Achieving Robust and Accurate Power Distribution Grid Damage Forecasting via a Two-Stage Forecasting Method

Abstract

This paper presents a method to forecast storm-induced power distribution grid damage. Three sets of historical data are used: storm data, local weather data, and power distribution grid damage data from January 2008 to March 2018. Before developing the damage forecasting method, the key explanatory variables are identified by using stepwise regression analysis to develop a simpler and robust forecasting model. Thereafter, this paper proposes a two-stage damage forecasting method. Random Forest (RF) and feed-forward neural network (FFNN) model are used for forecasting grid damages. RF is used to classify the no damage and damage cases before the damage forecasting and then FFNN is used to forecast the number of grid damages only for the damage cases. The actual storm event data is used to verify the proposed method by using Mean Absolute Error (MAE).