Paper Conference

Proceedings of Building Simulation 2017: 15th Conference of IBPSA


Recurrent Neural Network based Deep Learning for Solar Radiation Prediction

Fuxin Niu, Zheng O'Neill
Department of Mechanical Engineering, The University of Alabama, Tuscaloosa, USA

Abstract: In the past decade, the data based solar radiation prediction models such as artificial neural network (ANN) model, support vector machine (SVM) model, state space model (SS), Bayesian network model (BN), and autoregressive with exogenous terms (ARX) model appeared in abundance along with the advanced computing technologies and large amount of data storage devices. These inverse models performed relatively well in the solar radiation prediction. Currently, how to further improve the accuracy of the solar prediction algorithms for building application such as model predictive controls remains as a challenge. In the era of big data, deep learning is being explored widely. This is a new area of machine learning research with an objective of moving machine learning closer to one of its original goals: Artificial Intelligence. To further investigate the solar radiation prediction using neural network, recurrent neural network (RNN) based deep learning algorithm is proposed and compared with other data-driven methods: ARX, SS, ANN, and BN methods. It was concluded that RNN method has the best performance in terms of the accuracy of solar radiation prediction for the selected case study.
Pages: 1890 - 1897