A novel robust adaptive decomposition approach for solar energy potential using atmospheric transparency and UV radiation indicators

Abstract

In daily life, interest in solar radiation and related climatic factors has increased due to their impact on energy production, food security, agriculture, the ozone layer, and various industrial applications. A new Robust Adaptive Decomposition (RAD) method is introduced in this study to improve the forecasting precision of solar radiation time series data. With an adaptive weighting mechanism that reduces the impact of outliers and large deviations, the RAD method successfully decomposes high-frequency noise from significant signal features. Three solar radiation datasets of different types have been used in the research: ALLSKY_SFC_UV_INDEX (UV index), ALLSKY_KT (clearness index), and ALLSKY_SFC_SW_DWN (daily all-sky surface shortwave downward irradiance). Forecasting precision of the proposed RAD method is tested through cross-validation over three folds and contrasted with ARIMA, LSTM, and hybrid models like VMD-ARIMA and VMD-LSTM. It is hybridized with both conventional and deep learning models (ARIMA and LSTM). The results show RAD-LSTM to have the lowest error rates in Fold 2, whereas RAD-ARIMA performs well in comparison to all the other models on actual datasets, especially in Fold 1 and Fold 3. The ability of RAD-based models to decompose noisy time series into clean, stationary intrinsic mode functions (IMFs) has made them superior as the process becomes easier to model and increases prediction precision. Performance metrics like MAE, RMSE, and MAPE validate the proposed method's robustness and adaptability. The results show RAD's potential as a powerful tool for denoising and improving forecasting precision with far-reaching implications for climate-sensitive decision-making, agriculture, and energy planning.