Support Vector Regression for Predicting Intra-day Photovoltaic Generation in Algeria

In this study, we conducted an experimental assessment to predict intra-day values of a grid-connected photovoltaic power plant positioned within the Djelfa province, specifically in the Ain El Ibel commune of Algeria. The Support Vector Machine Regression is used. The experiments consist of investigating the best kernel function that fits accurately the processed data, in addition to the impact of the historical input layer length. Thus, the Polynomial kernel function, the Radial Basis Function, and the linear kernel function were individually tested. The results show that the Radial Bias function gives better results using a few historical inputs. However, when the historical input data becomes larger, the polynomial kernel function competes with the RBF. Despite this, the calculated mean average of the performance parameters reveals that the Radial Basis function gives slightly better statistical metrics (r = 0.8938, MAE =0.0690, RMSE = 0.0.0947, rRMSE=0.2635, NRMSE=0.0961) compared with the polynomial (r = 0.8886, MAE = 0.0702, RMSE = 0.0991, rRMSE=0.2748, NRMSE=0.0972). In this study, the linear kernel shows the weakest parameters (r = 0.8329, MAE =0.0918, RMSE = 0.1192, rRMSE = 0.3267, NRMSE = 0.1204).