Hydroelectric generating set degradation index construction method and system based on sparse auto-encoder and extreme gradient lifting

Through sparse autoencoder and extreme gradient enhancement algorithm, multi-dimensional characteristics of vibration signals of hydropower units are extracted and reduced, healthy models are established, and deterioration indicators are calculated, and the problem of difficulty in extracting fault features in the existing technology is solved, and the accuracy of hydropower units is achieved Fault prediction and early warning improve the reliability and operational efficiency of the equipment.

CN119089167AActive Publication Date: 2024-12-06CHINA YANGTZE POWER

Patent Information

Application Number
CN202411016670.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-29
Publication Date
2024-12-06
Estimated Expiration
2044-07-29

AI Technical Summary

Technical Problem

The prior art is difficult to accurately extract and identify fault characteristics in the vibration signals of the hydroelectric unit, resulting in low fault prediction and maintenance efficiency, affecting the reliability and operating efficiency of the unit.

Method used

The sparse autoencoder and extreme gradient enhancement algorithm are used to construct a multi-dimensional feature set through the time domain, frequency domain and energy feature extraction of vibration signals, and the sparse autoencoder is used to reduce the feature dimensions, and the unit vibration characteristics are established in combination with the extreme gradient enhancement algorithm. Health model, computer group deterioration indicators to achieve fault feature extraction.

Benefits of technology

It realizes accurate extraction and early warning of fault characteristics of hydropower units, improves the reliability and operating efficiency of the unit, and ensures the safe and stable operation of the equipment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089167A_ABST
    Figure CN119089167A_ABST
Patent Text Reader

Abstract

The invention discloses a hydroelectric generating set degradation index construction method and system based on a sparse auto-encoder and extreme gradient lifting, and the method comprises the following steps: S1, constructing a vibration signal multi-dimensional feature set according to the vibration data of a hydroelectric generating set through employing a vibration signal time domain, frequency domain and energy feature extraction method; s2, using the hydroelectric generating set vibration signal multi-dimensional feature sample to train a sparse auto-encoder, and obtaining a trained sparse auto-encoder model as a set multi-dimensional feature dimension reduction model; s3, collecting vibration monitoring data of the hydroelectric generating set in a normal state to form a health data set, and establishing a set vibration characteristic health model; and S4, calculating the degradation index of the unit. According to the method, the vibration signal characteristics of the hydroelectric generating set are taken as entry points, the vibration characteristics of the set are refined, the real-time degradation index of the set is constructed, the degradation trend of the set is reflected, and the method has important significance on state maintenance and fault early warning of the hydroelectric generating set and reduction of fault loss of the hydroelectric generating set.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of hydropower unit fault feature extraction, and in particular to a method and system for constructing a degradation index of a hydropower unit based on a sparse autoencoder and extreme gradient boosting. Background Art

[0002] Hydropower generating units are the core key equipment of hydropower stations. The scale and capacity of hydropower generating units are constantly increasing, and the structure is more complex. Their health status directly affects the safe and stable operation of the entire hydropower station. As the service life increases, hydropower generating units will have failure accidents of varying degrees. In order to avoid safety accidents, it is very important to conduct in-depth research on fault diagnosis and status prediction of hydropower generating units.

[0003] As an important monitoring indicator for evaluating the operating status of the unit, vibration signals can reflect the internal operation and fault information of the unit. By analyzing and processing vibration signals, equipment abnormalities can be discovered in a timely manner, potential faults can be predicted, downtime can be reduced, the availability and operating efficiency of the unit can be improved, and the safe and stable operation of the unit can be ensured. By extracting features from vibration signals and constructing degradation indicators, feature information can be highlighted, providing a good data foundation for subsequent signal research, helping engineering and technical personnel to more accurately judge the status of the unit, take corresponding maintenance and repair measures in a timely manner, and improve the reliability and operating efficiency of the unit. Therefore, it is crucial for the normal operation and maintenance of hydropower units to accurately and reliably extract and identify fault features in vibration signals and construct degradation indicators that can reflect the real-time status of the unit. Summary of the invention

