Adaptive Stacking ensemble model driven by multi-source data fusion for energy consumption prediction in forging production line

The Forging Production Line (FPL) is a representative energy-intensive industrial system where accurate energy consumption prediction is essential for effective energy management. However, nonlinear load fluctuations and frequent product specification transitions introduce significant dynamic complexity, posing substantial challenges to traditional single-model approaches and conventional ensemble methods that apply uniform architectures across all operational conditions. To address these challenges, this study develops a novel adaptive ensemble framework that systematically accommodates the phase-specific characteristics of multi-phase industrial processes. The framework divides FPL operations into four distinct phases according to their energy characteristics and constructs phase-specific feature spaces. An Adaptive Selection Mechanism (ASM) dynamically determines optimal base learner combinations for each phase using multiple evaluation metrics, overcoming the limitations of static ensemble architectures. Bayesian Optimization is integrated to enhance phase-specific prediction performance. Experimental validation demonstrates that the framework achieves error reductions of 70.81 %, 83.96 %, 84.48 %, and 78.43 % across all four operational phases compared with the Stacking model in single-step predictions. For multi-step forecasting, the framework maintains robust performance with minimal error accumulation across 2-step and 3-step predictions. Comparative evaluation against advanced models including Informer, Autoformer and TimesNet confirms competitive advantages across all phases. Feature importance analysis reveals that temperature-related variables are dominant contributors to energy consumption. Cross-production-line validation across multiple prediction horizons confirms superior generalization capability. The high-precision predictions enable substantial energy cost savings through improved production planning, equipment scheduling, and process optimization, contributing to intelligent energy management in forging operations.