Lightning arrester state on-line diagnosis method

Abstract

The invention discloses a lightning arrester state on-line diagnosis method, which comprises the following steps: carrying out an impact aging test on a ZnO valve plate of a lightning arrester, and recording initial electrical parameters and each group of electrical parameters after impact until the valve plate is damaged; the change condition between the leakage current under the operation voltage and typical aging parameters in regulations in the whole life cycle process of the ZnO valve plate in the impact aging test is analyzed, correlation analysis is carried out, and the characteristics and labels of a machine learning model are determined; based on a ZnO valve plate test data set, researching the performance of a relation model between online monitoring parameters and typical aging parameters in regulations under different machine learning models, and proposing an optimal relation model; and on the basis of the optimal relation model, leakage current under the operation voltage of the lightning arrester is monitored on line, typical aging parameters in the regulation are further predicted in real time, and intelligent on-line diagnosis of the state of the lightning arrester is realized.

Description

Technical Field

[0001] The invention relates to the technical field of intelligent detection, and in particular to an online diagnosis method for a lightning arrester state. Background Art

[0002] Lightning arresters are key equipment for suppressing overvoltage in key parts of power grids such as transmission and distribution lines, substations, and converter stations. Since they are often in strong electromagnetic environments and are frequently subjected to shocks, online diagnosis of the health status of lightning arresters throughout their life cycle can continuously improve the on-site lightning arrester operation and maintenance efficiency and improve the reliable operation capability of the power grid. At the same time, the use of artificial intelligence to achieve intelligent diagnosis of the status of lightning arresters will also provide basic methodological support for the digital twin of power equipment.

[0003] In the actual operation and maintenance of lightning arresters in my country, the relevant provisions of the power industry standard DL / T596-2016 "Preventive Test Procedures for Power Equipment" are mainly followed, and the DC reference voltage is used. U 1mA Is the rate of change greater than 5%, 0.75? U 1mA DC leakage current I 0.75U1mA Whether it exceeds 50μA, the resistive component of the leakage current under the operating voltage I xr Increase 100% as the failure criterion for arrester status diagnosis. I x In the domestic and foreign research on lightning arrester status diagnosis, specific numerical value / change rate criteria are mainly proposed based on experiments, or a state binary classification model is simply proposed using machine learning.

[0004] The disadvantages of the prior art are: Recommended by existing regulations U 1mA and I 0.75U1mA Only offline testing can be performed, and online monitoring cannot be achieved. However, lightning arresters are often subjected to frequent shocks during thunderstorm seasons and daily switching of circuit breakers. U 1mA , I 0.75U1mA Obviously, it cannot meet the actual needs; and the resistive component of the leakage current under the operating voltage I xr The monitoring requires the installation of additional voltage sensors to monitor the voltage at both ends of the lightning arrester, which will bring huge costs and poor economy.

[0005] Leakage current at operating voltage I xSince the increase is not large during the whole life cycle of the arrester, few countries have proposed specific thresholds for it. The absolute value / change rate threshold obtained from the experiment is relatively simple as the diagnosis criterion of the arrester status. The status classification model proposed in the existing research cannot reflect the gradual degradation process of the arrester and is not convenient for actual on-site operation and maintenance. Summary of the invention

[0006] In order to overcome the above-mentioned shortcomings and deficiencies of the prior art, an object of the present invention is to provide an online diagnosis method for the state of a lightning arrester.