[0004] The main purpose of the present invention is to provide a method and system for constructing a degradation index of a hydropower unit based on sparse autoencoder and extreme gradient boosting, and to use the sparse autoencoder model (SAE) and extreme gradient boosting (XGBoost) in the field of machine learning to construct a degradation index of a hydropower unit to reflect the degradation trend of the unit, so as to find a more effective new method in the field of fault feature extraction of hydropower units.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for constructing a degradation index of a hydropower unit based on a sparse autoencoder and extreme gradient boosting, comprising the following steps: S1. Based on the vibration data of the hydropower unit, a vibration signal time domain, frequency domain, and energy feature extraction method is used to construct a multi-dimensional feature set of the vibration signal; S2. Use the multi-dimensional feature samples of the vibration signal of the hydropower unit to train the sparse autoencoder, and obtain the trained sparse autoencoder model as the multi-dimensional feature dimensionality reduction model of the unit; S3. Collect vibration monitoring data of hydropower units under normal conditions to form a healthy data set, use normal data to train the extreme gradient boosting algorithm, find the mapping relationship between each vibration dimension reduction feature and the operating data of the hydropower unit, and establish a healthy model of unit vibration characteristics; S4. Calculate the degradation index of the unit, use S3 to obtain the deviation between the real-time vibration dimension reduction feature of the unit and the corresponding healthy feature, and integrate the deviations of each vibration dimension reduction feature as the degradation index of the vibration signal of the hydropower unit.

[0006] In S1, the time domain, frequency domain and energy characteristics of the vibration signal are extracted, specifically including: S11. Extract target signal The six characteristics of time domain mean, standard deviation, peak-to-peak value, root mean square, kurtosis, and skewness are as follows: ; (1) ; (2) ; (3) ; (4) ; (5) ; (6) in: represents the signal sequence, N is the number of sampling points, is the time domain mean, is the standard deviation, is the peak-to-peak value, is the root mean square, is the kurtosis, is the skewness; S12. Extract the target signal spectrum The seven indicators of mean frequency, spectrum skewness, spectrum kurtosis, spectrum first-order center of gravity, spectrum second-order center of gravity, spectrum second-order central moment, and standard deviation frequency are used as frequency characteristics, as follows: ; (7) ; (8) ; (9) ; (10) ; (11) ; (12) ; (13) in, Indicates the amplitude of each frequency corresponding to the signal spectrum, Represents the frequency of each point on the spectrum, N Indicates the number of sampling points; represents the mean frequency, represents the spectral skewness, represents the spectral kurtosis, represents the first-order center of gravity of the spectrum, represents the second-order center of gravity of the spectrum, represents the second-order central moment of the spectrum, represents the standard deviation frequency; S13. Extracting the energy entropy of the target signal as its energy feature; ; (14) in, E represents the signal energy, Represents the energy entropy of the target signal.

[0007] In S2, the multi-dimensional feature samples of the unit vibration signal are used to train the sparse autoencoder. By adding a sparsity limit to the hidden layer unit z in the autoencoder, the autoencoder can learn the key structure in the data. The steps are as follows: S21. Assuming the number of training samples is N, the optimization goal of the sparse autoencoder is: ; (15) in, Represents the vector composed of all training sample sequences at the input end, is the sparsity measurement function, W is the hyperparameter of the sparse autoencoder, represents the optimization goal, represents the nth input sample sequence, represents the output sample sequence obtained by decoding the sparse autoencoder, represents the weight coefficient of the sparsity constraint, Represents the weight attenuation coefficient, which is used to reduce the weight amplitude and prevent overfitting; S22. Given N After training samples, the hidden layer j The mean activity of a neuron is: ; (16) In the formula, represents the activation probability of the jth neuron in the hidden layer, so that Close to a given value ; Indicates the nth group of sample input The output value of the jth neuron in the hidden layer is Using relative entropy as a measure of probability differences, it is calculated as: ; (17) S23. When Equal to a given value hour, . Therefore, the sparsity metric function is defined as: ; (18) In the formula, p represents the number of neurons in the hidden layer, express KL Divergence is used to limit and The difference between represents a constant coefficient.

[0008] In S3, the operating data of the hydropower unit includes active power P , guide vane opening , Blade opening , Unit speed N , and water head H .

