A Data-Driven Stochastic Multi-Objective Energy Management System for Sustainable Interconnected Microgrids, Integrating LSTM Forecasts and Hybrid Storage Systems

The research proposes an EMS for grid-connected microgrids which incorporates deep learning predictions, stochastic optimization, and multi-carrier storage systems to optimize cost of operations and greenhouse gas emissions. By jointly reducing operational costs and carbon emissions, the suggested approach will facilitate the transition to sustainable, resilient, low-carbon energy systems. Quick-acting battery storage and slow-reacting hydrogen are used for shaving peaks and balancing energy supply, while decision-making is robustified against the uncertainty associated with renewables through Latin Hypercube Sampling (LHS). The LSTM-based model forecasts 24 h ahead multi-step PV and wind generation outputs. Five algorithms, namely, GA, PSO, GWO, NSGA-II, and the Hippopotamus optimizer (HO), are evaluated for two different microgrid structures (MG1, MG2), using hypervolume metrics and a Wilcoxon rank-sum test conducted on 30 independent runs. The findings indicated that HO provides the best spread and dispersion of the Pareto frontier, along with highest hypervolume, being statistically similar to PSO (p > 0.05), while proving significantly superior to GA and NSGA-II, which get stuck within a highly inefficient region. GWO is overestimated in terms of Pareto convergence, since it is possible at one certain point only. This is impossible in real life, as proven through a peer-to-peer trading counterfactual study. An EMS is capable of allocating renewable excesses into batteries or hydrogen storage and optimizing the profits from grid trading during the 24 h periods. The findings prove HO and PSO algorithms to be reliable energy management solvers for microgrids.