Toward Improved Load Forecasting in Smart Grids: A Robust Deep Ensemble Learning Framework

Abstract

This paper presents an advanced deep ensemble learning framework for short-term load forecasting (STLF). The refined deep ensemble model (DEM), complemented with a flexible error compensation (FEC) strategy, is introduced to improve both forecast accuracy and reliability. To address the challenges of ensemble pruning and aggregation (EPA), a worst-case (WC) robust approximation problem is formulated to accommodate the inherent uncertainty in predictions. The solution to this multifaceted problem employs a sophisticated methodology, integrating cardinality minimization and the augmented Lagrangian algorithm. Real-world empirical studies substantiate the enhanced STLF attained by the proposed framework.