[0009] The steps to establish the unit vibration characteristic health model are as follows: S31. Given a data set A containing samples and The specific calculation method of the extreme gradient boosting algorithm is as follows: ; (19) ; (20) ;(twenty one) In the formula, K represents the number of regression trees, Indicates The predicted value of samples, where the predicted value is accumulated by each tree, B is the solution space of regression tree, represents the i-th sample, represents the i-th feature, Represents the overall structure function of the CART tree. The input sample can be obtained by Mapped to leaf nodes, Represents a tree model, Represents the output value of the leaf node, represents the structure function of the tree, represents a sample set, Represents a collection of trees, Indicates the number of trees; S32. Given Represents the objective function of the extreme gradient boosting algorithm, and the objective function calculation method is as follows: ;(twenty two) ;(twenty three) In the formula, Represents the loss function, which is used to describe the degree of deviation between the true value and the predicted value. represents the regularization term of the function, represents the objective function of the algorithm, Indicates the difficulty of partitioning leaf nodes. represents the regularization coefficient, Indicates the score of each leaf node; S33. Loss function Use Taylor's formula to expand to the second order and simplify the objective function. The calculation formula is as follows: ;(twenty four) ; (25) In the formula, represents the sum of the linear gradients of the regression tree leaf nodes, Represents the sum of the quadratic gradients of the regression tree leaf nodes; In order to minimize the objective function, the derivative of the objective function is set to zero for solution, and the results are as follows: ; (26) ; (27) In the formula, represents the standard value of weight, represents the weight of j leaf nodes, Indicates the difficulty of leaf node partitioning, represents the penalty coefficient, represents the number of trees, represents the partial derivative of the loss function, represents the partial derivative of the predicted value; S34. Input the real-time operating data of the hydropower unit into the trained extreme gradient boosting algorithm to obtain the real-time health value of the vibration characteristics as follows: ; (28) ; (29) In the formula , is the two-dimensional feature after dimensionality reduction of multi-dimensional features, Indicates active power, Indicates the guide vane opening, Indicates the blade opening, Indicates the unit speed. Indicates water head.

[0010] In S4, the steps of calculating the degradation index of the computer group are as follows: S41. Calculate the deviations between the two-dimensional vibration characteristics after dimension reduction and the unit health standard as follows: ; (30) ; (31) In the formula, , Indicates the symptom index corresponding to the vibration characteristic component, , represents the actual value of the feature, , represents the two-dimensional feature health value, , Represents the two-dimensional feature health value; S42. The corresponding indicators of the vibration characteristic components are fused according to the weights to obtain the degradation indicator of the target signal. The steps are as follows: ; (32) In the formula, Indicates the vibration signal degradation index, , represents the weight coefficient, In physical terms, it indicates the degree of deviation of vibration characteristics from the healthy value; The unit vibration characteristics obtained by multi-dimensional feature reduction contain local vibration characteristic information. Different characteristics have different sensitivity to faults and may even change in opposite directions. Therefore, in order to avoid mutual cancellation between characteristics, it is stipulated that and Keep the notation consistent, that is: ; (33) When calculating the fusion degradation index of a single component, a weight is introduced, and the calculation method is as follows: ; (34) ; (35) In the formula, , represents the weight coefficient, represents the health value of feature one, represents the health value of feature 2, , is the actual value of the feature.

[0011] S43. After the vibration signal degradation index is constructed, the fault warning threshold is further set according to the average value of the degradation index of the normal state of the hydropower unit to define the normal state and fault warning state of the hydropower unit. The specific calculation method is as follows: ; (36) In the formula, is the fault warning threshold, is the median of the overall warning index of the normal state of the hydropower unit, and 𝛿 is the amplification factor; Among them, when the degradation index exceeds the health threshold, it means that the unit may have an early failure; otherwise, it means that the equipment is in normal condition.

[0012] A system using the method for constructing a hydropower unit degradation index based on sparse autoencoder and extreme gradient boosting comprises: Unit vibration signal data acquisition module: The unit vibration signal data acquisition module is used to obtain the vibration history data of the unit components under various working conditions to form a unit vibration signal sample set, each sample containing the vibration waveform data of key components; Multidimensional feature extraction module: The multidimensional feature extraction module is used to extract multidimensional features of the target signal and construct a multidimensional feature set of the vibration signal by using the vibration signal time domain, frequency domain and energy feature extraction methods; Sparse autoencoder feature dimensionality reduction module: The sparse autoencoder feature dimensionality reduction module is used to input the training samples of each test point into the sparse autoencoder, train the sparse autoencoder and use the trained sparse autoencoder model as the feature dimensionality reduction model, input the sample to be tested into the dimensionality reduction model, and obtain the dimensionality reduction features of each sample; Extreme gradient boosting health model construction module: The extreme gradient boosting health model construction module is used to select the normal operating condition data of the unit under different operating conditions and the vibration characteristics of the corresponding components as the input of the extreme gradient boosting algorithm for training, and input the real-time operating condition of the unit into the trained extreme gradient boosting algorithm to obtain the healthy vibration characteristics of the unit under the operating condition; Degradation index and its health threshold calculation module: The degradation index and its health threshold calculation module uses the deviation between each vibration real-time feature of the unit and the corresponding health feature, integrates each vibration feature deviation as the degradation index of the vibration signal of the hydropower unit, and selects the product of the median of the degradation index and the amplification factor in the healthy state as the fault threshold; Fault warning module: The fault warning module is used to calculate the sample degradation index. When the sample degradation index exceeds the warning threshold, the fault warning action is executed; otherwise, it means that the device is in a normal state, and the degradation index at the next moment is continued to be calculated.

