| Summary: | 碩士 === 國立臺灣科技大學 === 電機工程系 === 105 === The purpose of this study is to establish an underground power cable insulation real-time monitoring system using partial discharge data acquired from the straight cable joints to diagnose the insulation condition of the underground power cable. Timely maintenance can be carried out upon the real-time updates to prevent accidents caused by cable degradation which in turn results in economic losses and damages.
First, the research designs two types of sample underground power cable joints with defects, 9 sets for group A and 7 sets for group B, respectively, and perform long-term pressure degradation testing on the cable joints to acquire a large quantity of partial discharge data. Next, through procedures such as noise suppression and data reduction, 104 eigenvalues are acquired. Then the moving average method is applied to fitting the electric discharge tendency during the growth stage. However, considering the insulation diagnostic mechanism of the research team from the past, the preliminary screening of eigenvalues that precedes the insulation diagnosis is an assessment by experts who select several superior parameters from the 104 eigenvalues to pair degradation bicharacteristics. Nonetheless, this method can still leave out desirable eigenvalues. Therefore, this paper proposes the technique of brute-force search (exhaustive search) also known as haar cascades classifier or haar-like features. Take the 9 sets from group A as an example. First, pair the 104 eigenvalues to acquire 5356 degradation bicharacteristics and find the 9 breakdown points of every degradation bicharacteristic to obtain a rectangle which is the feature rectangular area. The procedure is to cut the 9 pairs of bicharacteristics in each set and determine the bicharacteristic data that first enter into the rectangular area and those that enter afterwards as in the crisis period. The values that enter prior to this should be regarded as in the attention period. Continue to converge the rectangular area until the cutting area reduces to its minimum to obtain the power cable service rate of that set. Next, based on the assessment of the computer, screen the eigenvalues with a service rate that is above 90%, 80%, and 70%, respectively, from the 5356 bicharacteristics of which the convergence has completed as the final results of this paper. Up to this point, the 104 eigenvalues have already been screened successfully by the computer assessment while the insulation diagnosis is carried out at the same time. This considerably improve the accuracy of the eigenvalue screening which also perfects the warning system for the power cable insulation condition.
|
|---|
