Long term load forecasting with hourly predictions based on long-short-term-memory networks

Abstract

The conventional methodology for long term load forecasting is mostly restricted to electricity load data with monthly or annual granularity. This leads to forecasts with very low accuracy. In this paper, a novel method for long term load forecasting with hourly resolution is proposed. The model is fundamentally centered on Recurrent Neural Network consisting of Long-Short-Term-Memory (LSTM-RNN) cells. The long term relations in a time series data of electricity load demand are taken into account using LSTM-RNN and hence results in more accurate forecasts. The proposed model is implemented on real time data of ISO New England electricity market. Precisely, publicly available data of twelve years from 2004 to 2015 have been collected to train and validate the model. Electricity demand predictions have been made for a period of five years from 2011 to 2015 on a rolling basis. The proposed model is found to be highly accurate with a Mean Absolute Percentage Error (MAPE) of 6.54 within a confidence interval of 2.25%. Moreover, the model has a computation time of approximately 30 minutes which is favorable for offline training to forecast electricity load for a period of five years.