[0013] The present invention has the following beneficial effects: The present invention proposes a method and system for constructing vibration degradation indicators of hydropower units based on sparse autoencoder (SAE) and extreme gradient boosting (XGBoost). Multidimensional features are extracted from the three aspects of vibration signal time domain, frequency domain and energy, and SAE is used to reduce the dimension of multidimensional features to obtain key features. XGBoost is used to find the mapping relationship between the operating data of the hydropower unit and the key features to build a healthy standard model, and the deviation between the real-time features and the standard features in the health model is used as the unit degradation indicator, and the fault threshold of the unit degradation indicator is determined to achieve early fault warning of the hydropower unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0015] Figure 1 A flow chart of a method for constructing a degradation index of a hydropower unit based on a sparse autoencoder and extreme gradient boosting is provided for the implementation of the present invention.

[0016] Figure 2 A diagram of the sparse autoencoder network architecture provided for the implementation of the present invention.

[0017] Figure 3 A flowchart of degradation index calculation provided for the implementation of the present invention.

[0018] FIG4 is a diagram showing the system architecture for constructing degradation indicators of hydropower units based on sparse autoencoders and extreme gradient boosting according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] Embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, but should not be construed as limiting the present invention.

[0020] Embodiment 1: A method for constructing a degradation index of a hydropower unit based on a sparse autoencoder and extreme gradient boosting is proposed in an embodiment of the present invention, referring to Figure 1-3 , the specific implementation steps are as follows: Step 1: Based on the vibration data of the hydropower unit, the vibration signal time domain, frequency domain, and energy feature extraction methods are used to construct a multi-dimensional feature set of the vibration signal. The construction process is as follows; (1) Extracting time domain features includes: Mean ; Standard Deviation ; Peak-to-Peak ; RMS ; Kurtosis ; Skewness ; in, represents the signal sequence, N is the number of sampling points.

[0021] (2) Extracting frequency domain features includes: Mean Frequency ; Standard Deviation Frequency ; Frequency skewness ; Spectral Kurtosis ; First-order center of mass of spectrum ; Second-order center of mass of spectrum ; Spectral second-order central moment ; in Indicates the amplitude of each frequency corresponding to the signal spectrum, Represents the frequency of each point on the spectrum, N is the number of sampling points.

[0022] (3) Extracting energy characteristics and energy entropy ; in, E represents the signal energy, Represents the energy entropy of the target signal.

[0023] Step 2: Use some samples of the multi-dimensional features of the vibration signal of the hydropower unit to train the sparse autoencoder (SAE) to obtain the trained sparse autoencoder (SAE) model as the multi-dimensional feature dimensionality reduction model of the unit.

[0024] Step 21: Assuming the number of training samples is N, the optimization goal of the sparse autoencoder is: ; in, Represents the vector composed of all training sample sequences at the input end, is the sparsity measurement function, W is the hyperparameter of the sparse autoencoder, represents the optimization goal, represents the nth input sample sequence, represents the output sample sequence obtained by decoding the sparse autoencoder, represents the weight coefficient of the sparsity constraint, Represents the weight attenuation coefficient, which is used to reduce the weight amplitude and prevent overfitting.

[0025] Step 22: Given N After training samples, the hidden layer j The mean activity of a neuron is: ; In the formula, represents the activation probability of the jth neuron in the hidden layer, so that Close to a given value ; Represents the nth group of sample input The output value of the jth neuron in the hidden layer is , and the relative entropy (Kullback-Leibe) divergence is used as a measure of the probability difference, which is calculated as follows: .

