Active load forecasting based on hybrid ensemble learning framework considering active-reactive coupling

Le résumé fourni par la source

Conventional load forecasting methods face dual challenges in handling time-frequency modal interference and neglecting active-reactive power interdependencies under increasingly stochastic loading conditions. This study develops a CEEMDAN-BiLSTM-Adaboost composite framework that explicitly incorporates active-reactive coupling characteristics. The methodology initiates with Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) processing to decompose active load profiles into intrinsic mode functions (IMFs) and residual components through noise-assisted ensemble decomposition. Bidirectional Long Short-Term Memory (BiLSTM) networks are subsequently employed as base learners, with reactive power time-series embedded as coupled input features within the multivariate predictor architecture. An AdaBoost meta-learner dynamically optimizes and aggregates these weak learners via iterative error-corrected weighting, followed by IMF reconstruction to generate final active load predictions. Experimental validation using residential smart meter datasets confirms the framework’s enhanced capability in capturing coupled load dynamics compared with conventional prediction paradigms.

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