Prognosis Of Underground Cable Via Online Data-Driven Method With Field Data

Underground distribution cable is widely used because of its safety and aesthetics. It is very important to keep the reliability of the underground cable. Many specialized testing and offline diagnostics have been well-developed aiming at cable fault detection and location. If the remaining life of distribution cables could be predicted, not only will the maintenance expenditure be reduced, but also the quality and reliability of delivered power will be improved. To predict the remaining life of a monitored cable, online prognosis method has yet to be developed.In this dissertation, in order to prevent unexpected electrical outage and to save on repair expenses, a prognosis method of the monitored underground cable is proposed via on-line data-driven method with field data. The field data of voltage and current was collected from an underground distribution cable lateral installed in a residential area. The aim of this dissertation is not to propose a new aging feature or mechanism but to propose a method, which can continually offer valuable remaining life prediction of the monitored underground cable as time goes on. The method is nondestructive and nonintrusive. For cable insulation aging, the degradation process is an accumulated process. From a view of monitoring data, the degradation effect is relative to increments of some features. When the accumulation of degradation effect reaches a threshold, breakdown occurs quickly and the life time of the cable comes to an end. Therefore, the remaining useful life of the cable could be forecasted by predicting the time for cumulative effect to reach the threshold. The main work is shown as follows.(1) The preprocessing module performs resampling, DC removal, denoising and data cleaning tasks to prepare the input data for the subsequent analysis. The wavelet-based denoising method is applied to remove noise from the signal. It is necessary to separate signals with load change transients, because the behavior of some abnormalities in distribution systems due to normal switching operations might resemble the behavior attributed to incipient behavior. The load change transients are recognized though a rule-based classifier. Since the recorded signals with load change transients are separated, former recorded signal data is used to pad the missing data. The data set is divided into training data set and testing data set.(2) Differing from determining information features based on aging mechanisms, potential useful features in this research are selected based on data-driven method using large amount of collected field data. A complete analysis of signals is carried out by extracting features in time, frequency, and time-scale domain. A few hundred features are extracted to describe the signals, such as statistical features, RMS features, spike features, FFT features, wavelet features and severity features. Then a data-driven feature subset selection method selects potential useful features from the extracted features. The rationality of selected features is verified in view of cable aging mechanisms, after that the useful features are finally determined.(3) A Sliding-Window regression method was proposed to predict the time for the cumulative effect to reach the threshold, and the time is the predicted fault time, meanwhile, the remaining useful life is forecasted. During the training process, onset values and threshold values are extracted from the training data.(4) The performance of the developed prognosis method was tested and evaluated using the field testing data. The standard deviation of the prognosis results at different arrival time is used as a source of uncertainty estimation.(5) During the proposing process of the prognosis method, it is found that the partial discharge of the monitored cable can be detected from the cable current. The monitoring of current data is easy to be online, nondestructive and nonintrusive. Control experiments are conducted to compare this online-current-PD-detection method with the ERA method. And the experimental results have verified the PD detection results of the proposed method.The results demonstrate that the prognosis method can continually offer valuable remaining life prediction of the monitored underground cable as time goes on. When the predicted fault times approach the actual fault time, the standard deviation value is small. When standard deviation value is small, there are steady forecast results and usually small residuals. The spike of neutral current can be use to detection the PD of the monitored cable...