[0026] Step 23: When Equal to a given value hour, . Therefore, the sparsity metric function is defined as: ; In the formula, p is the number of neurons in the hidden layer, express KL Divergence is used to limit and The difference between represents a constant coefficient.

[0027] Step 3: Collect vibration monitoring data of hydropower units under normal conditions to form a healthy data set. Use normal data to train extreme gradient boosting, and use the trained extreme gradient boosting algorithm to find the vibration features and active power extracted in step 2. P , guide vane opening , Blade opening , Unit speed N , water head H The mapping relationship between them is used to establish the unit vibration characteristic health model.

[0028] Step 31: Given a data set A containing samples and Features.

[0029] The specific calculation method of the extreme gradient boosting algorithm (XGBoost) is: ,; ; ; In the formula,K represents the number of regression trees, Indicates The predicted value of samples, where the predicted value is accumulated by each tree, B is the solution space of regression tree, represents the i-th sample, represents the i-th feature, Represents the overall structure function of the CART tree. The input sample can be obtained by Mapped to leaf nodes, Represents a tree model, Represents the output value of the leaf node, represents the structure function of the tree, represents a sample set, Represents a collection of trees, Indicates the number of trees.

[0030] Step 32: Given Represents the objective function of the extreme gradient boosting algorithm (XGBoost), and the objective function calculation method is obtained as: ; ; in, Represents the loss function, which is used to describe the degree of deviation between the true value and the predicted value. represents the regularization term of the function, represents the objective function of the algorithm, Indicates the difficulty of partitioning leaf nodes. represents the regularization coefficient, Represents the score of each leaf node.

[0031] Step 33: Convert the loss function Use Taylor's formula to expand to the second order and simplify the objective function. The calculation formula is: ; ; in, represents the sum of the linear gradients of the regression tree leaf nodes, Represents the sum of the quadratic terms of the regression tree leaf nodes. In order to minimize the objective function, the derivative of the objective function is set to zero for solution, and the result is: , ; In the formula, represents the standard value of weight, represents the weight of j leaf nodes, Indicates the difficulty of leaf node partitioning, represents the penalty coefficient, represents the number of trees, represents the partial derivative of the loss function, represents the partial derivative of the predicted value.

[0032] Step 34: Input the real-time operating data of the hydropower unit into the trained extreme gradient boosting algorithm (XGBoost) to obtain the real-time health value of the vibration feature: ; ; In the formula , is the two-dimensional feature after dimensionality reduction of multi-dimensional features, Indicates active power, Indicates the guide vane opening, Indicates the blade opening, Indicates the unit speed. Indicates water head.

[0033] Step 4: Using the deviation between the real-time vibration characteristics of the unit obtained in step 3 and the corresponding healthy characteristics, the deviation of each vibration characteristic is integrated as the vibration signal degradation indicator of the hydropower unit.

[0034] Step 41: Calculate the deviation between the two-dimensional vibration characteristics after dimension reduction and the unit health standard: , ; in, , is the symptom index corresponding to the vibration characteristic component, , is the actual value of the feature, , represents the two-dimensional feature health value, , Represents two-dimensional feature health.

[0035] Step 42: Fusion the corresponding indices of the vibration characteristic components according to the weights to obtain the degradation index of the target signal: ; in, is the vibration signal degradation indicator, , is the weight coefficient. In physical terms, it indicates the degree of deviation of vibration characteristics from healthy values.

[0036] The unit vibration characteristics obtained by multi-dimensional feature reduction contain local vibration characteristic information. Different characteristics have different sensitivity to faults and may even change in opposite directions. Therefore, in order to avoid mutual cancellation between characteristics, it is stipulated that and Keep the notation consistent, that is: .

[0037] Step 43: In order to enhance the sensitivity of the degradation index to abnormal data, weights are introduced in the calculation of the fusion degradation index of a single component to more specifically consider the importance of different features in reflecting the change of the unit state. Such weight selection can improve the ability to identify potential fault features. The calculation method is: ; ; In the formula, , represents the weight coefficient, represents the health value of feature one, represents the health value of feature 2, , is the actual value of the feature.

[0038] Step 5: After completing the construction of the vibration signal degradation index, the fault warning threshold is further set according to the average value of the degradation index of the normal state of the hydropower unit to define the normal state and fault warning state of the hydropower unit. The specific calculation method is: .

