A multivariate ensemble framework for short term solar photovoltaic output power forecast

Abstract

In emerging renewable energy resources, solar photovoltaic (PV) is substantially important to fulfil the future electricity demand. One of the major challenges for large scale integration of PV into the grid is intermittent and uncertain nature of its output. Therefore, it is utmost important to forecast the solar PV output power with higher accuracy. In this paper, a novel ensemble forecast framework is proposed based on autoregressive (AR), radial basis function (RBF) and forward neural network (FNN) predictors. The neural predictor (FNN and RBF) are trained with particle swarm optimization (PSO) to enhance the prediction performance. Furthermore, wavelet transform (WT) technique is applied to remove the sharp spikes and fluctuations in it. In addition, correlated variables such as PV output power data, solar irradiance, temperature, humidity and wind speed are applied as inputs to multivariate ensemble network. The performance of proposed framework is analyzed for one day and week ahead case studies. The selected days from each season are a clear day (CD), partial cloudy day (PCD) and cloudy day (CLD). The proposed forecast framework provides a reduction in forecast nRMSE in seasonal daily and week ahead case studies.