Experimental Evaluation of Microgrid Energy Management Using Surrogate-Assisted Optimization on PHIL and Smart-Grid Systems

In this work, we present an integrated hybrid approach of multi-objective evolutionary decomposition algorithm for a microgrid-connected smart grid test-bench. The optimization technique was tested in a power hardware-in-the-loop setup, using different distributed energy resources. During evaluation, the optimization ran in real time. To reduce computational time, we added a surrogate-assistance to help find the feasible operational region and detect knee points for decision-oriented prioritization of candidate solutions. We focused more on human decision preferences instead of only choosing the best solution from the Pareto set, giving priority to decision-based knee points for dynamic constraints. The research integrates an OPAL-RT microgrid power hardware-in-the-loop platform with a Lucas-Nülle smart-grid laboratory system to investigate energy management system optimization under various operational modes. The proposed hybrid framework also incorporates a stochastic survival strategy to maintain both convergence and diversity of the search process. The method was evaluated in both grid-connected and off-grid modes, where the microgrid operates as the primary energy source for the smart-grid unit during off-grid mode. In off-grid operation, grid-forming objectives were applied, while grid-following objectives were used in grid-connected scenarios. In the grid-following mode, a selective physical replay mechanism was employed to reduce computational complexity while maintaining stable decision quality. In the results for grid-connected mode, it observed constrained feasible regions because of the limited renewable utilization and reliance on battery and grid support which shows the influence of measure system conditions. In the grid-forming scenario, the results reveal stable convergence behavior with consistent Pareto set quality across independent runs. Experimental validation demonstrates objective-dependent modeling capability for nonlinear system behavior such as battery stress, efficiently reduce thousands of candidate solutions to a small and robust Pareto set, and provide a practical decision-intelligence layer linking evolutionary optimization with real-time energy management systems.