[0039] in, is the median of the overall warning index of the normal state of the hydropower unit, and 𝛿 is the amplification factor. Take 𝛿=1.6. When the degradation index exceeds the health threshold, it means that the unit may have an early failure, and the on-site staff needs to conduct a preliminary inspection; otherwise, it means that the equipment is in normal condition.

[0040] Embodiment 2: See also Figure 4 The embodiment of the present invention proposes a system for constructing degradation indicators of hydropower units based on sparse autoencoders and extreme gradient boosting, the system comprising: The unit vibration signal data acquisition module relies on the real-time monitoring system of the power station to obtain the historical vibration data of the unit components under various working conditions and form a sample set of unit vibration signals. (N is the number of samples, M is the number of monitoring variables contained in the sample), each sample It should include the vibration waveform data of key components, such as the swing waveform data of each guide bearing.

[0041] Multidimensional feature extraction module. Extracts multidimensional features of target signals. Uses vibration signal time domain, frequency domain, and energy feature extraction methods to construct a multidimensional feature set of vibration signals.

[0042] Sparse Autoencoder (SAE) feature dimensionality reduction module. Input the training samples of each test point into SAE, train SAE according to the process of power 2 and use the trained SAE model as the feature dimensionality reduction model, input the samples to be tested into the dimensionality reduction model, and calculate the dimensionality reduction features of each sample. ; Extreme Gradient Boosting (XGBoost) health model building module. The normal operating data of the unit under different operating conditions and the vibration characteristics of the corresponding components are selected as the input of the extreme gradient boosting algorithm for training, and the real-time operating conditions of the unit are input into the trained extreme gradient boosting algorithm to obtain the healthy vibration characteristics of the unit under this condition.

[0043] Degradation index and its health threshold calculation module. The deviation between the real-time vibration characteristics of the unit and the corresponding health characteristics is used to integrate the deviation of each vibration characteristic as the degradation index of the vibration signal of the hydropower unit. The product of the median of the degradation index and the amplification factor in the healthy state is selected as the fault threshold.

[0044] Fault warning module. For the sample to be tested, the sample degradation index is first calculated. When the sample degradation index exceeds the warning threshold, it means that the equipment has early signs of failure and the fault warning action is executed; otherwise, it means that the equipment is in a normal state, and the degradation index at the next moment is calculated.

[0045] It will be easily understood by those skilled in the art that the above descriptions are merely specific implementation steps of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for constructing degradation index of hydropower units based on sparse autoencoder and extreme gradient boosting, characterized in that: The following steps are involved: S1. Based on the vibration data of the hydropower unit, a vibration signal time domain, frequency domain, and energy feature extraction method is used to construct a multi-dimensional feature set of the vibration signal; S2. Use the multi-dimensional feature samples of the vibration signal of the hydropower unit to train the sparse autoencoder, and obtain the trained sparse autoencoder model as the multi-dimensional feature dimensionality reduction model of the unit; S3. Collect vibration monitoring data of hydropower units under normal conditions to form a healthy data set, use normal data to train the extreme gradient boosting algorithm, find the mapping relationship between each vibration dimension reduction feature and the operating data of the hydropower unit, and establish a healthy model of unit vibration characteristics; S4. Calculate the degradation index of the unit, use S3 to obtain the deviation between the real-time vibration dimension reduction feature of the unit and the corresponding healthy feature, and integrate the deviations of each vibration dimension reduction feature as the degradation index of the vibration signal of the hydropower unit.

2. According to claim 1, a method for constructing degradation index of hydropower units based on sparse autoencoder and extreme gradient boosting is characterized in that: In S1, the time domain, frequency domain and energy characteristics of the vibration signal are extracted, specifically including: S11. Extract target signal The six characteristics of time domain mean, standard deviation, peak-to-peak value, root mean square, kurtosis, and skewness are as follows: ;(1) ;(2) ;(3) ;(4) ;(5) ;(6) in: represents the signal sequence, N is the number of sampling points, is the time domain mean, is the standard deviation, is the peak-to-peak value, is the root mean square, is the kurtosis, is the skewness; S12. Extract the target signal spectrum The seven indicators of mean frequency, spectrum skewness, spectrum kurtosis, spectrum first-order center of gravity, spectrum second-order center of gravity, spectrum second-order central moment, and standard deviation frequency are used as frequency characteristics, as follows: ;(7) ;(8) ;(9) ;(10) ;(11) ;(12) ;(13) in, Indicates the amplitude of each frequency corresponding to the signal spectrum, Represents the frequency of each point on the spectrum, N Indicates the number of sampling points; represents the mean frequency, represents the spectral skewness, represents the spectral kurtosis, represents the first-order center of gravity of the spectrum, represents the second-order center of gravity of the spectrum, represents the second-order central moment of the spectrum, represents the standard deviation frequency; S13. Extracting the energy entropy of the target signal as its energy feature; ;(14) in, E represents the signal energy, Represents the energy entropy of the target signal.

