Daily wind-based interruption prediction in Enedis' Medium Voltage Overhead Lines using time series data

Abstract

Overhead lines represent about 50% of the French Medium Voltage/Low Voltage electrical network. Therefore, most of the French electrical network is vulnerable to weather conditions. In order to optimize financial investments, and plan maintenance campaigns, a reliability model would be helpful, especially if based on three main features of the electrical network: the intrinsic reliability linked to natural ageing, the sensibility to frequent weather conditions, and its vulnerability to extreme and rare climatic hazards which it is more difficult to hedge. The history of electrical failures is used to tune and build this model in order to propose a failure rate made of those three features. Then we will present a machine learning model which predict a few days ahead of time, the number of failures caused by the prevailing winds of a French area.