training the stacked ensemble learning model using a training set; the training set includes training samples and corresponding labels; the training samples are the historical operating condition data of the hydropower unit in a healthy state; the labels are the historical unit vibration parameters of the hydropower unit in a healthy state; Determining a vibration parameter deviation value of the unit according to the predicted vibration parameter value of the unit in the healthy state and the actual measured vibration parameter value of the unit; Acquire multiple unit vibration parameter deviation values, and obtain a unit vibration parameter deviation sequence according to the multiple unit vibration parameter deviation values; Decomposing the vibration parameter deviation sequence of the unit to obtain a plurality of stable components of the vibration parameters of the unit; Calculate the logarithmic energy value of the stationary component of the vibration parameters of each unit; Determine whether the logarithmic energy of the stable component of each vibration parameter of the unit is less than a preset logarithmic energy threshold, and if not, discard the corresponding stable component of the vibration parameter of the unit; If so, the corresponding stable component of the unit vibration parameter is reconstructed, and the reconstructed component of the unit vibration parameter is used as the prediction result of the long-term service performance evolution law of the hydropower unit, and the equipment performance of the hydropower unit is evaluated and predicted based on the prediction result.
2. The method for determining the long-term service performance evolution law of a hydropower unit according to claim 1, characterized in that: The stacked ensemble learning model includes a first base learner, a second base learner and a meta learner; the first base learner and the second base learner are connected in parallel and then connected in series with the meta learner.
3. The method for determining the long-term service performance evolution law of a hydropower unit according to claim 2, characterized in that: The first base learner is a support vector machine base learner; the second base learner is a random tree base learner; and the meta learner is an adaptive neural network fuzzy inference system meta learner.
4. The method for determining the long-term service performance evolution law of a hydropower unit according to claim 2, characterized in that: The training process of the vibration parameter model of the unit in a healthy state specifically includes: The training sample (X tr , Y tr ) are respectively input into the first basis learner and the second basis learner, and the first unit vibration parameter prediction value and the second unit vibration parameter prediction value are output; wherein the training set (X, Y) is divided into training samples (X tr , Y tr ) and test samples (X te , Y te ); X = {x i ,i=1,2,3,…,n},x i ={active power, working head}, x i represents the historical operating condition data corresponding to the i-th sample; Y = {y i ,i=1,2,3,…,n},y i represents the vibration parameter value of the unit corresponding to the i-th sample; (X tr , Y tr )={(x i ,y i ), i = 1, 2, 3, ..., nk} means selecting nk samples as training samples; (X te , Y te )={(x i ,y i ), i=1, 2, 3, ..., k}, indicating that k samples are selected as test samples; The first unit vibration parameter prediction value and the second unit vibration parameter prediction value are used as the first fusion training sample (P tr , Y tr ), input to the meta-learner, and output the predicted value of the vibration parameter of the third unit; The parameters of the first base learner and the second base learner are adjusted according to the error between the first unit vibration parameter prediction value and the second unit vibration parameter prediction value and the corresponding label, and the parameters of the meta learner are adjusted according to the error between the third unit vibration parameter prediction value and the corresponding label to obtain the vibration parameter model of the unit in a healthy state.
5. The method for determining the long-term service performance evolution law of a hydropower unit according to claim 4, characterized in that: After obtaining the vibration parameter model of the unit in a healthy state, the method further includes: testing the vibration parameter model of the unit in a healthy state using a test set, specifically including: The test set (X te , Y te ) are respectively input into the first basis learner and the second basis learner of the vibration parameter model of the unit in a healthy state to obtain the vibration parameter prediction value of the fourth unit and the vibration parameter prediction value of the fifth unit; the test set includes test samples and corresponding labels; the test samples are the historical operating condition data of the test water-generating unit in a healthy state; The predicted values of the vibration parameters of the fourth unit and the fifth unit are used as the second fusion training samples (P te , Y te ), input into the meta-learner of the vibration parameter model of the unit in a healthy state, and output the predicted value of the vibration parameter of the sixth unit; The prediction effect of the vibration parameter model of the unit in a healthy state is evaluated according to the predicted value of the vibration parameter of the sixth unit.
6. The method for determining the long-term service performance evolution law of a hydropower unit according to claim 1, characterized in that: The method for decomposing the vibration parameter deviation sequence of the unit is empirical Fourier decomposition.
7. The method for determining the evolution law of the long-term service performance of a hydropower unit according to claim 6, characterized in that: Performing empirical Fourier decomposition on the vibration parameter deviation sequence of the unit specifically includes: The vibration parameter deviation sequence of the unit is subjected to Fourier transform to obtain the Fourier spectrum of the vibration parameter deviation sequence of the unit; the calculation formula of Fourier transform is as follows: Where, f(t) is the representation of the unit vibration parameter deviation sequence in the time domain; ω is the frequency; j is the imaginary unit; t is the time; is the Fourier spectrum of the unit vibration parameter deviation sequence; Based on the Fourier spectrum of the vibration parameter deviation sequence of the unit, the boundary of each segment is determined by using an improved segmentation technique; the expression of the boundary of each segment is as follows: Among them, ω n is the boundary of the segment; Represents Ω n and Ω n+1 The Fourier spectrum amplitude between; N represents the total number of frequency segments obtained by segmentation; n is the number of frequency segments; 1≤n≤N; Based on the boundary ω of each segment n , construct a filter bank; the expression of the filter bank is as follows: in, represents a filter bank; Based on the filter bank, a filtered frequency domain signal is obtained; the expression of the filtered frequency domain signal is as follows: Perform inverse Fourier transform on the filtered frequency domain signal to obtain multiple unit vibration parameter steady components of the unit vibration parameter deviation sequence; the expression of the unit vibration parameter steady component is as follows: Among them, f n (t) is the stable component of the unit vibration parameter.
8. The method for determining the long-term service performance evolution law of a hydropower unit according to claim 1, characterized in that: Calculate the logarithmic energy value of the stationary component of the vibration parameters of each unit, including: Based on the stationary component of the vibration parameters of each unit, a plurality of data points are determined; Set the initial energy value to 0 and calculate the cumulative energy value of multiple data points; Based on the accumulated energy value, the logarithmic energy value of the stable component of the vibration parameter of each unit is obtained.
9. The method for determining the long-term service performance evolution law of a hydropower unit according to claim 8, characterized in that: The expression for calculating the accumulated energy value of multiple data points is: Y t =Y t-1 +((f n (t)^2)-f n (t-1)*f n (t+1)); Among them, Y t represents the accumulated energy value of the tth data point; Y t-1 represents the accumulated energy value of the t-1th data point; t represents the sequence number of the data point; f n (t) represents the stable component of the unit vibration parameters.
10. The method for determining the long-term service performance evolution law of a hydropower unit according to claim 9, characterized in that: The calculation formula of the logarithmic energy value of the stable component of the unit vibration parameters is as follows: h q =log(Y t ); Among them, h q It represents the logarithmic energy value of the stable component of the unit vibration parameter, and q represents the serial number of the stable component of the unit vibration parameter.
Citation Information
Cited By
Oil monitoring fault diagnosis method and system based on Internet of Things
CN119691646A
A method and system for fault diagnosis of oil monitoring based on the Internet of Things
CN119691646B