3. According to claim 1, a method for constructing degradation index of hydropower units based on sparse autoencoder and extreme gradient boosting is characterized in that: In S2, the multi-dimensional feature samples of the unit vibration signal are used to train the sparse autoencoder. By adding a sparsity limit to the hidden layer unit z in the autoencoder, the autoencoder can learn the key structure in the data. The steps are as follows: S21. Assuming the number of training samples is N, the optimization goal of the sparse autoencoder is: ; (15) in, Represents the vector composed of all training sample sequences at the input end, is the sparsity measurement function, W is the hyperparameter of the sparse autoencoder, represents the optimization goal, represents the nth input sample sequence, represents the output sample sequence obtained by decoding the sparse autoencoder, represents the weight coefficient of the sparsity constraint, Represents the weight attenuation coefficient, which is used to reduce the weight amplitude and prevent overfitting; S22. Given N After training samples, the hidden layer j The mean activity of a neuron is: ;(16) In the formula, represents the activation probability of the jth neuron in the hidden layer, so that Close to a given value ; Represents the nth group of sample input The output value of the jth neuron in the hidden layer is Using relative entropy as a measure of probability differences, it is calculated as: ; (17) S23. When Equal to a given value hour, . Therefore, the sparsity metric function is defined as: ;(18) In the formula, p represents the number of neurons in the hidden layer, express KL Divergence is used to limit and The difference between represents a constant coefficient.

4. According to claim 1, a method for constructing degradation index of hydropower units based on sparse autoencoder and extreme gradient boosting is characterized in that: In S3, the operating data of the hydropower unit includes active power P , guide vane opening , Blade opening , Unit speed N , and water head H .

5. The method for constructing a degradation index of a hydropower unit based on sparse autoencoder and extreme gradient boosting according to claim 4 is characterized in that: The steps to establish the unit vibration characteristic health model are as follows: S31. Given a data set A containing samples and The specific calculation method of the extreme gradient boosting algorithm is as follows: ;(19) ;(20) ; (21) In the formula, K represents the number of regression trees, Indicates The predicted value of samples, where the predicted value is accumulated by each tree, B is the solution space of regression tree, represents the i-th sample, represents the i-th feature, Represents the overall structure function of the CART tree. The input sample can be obtained by Mapped to leaf nodes, Represents a tree model, Represents the output value of the leaf node, represents the structure function of the tree, represents a sample set, Represents a collection of trees, Indicates the number of trees; S32. Given Represents the objective function of the extreme gradient boosting algorithm, and the objective function calculation method is as follows: ;(22) ; (23) In the formula, Represents the loss function, which is used to describe the degree of deviation between the true value and the predicted value. represents the regularization term of the function, represents the objective function of the algorithm, Indicates the difficulty of partitioning leaf nodes. represents the regularization coefficient, Indicates the score of each leaf node; S33. Loss function Use Taylor's formula to expand to the second order and simplify the objective function. The calculation formula is as follows: ;(24) ;(25) In the formula, represents the sum of the linear gradients of the regression tree leaf nodes, Represents the sum of the quadratic gradients of the regression tree leaf nodes; In order to minimize the objective function, the derivative of the objective function is set to zero for solution, and the results are as follows: ;(26) ;(27) In the formula, represents the standard value of weight, represents the weight of j leaf nodes, Indicates the difficulty of leaf node partitioning, represents the penalty coefficient, represents the number of trees, represents the partial derivative of the loss function, represents the partial derivative of the predicted value; S34. Input the real-time operating data of the hydropower unit into the trained extreme gradient boosting algorithm to obtain the real-time health value of the vibration characteristics as follows: ;(28) ; (29) In the formula , is the two-dimensional feature after dimensionality reduction of multi-dimensional features, Indicates active power, Indicates the guide vane opening, Indicates the blade opening, Indicates the unit speed. Indicates water head.

