A Comparative Study of ARIMA and RNN for Short Term Wind Speed Forecasting

Abstract

The requirement of electricity is increasing day by day due to industrial and technological development and due to that the gap between generation and demand is also increasing. To fulfil the current electricity demand we are switching towards renewable sources of energy as nonrenewable sources are limited in stock and are depleted day by day. Moreover, renewable sources are more cleaner and environment friendly as they cause small or no carbon emission. Wind energy plays a larger role in providing electricity to industrial and domestic consumers. But wind has a stochastic nature and so we are not sure about how much energy it will give in the coming minutes or hours or days or months or years. Thus, we need an effective forecasting model, which helps in minimizing the use of conventional power plants (by unit commitment) and also optimizing the plants (by economic dispatch). This paper presents a comparative study of a time series model (ARIMA, i.e. Auto Regressive Integrated Moving Average) and a deep learning model (RNN, i.e. Recurrent Neural Network). Here, RNN is used with the combination of LSTM (Long short-term memory).