Study On Prediction For Corona Onset And Breakdown Voltages Of Air Gap Based On Electric Field Features And Support Vector Machine

The ultra-high voltage (UHV) grid can significantly improve the security, reliability, flexibility and economy of power grid. Electromagnetic environment and external insulation characteristics are key focuses in the design and operation of UHV power transmission and transformation projects. Their corn contents are corona control and selection of air gap, and basics are the corona onset and breakdown characteristics, respectively. The assessment test is to determine if the design of UHV project meets the requirements. However, the corona and breakdown tests of air gap in the UHV projects are rather time-consuming and expensive and fail to include every kind of air gap, and almost all of the research conclusions are based on the scale of air gap, consequently their scope have been greatly restricted for complex gap structures which are difficult to be described by geometric scale in actual projects. So it is necessary to carry out studies on mechanism and model of air gap discharge. On this basis, corona onset and breakdown voltages of air gaps with various kinds of structures under different operating conditions are predicted by simulation, which are used to guide the design of power transmission and transformation projects.Currently, the prediction methods of corona onset and breakdown voltages of air gap mainly include empirical and semi-empirical formulas and physical models. However, the application scopes of empirical and semi-empirical formulas are limited, because some parameters are determined under some specific test conditions or assumptions. The physical model has made considerable progress in the past20years, which predicts the corona and breakdown characteristics according to the physical process of air discharge. Nevertheless, due to the extreme complexity, there are still some unsolved problems about the physical model, leading to some deviation between the prediction and experimental values. Therefore, the physical model is still difficult to effectively guide the design of actual power transmission and transformation projects. In response to these problems, this thesis proposes a prediction method based on electric field features and support vector machine (SVM) for corona onset and breakdown voltages of air gap. Using the proposed method, DC corona onset voltages of fittings such as rod and ball structures and stranded conductors, and breakdown voltages of air gaps with different electrode structures are predicted, also some related influencing factors are analyzed. Meanwhile, the corona onset and breakdown voltages of some structures are measured. Combining with test data obtained from related literatures, the proposed prediction method for corona onset and breakdown voltages of air gap is proved to be valid. Finally, a measure is proposed to improve the insulation strength of a long air gap, and corresponding tests are carried out to verify it. Research results of this thesis have certain theory and practical application values for the corona control and selection of air gap in the UHV power transmission and transformation projects. The main contents and achievements of the thesis are as follows:(1) A prediction method based on electric field features and support vector machine for corona onset and breakdown voltages of air gap is proposed. And electric field features are defined to characterize the gap structure. The corona onset and breakdown voltage prediction model of air gap is established by using electric field features as input parameters to SVM, and whether corona onset or breakdown would happen to the air gap as output parameters to SVM, which changes the regression problem to a binary classification problem.(2) Using the proposed method, the positive DC corona onset voltage of a rod-plane gap is predicted. The effectiveness and superiority of the proposed method have been proved by comparing its predicted values with the experimental values and the predicted values of existing methods. An ozone detection method is presented to measure the corona onset voltage. Combining with the proposed prediction method, the influence of a dielectric enclosure on the negative DC corona onset voltage of rod-plane gap is analyzed. The results show that the space charge accumulated on the surface of dielectric enclosure increases negative DC corona onset voltage of rod-plane gap when the gap distance is greater than a certain value, and the corona onset voltage gain increases with the increase of gap distance and the decrease of inner diameter of the dielectric enclosure. The corona onset voltages of grading balls in a valve hall of±660kV DC converter station are studied by experiment and prediction, and control values of electrical field without corona for grading balls are obtained. By comparing the numerical calculation values of electrical field through finite element method (FEM) with the control values, it is proved that the grading balls can be operated without corona in the valve hall.(3) The proposed method is applied to predict the negative DC corona onset voltage of strand conductors, and its predicted values are compared with the experimental values and the predicted values of existing methods. Combining with the photoionization prediction model of corona onset voltage, the relationships between corona onset voltage of strand conductors and atmospheric parameters like pressure, altitude and temperature, and structural parameters like radius, height, bundle number and spacing, are analyzed. Also the relationships between the surface irregularity factor of a strand conductor and its radius and number of outmost thin wire are discussed.(4) The power frequency breakdown voltage of short air gap in a slightly uneven electrical field is predicted, and power frequency withstand voltage test is carried out. In this research, typical electrode structures including sphere-sphere, rod-plane and sphere-plane gaps and atypical electrode structure of sphere-plane-sphere gap are considered. The proposed method has been proved valid to the prediction for breakdown voltage of short air gap in a slightly uneven electrical field, by comparing its predicted values with the experimental values. The BP neural network, RBF neural network and SVM are adopted to predict the power frequency breakdown voltage of sphere-sphere air gap considering the effects of temperature and humidity, and the results are compared and analyzed. The SVM method is proved to be superior in nonlinear approximation and generalization ability.(5) The breakdown characteristic of long air gap in an extremely uneven electrical field is significantly affected by the space charge produced by corona discharge. In response to this issue, two methods of considering the effect of corona are proposed and compared on the basis of breakdown voltage prediction model of short air gap in a slightly uneven electrical field. The proposed method is successfully applied to the prediction for the50%positive switching impulse breakdown voltage of a long sphere-plane air gap. Therefore, the test can be replaced by prediction within a certain range of accuracy to reduce the number and cost of test. Finally, a measure of multi-gap structure is proposed to improve the insulation strength of long air gap, and corresponding tests are carried out to verify it.