Un quadro di gestione energetica predittiva e resiliente per reti intelligenti che integrano sistemi IoT intermittenti

Energy systems are transitioning from centralized, deterministic infrastructures toward distributed, data-driven, and increasingly autonomous ecosystems. While renewable and energy-harvesting sources improve sustainability, they also introduce intrinsic variability and uncertainty that threaten reliability across scales, from large smart grids to intermittently powered IoT devices. Addressing these challenges requires more than reactive control; it demands predictive intelligence that anticipates energy dynamics and adapts system behavior before disruptions occur. This thesis proposes the Predictive Energy-Aware Paradigm (PEAP), a conceptual framework that unifies forecasting, power modeling, and predictive checkpointing as complementary mechanisms for anticipatory energy management in heterogeneous systems. PEAP conceptualizes energy management as a predictive–adaptive feedback loop that links foresight, quantification, and action under a unified control logic. In this study, the macro- and micro-layers of PEAP were experimentally validated, while the meso coordination layer, which connects the two domains, remains theoretical and is identified as future work. At the macro level, data-driven forecasting models, including LSTM, RNN, and Transformer architectures, were developed and evaluated using real-world data from an industrial partner in Bologna, Italy. The results demonstrate how predictive learning enhances load scheduling, renewable integration, and proactive grid management. At the micro level, predictive intelligence was extended to solar-powered, intermittently powered IoT nodes using LightGBM and Random Forest models to forecast short-term battery voltage and detect drops below 400 V. These forecasts enable a predictive checkpointing mechanism that safeguards computational state before energy depletion, reducing unnecessary checkpoints and improving system uptime. The meso layer is defined as a coordination tier where predictive insights can be shared between macro and micro systems, enabling federated, cooperative, and privacy-preserving learning. Overall, PEAP demonstrates how foresight-driven decision-making enhances efficiency, reliability, and autonomy while remaining bounded by uncertainty and supporting human-supervised control. The framework provides both theoretical and experimental foundations for next-generation cyber-physical energy systems.