[0007] The purpose of the present invention is achieved through the following technical solutions: A method for online diagnosis of lightning arrester status comprises the following steps: Conduct an impact aging test on the ZnO valve plate of the arrester, and record the electrical parameter values ​​at the beginning and after each set of impact aging tests until the valve plate is damaged; Analysis of the impact aging test, leakage current under operating voltage during the entire life cycle of the ZnO valve I x and its resistive component I xr The changes between the typical aging parameters in the specification and the DC reference voltage are analyzed and the characteristics and labels of the machine learning model are determined. U 1mA and 0.75 U 1mA DC leakage current I 0.75U1mA ; Based on the ZnO valve test data set, the leakage current under online monitoring of operating voltage under different machine learning models is studied I x The performance of the relationship model between the typical aging parameters in the regulations and the optimal relationship model are proposed; Online monitoring of leakage current under arrester operating voltage using the optimal relationship model I x , further predict the typical aging parameters in the regulations in real time and realize intelligent online diagnosis of the arrester status.

[0008] Furthermore, the arrester ZnO valve plate impact aging test is carried out, and the electrical parameter values ​​at the beginning and after each group of impacts are recorded until the valve plate is damaged, specifically: According to the principle of repeated charge transfer in GB / T11032-2020, determine the impact current, number of impacts and interval between adjacent impacts of the impact aging test; After each set of shocks, the electrical parameters are measured after cooling to room temperature. The electrical parameters include the DC reference voltage U1mA , 0.75 U 1mA DC leakage current I 0.75U1mA , leakage current under operating voltage I x and its resistive component I xr , the leakage current under the operating voltage I x and resistive component I xr Perform FFT transformation to obtain the fundamental amplitude of leakage current under operating voltage I x1 , fundamental amplitude of the resistive component of leakage current under operating voltage I xr1 The third harmonic amplitude of leakage current under operating voltage I x3 , the third harmonic amplitude of the resistive component of the leakage current under the operating voltage I xr3 ; The electrical parameter values ​​in multiple impact aging tests were recorded to construct a ZnO valve plate test data set.

[0009] Furthermore, in the analysis of the impact aging test, during the entire life cycle of the ZnO valve sheet, the leakage current under the operating voltage I x Its fundamental amplitude I x1 , third harmonic amplitude I x3 , the resistive component of the leakage current under the operating voltage I xr Its fundamental amplitude I xr1 , third harmonic amplitude I xr3 The changes between the typical aging parameters in the regulations are analyzed, and the characteristics and labels of the machine learning model are determined, specifically: Analysis of DC reference voltage in ZnO valve plate impact aging test U 1mA and 0.75 U 1mA DC leakage current I 0.75U1mA According to the change of the arrester status, the parameters that best represent the arrester status are determined as the labels of the machine learning model; Calculate the selected tags, I x , I x1 , Ix3 , I xr , I xr1 and I xr3 Pearson coefficients between parameters, correlation analysis, verification of leakage current under operating voltage I x and its harmonics I x1 , I x3 The resistive component of the leakage current under the operating voltage I xr and its harmonics I xr1、 I xr3 Equivalence in machine learning models, i.e. validation I x and I xr , I x1 and I xr1 , I x3 and I xr3 Is the Pearson coefficient greater than the threshold of 0.9? If so, further consideration I x , I x1 , I x3 The relative independence between them determines the characteristics of the machine learning model.

[0010] Furthermore, the label selection of the machine learning model is positive and negative. U 1mA .

[0011] Furthermore, the fundamental amplitude of the leakage current under the operating voltage is I x1 , third harmonic amplitude I x3 As features for machine learning models.

[0012] Furthermore, based on the ZnO valve test data set, the leakage current under online monitoring of operating voltage under different machine learning models is studied. I x The performance of the relationship model between the typical aging parameters in the regulations is analyzed, and the optimal relationship model is proposed, which is as follows: The ZnO valve plate test data set is divided into training set, validation set and test set; Construct hyperparameter optimization objective function: R 2 combine1 = 0.6 R 2 train ﹢ 0.4 R 2 validation R 2 combine2 = 0.5 R 2 train ﹢ 0.5 R 2 validation R 2 combine3 = 0.4R 2 train ﹢ 0.6 R 2 validation in R 2 train is the training set fit, R 2 validation is the goodness of fit of the validation set; According to the training set and validation set data, the optimization objective function is used to optimize the hyperparameters using a variety of optimization algorithms; the degree of fit on the validation set is used R 2 To measure the indicators, select the best objective function and optimization algorithm; The training set and the validation set are combined into a new training set, and training is performed on different machine learning models. The optimal relationship model is determined based on the effect evaluation of the trained different machine learning models on the test set, which is used to characterize the fundamental amplitude of the leakage current under the operating voltage of online monitoring. I x1 , third harmonic amplitude I x3 With DC reference voltage U 1mA The relationship between them.

