Time series forecasting of total daily solar energy generation: A comparative analysis between ARIMA and machine learning techniques
Sharif Atique, Subrina Sultana Noureen, Vishwajit Roy, Stephen Bayne, Joshua Macfie
Abstract
In this paper, the potential of machine learning based methods for time series forecasting of total daily solar energy generation has been explored. Firstly, the time series is modeled using the seasonal version of well known classical method auto regressive integrated moving average (ARIMA) and its performance is later compared to two other popular machine learning methods, support vector machine (SVM) and artificial neural network (ANN). The potential of machine learning based methods in this line of work is demonstrated by the superior performance of SVM. However, the reasons behind the low yield of ANN need to be inspected to enhance our understanding. In spite of SVM's relative success in prediction of solar generation, the overall accuracy still needs to be improved and the methods to achieve this objective should be researched in future.