Using LSTM To Perform Load Modelling For Residential Demand Side Management

Energy consumption from the residential sector forms a large portion of the grid demand. The growing accessibility of residential load profile data presents an opportunity for improved residential load forecasting and more effective demand side management strategies. However, the stochastic nature of residential power usage limits the effectiveness of traditional load forecasting techniques. Machine learning is a tool well-suited for predicting processes containing stochastic elements. Long short-term memory (LSTM) neural networks are especially suited for predicting time-series data such as electrical load profiles. This paper investigates the efficacy of an LSTM to perform residential load forecasting. Two LSTM-based prediction strategies are compared by forecasting the load profile for a set of five simulated, periodic, temperature profiles representing a population of electric water heaters (EWH). In Method A an aggregated load profile representing a feeder-point measurement is applied to a single LSTM model to forecast the load. In Method B each simulated temperature profile, representing appliance-level measurements, is applied to its own LSTM temperature prediction. The predicted temperature profiles are combined and used to determine the load forecast. Method A outperforms Method B and is able to predict the EWH energy with a percentage error of approximately 0.76%.