Solar Radiation Forecast Using Cloud Velocity for Photovoltaic Systems

Calvin Kong Leng Sing (Malaysia – Japan International Institute of Technology (MJIIT), Universiti Teknologi Malaysia, Jalan Sultan Yahya Petra, 54100 Kuala Lumpur,)
Tan Lit Ken (Malaysia – Japan International Institute of Technology (MJIIT), Universiti Teknologi Malaysia, Jalan Sultan Yahya Petra, 54100 Kuala Lumpur, eknologi Malaysia, Kuala Lumpur, Malaysia)
Lai Kok Yee (Malaysia – Japan International Institute of Technology (MJIIT), Universiti Teknologi Malaysia, Jalan Sultan Yahya Petra, 54100 Kuala Lumpur,)
Jane Oktavia Kamadinata (Malaysia – Japan International Institute of Technology (MJIIT), Universiti Teknologi Malaysia, Jalan Sultan Yahya Petra, 54100 Kuala Lumpur,)
Nor Azwadi bin Che Sidik (Malaysia – Japan International Institute of Technology (MJIIT), Universiti Teknologi Malaysia, Jalan Sultan Yahya Petra, 54100 Kuala Lumpur,)
Yutaka Asako (Malaysia – Japan International Institute of Technology (MJIIT), Universiti Teknologi Malaysia, Jalan Sultan Yahya Petra, 54100 Kuala Lumpur,)
Lee Kee Quen (Malaysia – Japan International Institute of Technology (MJIIT), Universiti Teknologi Malaysia, Jalan Sultan Yahya Petra, 54100 Kuala Lumpur,)

Journal of Engineering and Technological Sciences
Vol. 50 No. 4 (2018)

Today, solar energy is used in a many different ways. One of the most popular technological developments for this purpose is photovoltaic conversion to electricity. However, power fluctuations due to the variability of solar energy are one of the challenges faced by the implementation of photovoltaic systems. To overcome this problem, forecasting solar radiation data several minutes in advance is needed. In this research, a methodology to forecast solar radiation using cloud velocity and cloud moving angle is proposed. Generally, a red-to-blue ratio (RBR) color model and correlation analysis are used for obtaining the cloud velocity and moving angle. Artificial neural network (ANN) forecast models with different input combinations are established. This methodology requires lower computational time since it only uses part of the pixels in the sky image. Based on R-squared analysis, it can be concluded that the ANN model with inputs of cloud velocity and moving angle and average solar radiation showed the highest accuracy among other combinations of inputs. The R-squared value was 0.59 with only a relatively small sample size of 42. The proposed model showed a highest improvement of 75.79% when compared to the ANN model based on historical solar radiation data only.