A Longitudinal Study of Customer Electricity Load Profiles

Abstract

We present a novel approach for studying evolving customer electricity load profiles. Based on the daily changes that may happen in a power grid, we devise a network-based method to identify and track the evolution of electricity consumption patterns over days. The tracking of these evolving patterns enables us to: (a) use Cox regression and LSTM recurrent neural network for modeling the lifetime of electricity consumption profiles and (b) trace the trajectories of customer electricity consumption behaviors to perform load forecasting.