6-104
On the Use of an Ensemble-based Approach to Improve Forecast of Photovoltaic Power Generation with Support Vector Machines
○ガリ ダ シルバ フォンセカ ジュニアジョン・大関 崇・大竹秀明・髙島 工（産業技術総合研究所）・荻本和彦（東京大学）
In a previous case study it was shown that an ensemble-based approach, using only the training data, can yield forecast accuracies as high as the ones obtained with 1 year of spare data and the grid search in the problem of insolation forecasting. In this study we extend such application for the PV power forecasting problem evaluating the technique in hourly forecasts for 586 PV systems in 2009 and 756 PV systems in 2010. The results validated the use of the approach. They showed that in average more than 8% of reduction of the annual absolute forecast error can be achieved with the proposed approach. Regarding the rmse the reduction was smaller, 3.25%, but still consistent.