Outage Prediction of Overhead Line Based on Regional Ensemble Learning

Severe convective weather seriously affect the operation of power system, so the forecast of overhead line outage is important. The research on this problem has the following difficulties. First, large-scale forecast problem, the objective environment of overhead lines in different regions is quite different, and the single model has limited accuracy. Second, the training sample is imbalanced. Overhead line outage sample account for a small proportion in the actual operation data, so the problem of sample imbalance exists in model training. To solve the above problems, this paper proposes an adaptive regional ensemble learning algorithm to predict the outage probability of overhead line in severe convective weather. Aiming at sample imbalance, this paper designs a improved cross entropy loss function to reduce the loss weight of a large number of simple negative samples. Our method has been verified by actual data, which can provide technical support for the grid digital forecast.