Time Series Auto-Regressive Integrated Moving Average Model for Renewable Energy Forecasting

Abstract

Due to the rapid pace of industrialization and growing demand for energy consumption, forecasting of renewable energy has become an inevitable focus of many recent studies. In this paper, our aim is to develop a univariate auto-regressive integrated moving average (ARIMA) model to forecast daily and monthly wind speed and temperature based on 15 years (2000–2014) of hourly data at Charanka Solar Park, Gujarat. To check the stationarity of time series, Dickey fuller test and rolling statistics plots are employed. Autocorrelation and partial autocorrelation plots are used to determine potential models, whereas Akaike information criterion (AIC) and Bayesian information criterion (BIC) are utilized to establish ARIMA (2, 1, 2) model. After rigorous training, model performance is validated using root means square (RMS) errors. The entire methodology is implemented in python using pandas for data exploration, and stats and scikit-learn libraries for model building and validation. On comparing results based on the log-likelihood, AIC and BIC values, we conclude that the ARIMA model provides better accuracy to the wind power forecasting as compared to solar power on the selected dataset.