| Abstract: |
Smart Microgrid Energy Management Systems (MEMS) are vital due to the rapid growth and increasing complexity of modern power grids, as well as the plethora of distributed energy resources (DERs). This empirical research presents a comparison of the performances offered by artificial intelligence (AI) and automation techniques for real-time energy dispatch, demand-side management, and fault resilience against microgrid schedule optimization applications, specifically reinforcement learning (RL), deep neural networks (DNN), and fuzzy logic controllers [11]. The research uses primary data collected from a hybrid solar-wind-battery microgrid testbed hourly over 12 months and supported with simulation validated datasets to assess key performance metrics [energy cost savings, grid stability index, renewable energy penetration/so-called curtailment ratio (CR), response latency of run-time dispatch mechanism and battery state-of-health (SoH)]. Five quantifications yield systematic insights to demonstrate that AI MEMS can reduce operational energy costs up to 34.7% and improve renewable utilization by 28.3%, outperforming benchmarks based on conventional rule-based control strategies. Statistical regressions support strong correlations between AI algorithm selection and system performance indices (R² = 0.91, p < 0.001). The results prove that multi-agent deep reinforcement learning processes can excel over classical optimization methods in dynamic load-balancing environments. The existing literature is presented in relation to proposed results to further clarify how this study advances the domain beyond previous work. The study concludes with practical implications for grid operators and policymakers looking to deploy sustainable microgrids. |
