Forecasting of Electricity Prices for Scheduling of Residential Loads Using LSTM Neural Networks

Abstract

The electricity prices find wide applications in the present electricity markets. Generation companies use the forecasted electricity prices to plan their expenses, which helps the aggregator provide better consumer services. The market players also use it to strategize selling electricity to the distribution companies in the energy exchange market. The forecasted electricity prices are also used to implement Demand Response (DR) programs. DR programs reduce peak demand, which can be achieved by scheduling the loads. All these applications require accurate forecasting of electricity prices, but its volatile nature makes it challenging to make accurate predictions. Electricity price forecasting on hourly basis is presented using the Long Short Term Memory (LSTM) Neural Network model. LSTM model can extract highly complex relationships between parameters. It uses feedback property to predict electricity prices accurately. The proposed model predicts the prices based on the historical hourly prices, historical load demand data, and other quantities on which the prices depend. The accuracy of the model is compared with other baseline models. The model comes out to be the most accurate of all and has the lowest Mean Absolute Percentage Error (MAPE) and the highest R2 value.