Deep Residual MLP Networks: Enhancing Precision and Reliability in Energy Forecasting

Le résumé fourni par la source

Despite the long-standing success of Multilayer Perceptrons (MLPs) across diverse applications, increasing their depth often introduces overfitting and gradient degradation. To overcome these limitations, this paper proposes a novel hybrid architecture that synergistically integrates MLPs with Residual Networks (ResNet). Specifically, MLPs serve as nonlinear mapping functions within ResNet blocks, while skip connections preserve gradient flow to enable stable training in deeper networks. The model is optimized using the Adam algorithm for its rapid convergence and further enhanced through systematic hyperparameter tuning via grid search. Comprehensive evaluations are performed on three critical energy forecasting domains: electricity demand, petroleum products, and renewable energy generation, with comparisons against 10 state-of-the-art models. The proposed framework demonstrates superior predictive accuracy, achieving a mean absolute percentage error (MAPE) of 4.495% in petroleum demand forecasting, significantly outperforming all baseline methods. These results underscore the model’s robustness and practical relevance for real-world energy forecasting applications.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.