Power Generation Voting Prediction Model of Floating Photovoltaic System

Abstract

Solar energy is the most promising renewable energy within the Gulf area as annual solar irradiance is among the highest in the world (>2000kWh/m2). Therefore, countries within the Gulf area have focused their energy investment on solar energy harvesting, especially Photovoltaics (PV). Photovoltaics (PV) power output is highly dependent on environmental conditions variability. Accurate PV generation power prediction models are essential to investigate the effects of varying environmental conditions and ensure solar power converters' optimum performance whilst meeting peak demand through various environmental conditions. The environmental data which is analysed and discussed in this paper includes air temperature, relative humidity, Photovoltaics (PV) surface temperature, irradiance, dust, wind speed, and output power. The model proposed in this paper optimises and trains three prediction algorithms, including Artificial Neural Network (ANN), Multi-Variate (MV), and Support Vector Machine (SVM). The model deploys three well-known prediction algorithms and voting algorithm to decide the optimum prediction of PV generation power. Furthermore, the voting algorithm shows high prediction accuracy of the output power given the environmental conditions. The Mean Square Error (MSE) for the Artificial Neural network (ANN), Multi-variate (MV), and Support Vector Machine (SVM) are 98, 81, and 82, respectively. In comparison, Mean Squared Error (MSE) of the voting algorithm is significantly lower which is just above 53. The proposed PV power generation prediction algorithm shows reliable outcome with respect to the environmental conditions in Qatar. This tool is expected to assist in the design process of Photovoltaics (PV) plants design where energy generation is highly predictive using proposed voting algorithm.