[0013] Furthermore, using genetic algorithm with objective function R 2 combine1 Perform hyperparameter optimization.

[0014] Furthermore, the effect evaluation adopts the fit R 2 , mean absolute percentage error RMSE and RMS error MAPE .

[0015] Furthermore, the optimal relationship model is a GA-KNN model.

[0016] Furthermore, the optimal relationship model is used to monitor the leakage current of the arrester under the operating voltage online. I x , real-time prediction of typical aging parameters in the regulations, and intelligent online diagnosis of lightning protection status, specifically: Online monitoring of leakage current under arrester operating voltage I x , and further FFT changes to obtain its fundamental amplitude I x1 , third harmonic amplitude I x3 ; Fundamental amplitude I x1 and the third harmonic amplitude I x3 Input the optimal relationship model, and output labels are positive and negative. U 1mA Predicted value: when the predicted value is greater than 5%, immediate maintenance is required.

[0017] Compared with the prior art, the present invention has the following advantages and beneficial effects: The present invention can avoid the increase in voltage sensor cost caused by online monitoring of the resistive component of leakage current under the operating voltage, and realizes online diagnosis of the arrester state by using multiple monitoring parameters in combination with the regulation parameters, thereby improving the quality of the arrester state diagnosis.

[0018] At the same time, the diagnostic results of the present invention can directly reflect the changes in the procedure parameters, making the front-line staff's assessment of the arrester status more intuitive, which is conducive to the intelligent online operation and maintenance of on-site arresters, and further reducing the risks of arrester degradation and failure to the safe and stable operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of an implementation method of an online intelligent diagnosis method of a lightning arrester state according to an embodiment of the present invention; Figure 2 This is the on-site application method of the optimal relationship model of the embodiment of the present invention.

[0020] Figure 3 Embodiment of the present invention U 1mA and I 0.75U1mA Changes in circumstances; FIG. 4 (a) and FIG. 4 (b) are respectively embodiments of the present invention. I x and Ix1 , I x3 and I xr and I xr1 , I xr3 Rate of change; Figure 5 is the calculation result of the correlation coefficient of the embodiment of the present invention; Figure 6 It is a schematic diagram of optimization capabilities under different objective functions and optimization algorithms in an embodiment of the present invention; FIG. 7 (a) and FIG. 7 (b) are core bodies for lightning arresters according to an embodiment of the present invention. U 1mA Schematic diagram comparing predicted values ​​and true values. DETAILED DESCRIPTION

[0021] The present invention will be further described in detail below in conjunction with examples, but the embodiments of the present invention are not limited thereto.

[0022] Example like Figure 1 As shown, the embodiment of the present invention proposes an online diagnosis method for the arrester status, which uses an artificial intelligence algorithm to capture the leakage current under the operating voltage I x The changes of the arrester during its entire life cycle are predicted, typical aging electrical parameters in the regulations are predicted, and online intelligent diagnosis of the arrester status is achieved.

[0023] The specific steps include: S1 carries out the impact aging test of the ZnO valve plate of the arrester, records the relevant electrical parameters at the beginning and after each group of impacts until the valve plate is damaged, and carries out multiple impact aging tests, records the corresponding electrical parameters to construct the ZnO valve plate test data set.

