ANALYSING OVERHEAD MEDIUM VOLTAGE NETWORK CRITICITY WITH AI

With a network of 1.4 million km, Enedis spends several hundred million Euros every year to maintain and renew its existing network. Properly selecting which part of the network to renew is key to optimize the efficiency of that investment.For the past three years, Enedis has been using AI to optimally choose the parts of the network to be renewed. Machine learning algorithms have been successfully deployed to target underground networks replacement, both for LV and MV. The methods used to target underground faulty lines are almost exclusively based on the intrinsic network characteristics and the history of the line faults. Those methods do not perform well on the overhead networks as the external environment has a much stronger impact on the line faults than in the underground use case.To successfully discriminate risky overhead networks with machine learning algorithms, new features abstracting the environment have been created, mostly using geographical open data available for France.Part of the challenge is to combine both long-term (network renewal) and short-term (resilience to storms) approaches to improve overhead MV network management.