A predictive and adaptive energy management framework for internet of things networks under dynamic and uncertain operating conditions

Energy management is a major challenge in Internet of Things (IoT) networks, particularly in large-scale and dynamic environments where node behavior, traffic patterns, and energy consumption are uncertain. Many existing approaches rely on static control policies or fully control-driven deep reinforcement learning (DRL) schemes, which may suffer from limited adaptability, sensitivity to parameter tuning, or unnecessary computational overhead under rapidly changing conditions. Consequently, achieving stable and energy-efficient operation remains an open problem. This paper proposes a predictive and adaptive energy management framework for IoT networks that explicitly incorporates behavioral uncertainty into the decision-making process. A hybrid forecasting model combining Prophet and long short-term memory (LSTM) networks is employed to capture both long-term trends and short-term temporal variations in node-level energy consumption. The deviation between predicted and observed energy behavior is interpreted as a dynamic behavioral signal rather than a fixed anomaly. This residual information is then integrated into a deep reinforcement learning–based decision layer to adapt energy management policies online without predefined thresholds or rule-based interventions. The effectiveness of the proposed framework is evaluated through simulations conducted in MATLAB and NS-3 under varying network densities and traffic dynamics. Performance is assessed in terms of average energy consumption, network lifetime, and energy usage stability. Results show consistent improvements over non-adaptive methods, rule-based strategies, and conventional DRL baselines, achieving approximately 3–4% reductions in average energy consumption and corresponding extensions in network lifetime, confirming the practical effectiveness of the proposed approach.