6. The method for constructing degradation index of hydropower units based on sparse autoencoder and extreme gradient boosting according to claim 1 is characterized in that: In S4, the steps of calculating the degradation index of the computer group are as follows: S41. Calculate the deviations between the two-dimensional vibration characteristics after dimension reduction and the unit health standard as follows: ;(30) ; (31) In the formula, , Indicates the symptom index corresponding to the vibration characteristic component, , represents the actual value of the feature, , represents the two-dimensional feature health value, , Represents the two-dimensional feature health value; S42. The corresponding indicators of the vibration characteristic components are fused according to the weights to obtain the degradation indicator of the target signal. The steps are as follows: ; (32) In the formula, Indicates the vibration signal degradation index, , represents the weight coefficient, In physical terms, it indicates the degree of deviation of vibration characteristics from the healthy value; The unit vibration characteristics obtained by multi-dimensional feature reduction contain local vibration characteristic information. Different characteristics have different sensitivity to faults and may even change in opposite directions. Therefore, in order to avoid mutual cancellation between characteristics, it is stipulated that and Keep the notation consistent, that is: ;(33) When calculating the fusion degradation index of a single component, a weight is introduced, and the calculation method is as follows: ;(34) ;(35) In the formula, , represents the weight coefficient, represents the health value of feature one, represents the health value of feature 2, , is the actual value of the feature; S43. After the vibration signal degradation index is constructed, the fault warning threshold is further set according to the average value of the degradation index of the normal state of the hydropower unit to define the normal state and fault warning state of the hydropower unit. The specific calculation method is as follows: ; (36) In the formula, is the fault warning threshold, is the median of the overall warning index of the normal state of the hydropower unit, and 𝛿 is the amplification factor; Among them, when the degradation index exceeds the health threshold, it means that the unit may have an early failure; otherwise, it means that the equipment is in normal condition.

7. A system using a method for constructing a degradation index of a hydropower unit based on a sparse autoencoder and extreme gradient boosting as described in any one of claims 1 to 6, characterized in that: The system includes; Unit vibration signal data acquisition module: The unit vibration signal data acquisition module is used to obtain the vibration history data of the unit components under various working conditions to form a unit vibration signal sample set, each sample containing the vibration waveform data of key components; Multidimensional feature extraction module: The multidimensional feature extraction module is used to extract multidimensional features of the target signal and construct a multidimensional feature set of the vibration signal by using the vibration signal time domain, frequency domain and energy feature extraction methods; Sparse autoencoder feature dimensionality reduction module: The sparse autoencoder feature dimensionality reduction module is used to input the training samples of each test point into the sparse autoencoder, train the sparse autoencoder and use the trained sparse autoencoder model as the feature dimensionality reduction model, input the sample to be tested into the dimensionality reduction model, and obtain the dimensionality reduction features of each sample; Extreme gradient boosting health model construction module: The extreme gradient boosting health model construction module is used to select the normal operating condition data of the unit under different operating conditions and the vibration characteristics of the corresponding components as the input of the extreme gradient boosting algorithm for training, and input the real-time operating condition of the unit into the trained extreme gradient boosting algorithm to obtain the healthy vibration characteristics of the unit under the operating condition; Degradation index and its health threshold calculation module: The degradation index and its health threshold calculation module uses the deviation between each vibration real-time feature of the unit and the corresponding health feature, integrates each vibration feature deviation as the degradation index of the vibration signal of the hydropower unit, and selects the product of the median of the degradation index and the amplification factor in the healthy state as the fault threshold; Fault warning module: The fault warning module is used to calculate the sample degradation index, and when the sample degradation index exceeds the warning threshold, the fault warning action is executed; Otherwise, it means that the equipment is in normal state, and the degradation index at the next moment is calculated.

Citation Information

Patent Citations

  • Model training method and device based on kernel auto-encoder and storage medium

    CN111860552A

  • Hydroelectric generating set fault early warning method, device and terminal

    CN115952404A

  • Sparse self-coding fault diagnosis method and system based on information fusion

    CN116337449A

  • Industrial equipment health state identification method based on unbalanced data

    CN116502146A

  • Method and system for training a big data machine to defend

    US20170169360A1

Cited By

  • Method and system for monitoring state of key component of hydroelectric generating set

    CN121637030A