Data-driven methodology to predict distribution lines failure location

Distribution lines are one of the most critical assets in distribution networks. In fact, overhead lines and subterranean cables are subjected to numerous internal and external factors that can cause failures and degradation over time. To prevent customer disconnection and to ensure continuous electricity delivery, the Distribution System Operator (DSO) strives to minimize the number of distribution line failures by carrying out inspection and preventive maintenance actions. Typically, HV and MV networks cover wide areas of the territory and comprise many lines of different types (overhead & subterranean) and equipment (e.g. conductor, isolator, poles), which makes it difficult to predict when failures will occur, which distribution line will fail and its location. The latter information is especially relevant since some distribution lines can cover a considerable distance. Motivated by a real-world application, this work presents a methodology to predict and locate future HV and MV distribution line failures. The methodology encompasses clustering techniques to group lines sharing similar characteristics, identifying the most relevant factors on the lines' degradation. In addition, historical records are leveraged by a prediction algorithm to estimate the number of failures and the failed section of the line. The approach was validated using data from the Portuguese HV and MV DSO (E-Redes). The results highlight the advantages of the proposed method compared with benchmark approaches.