SAE-MRHL: a Self-Adaptive Explainable Meta-Reinforcement Hybrid Learner for remaining battery life prediction in IoE systems

The rapid evolution of the Internet of Everything (IoE) has accelerated the deployment of battery-powered sensors for intelligent environmental monitoring and decision-making. In dynamic marine environments, accurate prediction of Remaining Battery Life (RBL) is critical to ensure system reliability while minimizing maintenance costs. However, conventional estimation models often fail to capture nonlinear discharge behavior and temporal drift induced by environmental factors such as temperature, turbidity, and salinity. To address these challenges, this study proposes SAE-MRHL, a Self-Adaptive Explainable-Meta Reinforcement Hybrid Learner that integrates ensemble regression, Long Short-Term Memory (LSTM) networks, and a TD3-based Adaptive Learning controller based on the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm. The framework dynamically adapts ensemble weights and learning rates in response to environmental drift while incorporating explainability through entropy-based feature importance, SHapley Additive exPlanations (SHAP) analysis, and uncertainty quantification. Experimental evaluation demonstrates that SAE-MRHL achieves superior predictive performance, with a coefficient of determination ( R 2 ) of 0.9383, Root Mean Squared Error (RMSE) of 0.1744, and Mean Absolute Percentage Error (MAPE) of 0.57%. Comparative analysis against baseline models, including LSTM, eXtreme Gradient Boosting (XGBoost), and Random Forest (RF), confirms significant improvements in accuracy, adaptability, and robustness under non-stationary conditions. Overall, SAE-MRHL provides a unified, drift-aware, and interpretable framework for intelligent energy management in IoE systems, with strong potential for deployment in marine monitoring, smart cities, and autonomous industrial environments.