Abstract
Traditional controllers frequently perform poorly when managing complex industrial systems because they must balance multiple goals like cost, pollution, and efficiency. For higher performance and efficiency, model predictive controller addresses this but necessitates efficient cost function tuning, which is currently frequently enhanced utilising metaheuristic algorithms. This work aims to develop and validate a data-driven weight optimisation method for multivariable model predictive controller in a DC microgrid comprising photovoltaic panels, battery, supercapacitor, grid, and load-by systematically balancing control effort and accuracy. Four algorithms-particle swarm optimization, genetic algorithm, pareto search, and pattern search-selected for their capacity to resolve complex, multi-objective problems are used in this study's automated, set-point tracking-based weight optimisation approach. The results show that incorporating parameter interdependency reduces genetic algorithm's power load tracking error from 16% to 8%, while particle swarm optimization achieves an error of under 2% even without considering interdependency. Fast convergence and trade-offs are supported by pareto and pattern search, but they are less responsive to sudden changes. These methods provide a feasible, repeatable way to boost model predictive controller performance.