Implementation of Long Short Term Memory Neural Network Model for Electrical Load

This paper describes implementation of long short-term memory neural network model for electrical load forecasting. The accurate forecasting of energy demand is the key to build impeccable energy management strategies that help in efficiently generating and distributing energy among users. A number of conventional models, machine learning models, artificial neural networks (ANN), recurrent neural networks (RNN) have been adapted over decades. But due to nonlinear, non-stationary and nonseasonal nature of the electric load time series, those methods couldn't provide accurate results for longterm load forecasting. This paper explores LSTM (Long shortterm memory) network, a special type of recurrent neural network, which is observed to exhibit it's effective potential in forecasting both short-term and long-term energy demands accurately. A dynamic model based on LSTM is developed to forecast energy demand of an electric utility company in USA, by automatically learning temporal sequences and long term dependencies from its historic data over years. This novel method exhibited good potential & accuracy, which is substantiated by RMSE (Root mean square error) method.