| Summary: | 碩士 === 國立臺灣科技大學 === 電機工程系 === 107 === Underground cable is the main equipment of power system .It is not easily damaged by natural disasters. Once the fault occurs, in addition to the maintenance time, the cost is relatively expensive, and the impact on people's livelihood electricity consumption is very obvious. Therefore, the method of detecting partial discharge of underground cable is relatively important.This research is to study the insulation state of the underground cable straight joint. Partial discharge detection can be used to evaluate the insulation state of underground cables, so as to avoid unexpected accidents.
In this paper, two kinds of 15 sets of underground cable joint insulation damage test data were used. After dissecting the cable test body, the damage path was observed, the abnormal types of cable were analyzed and classified, and 104 characteristic parameters were taken after noise suppression and feature extraction. Finally, the phase center of gravity and standard deviation of the number of discharges in the positive discharge area are selected as the characteristic values of the main use. It is found that the accelerated deterioration test of each group of cable joints shows that the trend of partial discharge is irregular, and it is difficult to sort out the general rules for all test data. In this study, the findchangepts function is used to define the turning point of the change of the insulation state. Instead of using the judgment method of the slope change of the discharge times, the double characteristic curves of the 15 sets of cable test data are initially defined in the attention area and the danger zone, and then all the data are collected. The random method is divided into training and test data. After the training data is calculated by the support vector machine to calculate the classification hyperplane, the test data is used to verify the prediction success rate. After verification, the risk can be predicted before the insulation failure, and the success rate can reach 82%. The classification hyperplane can also be followed by a curve fitting tool to derive multi-dimensional equations, which can be used in the future to determine the effect of the cable state on the site.
|
|---|
