太陽光発電出力予測における時間推移のパターン抽出を用いた誤差分析手法

抄録

Information about forecasting errors of photovoltaic power generation (PV) output is useful in daily supply and demand operation of power systems for large-scale integration of PV. In the conventional error analyses, average is generally used which implies the summary of errors in a certain period. It does not represent actual transitions of errors that occurred every day and hence likelihood of future transitions with time cannot be grasped in the forecasting error. In this paper, we propose an error analysis method for PV output forecasting to grasp characteristics of error transitions such as the shapes and their occurrence probabilities. We enable such an analysis by extracting representative patterns in shape of error transitions in the following way. All the days are classified to some clusters by using a clustering method based on a similarity of shape of the error transition in that day, and then representative patterns are obtained by averaging the error transitions for each cluster. The numerical result shows that the k-means method, which is one of the clustering methods, is suitable for helping understand annual trends of error transitions. We also clarify how we should estimate the appropriate number of patterns of error time transitions.