[0024] Specifically: S11 numbers the ZnO valve plate samples of the lightning arrester respectively. Taking the ZnO valve plate for the lightning arrester YH5WS-17 / 50 as an example, 10 valve plates are selected and numbered #A1~#A10 respectively.

[0025] S12, set the impulse current waveform to 8 / 20μs, according to the guidance of the repeated transfer charge test in GB / T11032-2020 "AC gapless metal oxide arrester", calculate the waveform is 8 / 20μs, the rated transfer charge is 1.2C, and the corresponding impulse current amplitude of the repeated equivalent charge of 0.3C is 22.32kA. Use the above impulse current to carry out the impulse test on the arrester ZnO valve plate test piece.

[0026] In this embodiment, two shocks are set as one group, with an interval of 50 to 60 seconds between each shock. After each group of tests, the test is cooled to room temperature and the electrical parameters are recorded. The electrical parameters include the DC reference voltage U 1mA , 0.75 U 1mA DC leakage current I 0.75U1mA , leakage current under operating voltage I x and its resistive component I xr ; S13. Before impact, measure the initial electrical parameters of the ZnO valve plate. After each group of impact, measure the relevant electrical parameters. The specific measurement parameters are: DC reference voltage U 1mA , DC leakage current I 0.75U1mA , leakage current under operating voltage I x and its resistive component I xr .

[0027] S14. Repeat S12 and S13 until the valve plate is damaged by impact.

[0028] Further explanation: the electrical parameters measured multiple times are transformed by FFT to obtain the fundamental amplitude of leakage current under the operating voltage. I x1 , fundamental amplitude of the resistive component of leakage current under operating voltage I xr1 The third harmonic amplitude of leakage current under operating voltage I x3 , the third harmonic amplitude of the resistive component of the leakage current under the operating voltage I xr3 , construct the ZnO valve plate test data set.

[0029] The ZnO valve test data set includes the ZnO valve sample number, impact group number, DC reference voltage U 1mA , DC leakage current I 0.75U1mA , leakage current under operating voltage I x and its resistive component I xr ; At the same time, it also includes I x , I xr The FFT transformation results are I x1 , Ixr1 and I x3 , I xr3 .

[0030] In the S2 analysis shock aging test, the leakage current under the operating voltage during the entire life cycle of the ZnO valve piece I x and its resistive component I xr The typical aging parameters in the specification include DC reference voltage, ... U 1mA and 0.75 U 1mA DC leakage current I 0.75U1mA , specifically: S21 analyzes the process of ZnO valve plate during impact aging until failure. Typical aging parameters include DC reference voltage U 1mA , 0.75 U 1mA DC leakage current I 0.75U1mA And leakage current under operating voltage I x and its resistive component I xr The parameters that best characterize the state of the arrester are obtained based on the changes in the parameters and used as labels for machine learning.

[0031] The specific selection basis is to determine before the ZnO valve plate is damaged by impact. U 1mA and I 0.75U1mA The parameters that can better characterize the state of the ZnO valve plate are determined as the label. The label selection of the machine learning model in this embodiment is positive and negative. U 1mA .

[0032] In this embodiment, the impact current injection surface is defined as the front side. Taking the typical valve plate #A1 as an example, during the impact aging process, the positive and negative U 1mA With forward and reverse I 0.75U1mA The changes with the number of impact groups are as follows Figure 3 As shown. It can be seen that the ZnO valve plate is damaged before the valve plate is damaged. U 1mA The change rate has reached the aging threshold of 5% recommended by our regulations.I 0.75U1mA is still far less than the threshold of 50μA, then it is considered U 1mA It can more accurately reflect the state of the arrester. Therefore, choose the forward and reverse U 1mA Labels for machine learning models.

