Solar Power Prediction using Machine Learning Algorithms: A Comparative Study
Sana Mohsin Babbar, Lau Chee Yong
Abstract
Energy industry has been revolutionized rapidly over past few decades. The consumption of electricity from the renewable sources is increasing speedily due to the low maintenance and environmental friendly nature. Now a days, installment of PV solar cells have captured the energy market in an unparalleled way. The big challenge behind integration of PV arrays to the grid is its unpredictable and fragmented variation in terms of having PV power output. Therefore, construction of an effective predications models are needed as a solution for planning and controlling of the power grid. In the past, ample of machine learning algorithms i.e. Artificial Neural Networks (ANN), Moving Average and Fuzzy logics etc. has been used to overcome the volatile nature of solar energy. The Challenges can be effectively met by choosing robust machine learning algorithm. In this paper, comparative analysis has been made in terms of prediction accuracy by using different machine learning algorithms. Recurrent Neural Networks (RNN), Support Vector Machine (SVM), Autoregression with exogenous variable (ARX) Feedforward Neural Network with gradient descent momentum (FFNN-gdx) and Least Absolute Shrinkage and Solution Operator (LASSO) have been adopted as a predictive tool for bringing efficacy and precision in the PV power output. The results confirms that non-linear techniques i.e. RNN and SVM outperforms very well as compare to other linear models for predicting solar power output. Moreover, a comparative study helps out to select an appropriate and effective solution for forecasting and prediction models.