Reactive to Proactive: Using Outage Data and System Parameters to Predict URD Cable Failures Open Access
Downloadable ContentDownload PDF
As Underground Residential Development (URD) cable systems approach the end of expected service life, electric utilities face millions of dollars in operational and maintenance costs and lower reliability performance. Early URD cable systems installed prior to the mid-1980s typically incorporated direct-buried, unjacketed polyethylene cable, which has been proven to experience significantly higher failures rates and provides a shorter overall service life than updated modern installations. URD cable system data was compiled, cleaned, and fit to discrete and zero-inflated probability distributions to identify operational and environmental sources of cable faults, with the zero-inflated negative binomial model providing the best overall fit in terms of Akaike Information Criteria (AIC) and Bayesian Information Criteria (BIC). The covariates identified as the most significant drivers for cable events were used to test manual methods of failure prediction against machine learning algorithms. Artificial Neural Network and Random Forest machine learning tools were found to be significantly more accurate than all other manual and machine learning methods for forecasting the number and impact of URD cable system failures in terms of common reliability metrics.