One-day ahead PV power forecasts using 3D Wavelet Decomposition

The paper presents a spatio-temporal forecasting for the photovoltaic (PV) power generation by combining the three-dimensional wavelet transform (3D-DWT) and the Least Square Support Vector Machines (LS-SVM) to deal with historical time series data of distributed PV plants in both spatial and temporal domain. The proposed forecasting method applies the wavelet decomposition to the PV power data collected from several PV installations in a three-dimensional space taking into account the spatial distribution of the PV locations and the related PV output power in a defined time framework. The wavelet decomposition output is then used as input for a forecasting model based on the LS-SVM to predict the solar PV power of each plant. A case study is presented using hourly time series of 9 PV installations located in the Greek island of Rhodes in order to predict the power generation of each individual PV plant at 24 hours-ahead time horizon. The forecast performance of the proposed approach is investigated by error metrics and compared with a prediction model based on simple LS-SVM to quantify the improvements achieved by the proposed method.