Simple model predictive control strategies coupled to artificial intelligence-based disturbance modeling for microgrid energy management system

Abstract

Electricity generation still relies largely on fossil fuels; however, growing concerns about climate change, resource depletion, and increasing energy demand have accelerated the transition to cleaner alternatives. Renewable Energy Resources (RERs) offer advantages such as abundance, diversity, and low carbon dioxide emissions, but their intermittency and variability demand effective integration strategies. In this scenario, microgrids emerge as a promising solution by integrating generation, storage, and loads into modular systems capable of operating either grid-connected or in islanded mode. To ensure efficient and reliable performance, it is essential to adopt advanced control and optimization strategies, particularly in managing power exchange with economic objectives -- an operational layer known as the Energy Management System (EMS). In this context, this work proposes Model Predictive Control (MPC) strategies for EMSs focused on economic operation, including: (i) the development of a two-layer MPC based on Quadratic Programming (QP), whose results demonstrated improvements in numerical robustness, reduced computational costs, and faster optimization problem resolution compared to traditional Nonlinear Programming (NLP)-based MPC approaches; (ii) the application of the QP-based MPC strategy to an industrial-scale microgrid, together with the introduction of Artificial Neural Networks (ANNs) for disturbance modeling, such as renewable generation and demand, incorporating these forecasts into the control framework, with results showing greater robustness in decision-making under the operational variability imposed by disturbances; and (iii) the proposal of a single-layer, zone-based MPC strategy designed to operate within economically viable control regions defined by the microgrid operator, with a closed-loop stability proof based on Lyapunov theory, whose simulation analysis demonstrated that the cost function behaves as a Lyapunov function, ensuring controller scalability, recursive feasibility, and nominal convergence.