16:45 〜 17:00
[MAG32-11] Enhancing Accuracy of PV Output Probabilistic Forecast through Synergistic Integration of Machine Learning and Ensemble Weather Prediction
キーワード：太陽光発電、確率予測、アンサンブル予測、機械学習、信頼区間
As the adoption of renewable energy sources such as photovoltaics (PV) increases, the use of PV output forecasts becomes essential for maintaining the balance between electricity supply and demand. However, since forecast errors are inevitable, probabilistic forecasts that can quantitatively evaluate these errors have become increasingly important. The proposed PV output probabilistic forecasting method is designed based on an analysis of the factors contributing to forecast errors. For stochastic uncertainties arising from the nonlinearity of atmospheric dynamics, ensemble weather prediction is applied, while machine learning is used to address epistemic uncertainties due to the imperfections of the forecasting method. Gradient boosting is employed as a machine learning technique. The proposed probabilistic forecast includes the median of the probability distribution, 50%, 90%, and 99.73% confidence intervals. The width of these confidence intervals dynamically changes according to daily weather variations. Validation of the forecasts over a year showed that the proposed method reduces the forecast error of time series forecasts compared to conventional methods using empirical power curves for PV output conversion. Additionally, the width of the confidence intervals in the proposed method is narrower than that of conventional methods, confirming an improvement in forecast reliability. Furthermore, the proposed method better represents the probability distribution of PV output, as evidenced by improved occupancy rates of actual PV output values within the probabilistic classes compared to conventional methods.