Enhanced load forecasting for distributed multi-energy system: A stacking ensemble learning method with deep reinforcement learning and model fusion

Accurate multi-energy load forecasting for distributed multi-energy systems is facing challenges due to the complexity of multi-energy coupling and the inherent stochasticity. In this regard, a novel stacking ensemble learning model based on reinforcement learning (RL) and model fusion is proposed. First, feature selection is performed using the maximal information coefficient (MIC), load data is decomposed and reconstructed through the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and sample entropy (SE). Subsequently, the fused models with strong predictive capabilities are selected as base learners, and the RL based on deep deterministic policy gradient (DDPG) with excellent adaptive learning ability is selected as meta-learner. Next, the hyperparameters of the base learners are optimized using an improved arctic puffin optimization (APO) algorithm. To reduce overfitting and improve the generalization ability of the model, k-fold cross-validation is adopted to construct the model. Tests on real-world datasets demonstrate that the proposed method achieves smaller prediction errors, enhanced robustness and reliability. Moreover, through careful base learner selection and the utilization of RL as the meta-learner, the model achieves up to a 1.53 % improvement in coefficient of determination, a 36.09 % increase in residual prediction deviation, and a 102.96 % reduction in rooted mean square error.