SIFA: A two-stage adaptive ensemble framework for solar irradiance forecasting using a wrapper-based feature...

Area of Science:

• Renewable Energy Systems
• Machine Learning Applications
• Data Science

Background:

• Accurate solar irradiance (SI) forecasting is vital for effective solar energy system management.
• Existing SI forecasting models struggle with nonlinear data, high computational demands, and optimal feature selection.
• These limitations impact the stability and efficiency of photovoltaic (PV) power plants.

Purpose of the Study:

• To propose a novel multi-stage forecasting approach, SIFA, for enhanced solar irradiance prediction.
• To improve the accuracy and efficiency of PV power plant operations.
• To address the limitations of existing SI forecasting models.

Main Methods:

• A hybrid feature selection strategy combining Random Forest (RF) and Sequential Forward Selection (SFS) to identify informative features.
• Optimization of RF performance using an enhanced manta ray foraging optimizer (IWMRFO) with chaotic maps for improved exploration and convergence.
• An ensemble approach combining Huber Regressor (HR), Extra Trees (ET), and Extreme Gradient Boosting (XGB), with weights optimized by IWMRFO for adaptive prediction.

Main Results:

• The SIFA approach demonstrated lower average forecasting errors compared to competing models across three datasets (San Diego, Islamabad, NASA SI).
• The optimized IWMRFO effectively balanced exploration and exploitation, accelerating convergence and avoiding local optima.
• The adaptive ensemble approach enhanced predictive accuracy and maintained stable generalization capability.

Conclusions:

• SIFA offers a robust and accurate solution for solar irradiance forecasting.
• The proposed method effectively models nonlinear data and optimizes feature selection, outperforming existing models.
• SIFA is a strong alternative for improving the accuracy and efficiency of solar energy management systems.