[0033] S22 calculates the selected tags, I x , I x1 , I x3 , I xr , I xr1 and I xr3 Pearson coefficients between parameters, correlation analysis, verification of leakage current under operating voltage I x and its harmonics I x1 , I x3 and the resistive component of the leakage current I xr and its harmonics I xr1、 I xr3 Equivalence in machine learning models, i.e. validation I x and I xr , I x1 and I xr1 , I x3 and I xr3 Is the Pearson coefficient greater than the threshold of 0.9? If so, further consideration I x , I x1 , I x3 The relative independence between them determines the fundamental amplitude of the leakage current under the operating voltage. I x1 , third harmonic amplitude I x3 Features of machine learning models.

[0034] The recorded I x and I xrPerform FFT transformation to obtain the fundamental amplitude I x1 , I xr1 and the third harmonic amplitude I x3 , and its changes are shown in Figure 4 (a) and Figure 4 (b).

[0035] Count selected tags U 1mA and I x , I x1 , I x3 , I xr , I xr1 and I xr3 The Pearson coefficient between each parameter was used for correlation analysis, such as Figure 5 As shown. I xr and I x , I xr1 and I x1 , I xr3 and I x3 There is a strong correlation between them. Selecting one of the parameters can enable the machine learning model to complete the learning of relevant knowledge. However, when they are used together as features, the performance improvement of the model is limited. I xr , I xr1 , I xr3 The above analysis further verifies from the correlation analysis of machine learning feature selection that the resistive component can be directly replaced by the leakage current under the operating voltage in artificial intelligence learning without affecting the model effect.

[0036] exist I x , I x1 , I x3 In, now I x and I x1 The correlation coefficient is large, and there is a risk of multicollinearity for some machine learning models. I x3 Depend on I xObtained by FFT, in order to ensure data consistency and relative independence between variables, select I x1 , I x3 As machine learning model features.

[0037] S3 is based on the ZnO valve test data set to study the performance of the relationship model between online monitoring parameters and typical aging parameters in the regulations under different machine learning models, and propose the optimal relationship model; S31. The ZnO valve plate impact aging test data set is divided into 64% for training set, 16% for validation set and 20% for test set; five machine learning models with different principles are selected: support vector machine (SVR), gradient boosting decision tree (GBDT), adaptive boosting (Adaboost), random forest (RF) and k nearest neighbor method (KNN) as the selection objects.

[0038] S32. Construct multiple hyperparameter optimization objective functions: R 2 combine1 = 0.6 R 2 train ﹢ 0.4 R 2 validation R 2 combine2 = 0.5 R 2 train ﹢ 0.5 R 2 validation R 2 combine3 = 0.4 R 2 train ﹢ 0.6 R 2 validation in R 2 train is the training set fit, R 2 validation The goodness of fit of the validation set is GA, BO and Whale Algorithm (WOA) are selected as optimization functions. Adaboost, GBDT and RF are used as examples to analyze the optimization capabilities under different objective functions and optimization algorithms.

[0039] like Figure 6The results show that the GA algorithm performs better than the other two methods in the test set. The selection of different objective functions in the GA optimized model has little effect on the model performance. From all the optimization results, R 2 combine1 The objective function is stable and robust under all optimization methods, and it takes the overall fitting of the data into account more in small sample situations.

[0040] Therefore, GA is selected as the optimization algorithm. R 2 combine1 Taking as the objective function, the above five machine learning models are optimized, and the results are shown in Table 1.

[0041] Table 1 Hyperparameter optimization results

[0042] S33. After the optimization is completed, the training set and the validation set are merged into a new training set, and training is performed on different machine learning models. R 2 , mean absolute percentage error RMSE , Root Mean Square Error MAPE To measure the indicators, the model is evaluated on the test set, and the relevant results are shown in Table 2.

[0043] Table 2 Test set performance

[0044] S34. According to the performance of different machine learning models on the test set in S33, the GA-KNN model is selected as the optimal model to characterize the leakage current and typical aging parameters under the operating voltage of online monitoring. U 1mA The relationship model between them.

[0045] S4. Use the optimal relationship model to monitor the leakage current of the arrester under the operating voltage online I x , real-time prediction of typical aging parameters in the process U 1mA , realize intelligent online diagnosis of lightning protection status, specifically: S41. Propose the field application method of the optimal relationship model in S3 to realize the intelligent online diagnosis of the arrester status, such as Figure 2 . Through online monitoring of the leakage current under the operating voltage of the arrester I x , and further FFT changes to obtain its fundamental amplitude I x1 , third harmonic amplitude I x3.

[0046] S42, the features selected by S2 I x1 , I x3 The corresponding values ​​are used as the input of the optimal model obtained in S3 to predict the label selected by S2, that is, the positive and negative typical aging parameters of lightning arresters that can only be monitored offline in my country's regulations. U 1mA .

[0047] S43. Calculate the predicted U 1mA Compared with the initial value, the change rate is combined with the recommended U 1mA The threshold is 5%. When the change rate is greater than 5%, immediate maintenance is required. When the change rate is less than 5%, a reasonable operation and maintenance plan is arranged according to the specific value. In this example, the arrester core (number #C1, #C2) with the same electrical performance as the YH5WS-17 / 50 arrester is used as an example to give the forward and reverse impact aging test results. U 1mA The comparison between the predicted value and the actual value is shown in Figure 7 (a) and Figure 7 (b). U 1mA The change trend of the predicted value is the same as that of the true value. When the predicted values ​​of #C1 and #C2 reach 5%, the true values ​​are 4.44% and 6.52% respectively, reaching the critical / maintenance-required state, demonstrating the effectiveness and rationality of the method proposed in the present invention.

[0048] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A method for online diagnosis of arrester status, characterized in that: The steps include: Conduct an impact aging test on the ZnO valve plate of the arrester, and record the electrical parameter values ​​at the beginning and after each set of impact aging tests until the valve plate is damaged; Analysis of the impact aging test, leakage current under operating voltage during the entire life cycle of the ZnO valve I x and its resistive component I xr The changes between the typical aging parameters in the specification and the DC reference voltage are analyzed and the characteristics and labels of the machine learning model are determined. U 1mA and 0.75 U 1mA DC leakage current I 0.75U1mA ; Based on the ZnO valve test data set, the leakage current under online monitoring of operating voltage under different machine learning models is studied I x The performance of the relationship model between the typical aging parameters in the regulations and the optimal relationship model are proposed; Online monitoring of leakage current under arrester operating voltage using the optimal relationship model I x , further predict the typical aging parameters in the regulations in real time and realize intelligent online diagnosis of the arrester status.

2. The method for online diagnosis of arrester status according to claim 1, characterized in that: The arrester ZnO valve plate impact aging test is carried out, and the electrical parameter values ​​at the beginning and after each group of impacts are recorded until the valve plate is damaged, specifically: According to the principle of repeated charge transfer in GB / T11032-2020, determine the impact current, number of impacts and interval between adjacent impacts of the impact aging test; After each set of shocks, the electrical parameters are measured after cooling to room temperature. The electrical parameters include the DC reference voltage U 1mA , 0.75 U 1mA DC leakage current I 0.75U1mA , leakage current under operating voltage I x and its resistive component I xr , the leakage current under the operating voltage I x and resistive component I xr Perform FFT transformation to obtain the fundamental amplitude of leakage current under operating voltage I x1 , fundamental amplitude of the resistive component of leakage current under operating voltage I xr1 The third harmonic amplitude of leakage current under operating voltage I x3 , the third harmonic amplitude of the resistive component of the leakage current under the operating voltage I xr3 ; The electrical parameter values ​​in multiple impact aging tests were recorded to construct a ZnO valve plate test data set.

3. The method for online diagnosis of arrester status according to claim 2, characterized in that: In the analytical impact aging test, during the entire life cycle of the ZnO valve sheet, the leakage current under the operating voltage I x Its fundamental amplitude I x1 , third harmonic amplitude I x3 , the resistive component of the leakage current under the operating voltage I xr Its fundamental amplitude I xr1 , third harmonic amplitude I xr3 The changes between the typical aging parameters in the regulations are analyzed, and the characteristics and labels of the machine learning model are determined, specifically: Analysis of DC reference voltage in ZnO valve plate impact aging test U 1mA and 0.75 U 1mA DC leakage current I 0.75U1mA According to the change of the arrester status, the parameters that best represent the arrester status are determined as the labels of the machine learning model; Calculate the selected tags, I x , I x1 , I x3 , I xr , I xr1 and I xr3 Pearson coefficients between parameters, correlation analysis, verification of leakage current under operating voltage I x and its harmonics I x1 , I x3 The resistive component of the leakage current under the operating voltage I xr and its harmonics I xr1、 I xr3 Equivalence in machine learning models, i.e. validation I x and I xr , I x1 and I xr1 , I x3 and I xr3 Is the Pearson coefficient greater than the threshold of 0.9? If so, further consideration I x , I x1 , I x3 The relative independence between them determines the characteristics of the machine learning model.

4. The method for online diagnosis of arrester status according to claim 3, characterized in that: Label selection of the machine learning model is positive and negative U 1mA .

5. The method for online diagnosis of arrester status according to claim 3, characterized in that: The fundamental amplitude of leakage current under the operating voltage I x1 , third harmonic amplitude I x3 As features for machine learning models.

6. The method for online diagnosis of the arrester status according to any one of claims 1 to 5, characterized in that: Based on the ZnO valve test data set, the leakage current under online monitoring of operating voltage under different machine learning models is studied I x The performance of the relationship model between the typical aging parameters in the regulations is analyzed, and the optimal relationship model is proposed, which is as follows: The ZnO valve plate test data set is divided into training set, validation set and test set; Construct hyperparameter optimization objective function: R 2 combine1 = 0.6 R 2 train ﹢ 0.4 R 2 validation R 2 combine2 = 0.5 R 2 train ﹢ 0.5 R 2 validation R 2 combine3 = 0.4 R 2 train ﹢ 0.6 R 2 validation in R 2 train is the training set fit, R 2 validation is the goodness of fit of the validation set; Based on the training set and validation set data, we use the optimization objective function and various optimization algorithms to optimize hyperparameters. The fitness on the validation set R 2 To measure the indicators, select the best objective function and optimization algorithm; The training set and the validation set are combined into a new training set, and training is performed on different machine learning models. The optimal relationship model is determined based on the effect evaluation of the trained different machine learning models on the test set, which is used to characterize the fundamental amplitude of the leakage current under the operating voltage of online monitoring. I x1 , third harmonic amplitude I x3 With DC reference voltage U 1mA The relationship between them.

7. The method for online diagnosis of arrester status according to claim 6, characterized in that: Using genetic algorithm with objective function R 2 combine1 Perform hyperparameter optimization.

8. The method for online diagnosis of arrester status according to claim 6, characterized in that: The effect evaluation was performed using the goodness-of-fit R 2 , mean absolute percentage error RMSE and RMS error MAPE .

9. The method for online diagnosis of arrester status according to claim 1, characterized in that: The optimal relationship model is a GA-KNN model.

10. The method for online diagnosis of arrester status according to claim 1, characterized in that: The optimal relationship model is used to monitor the leakage current of the arrester under the operating voltage online. I x , real-time prediction of typical aging parameters in the regulations, and intelligent online diagnosis of lightning protection status, specifically: Online monitoring of leakage current under arrester operating voltage I x , and further FFT changes to obtain its fundamental amplitude I x1 , third harmonic amplitude I x3 ; Fundamental amplitude I x1 and the third harmonic amplitude I x3 Input the optimal relationship model, and output labels are positive and negative. U 1mA Predicted value: when the predicted value is greater than 5%, immediate maintenance is required.

Images