Real-time energy management optimization using model predictive control on a microgrid demonstrator

This paper presents the implementation of a real-time control algorithm based on model predictive control and neural network load forecasting for microgrid optimization. At each stage, the algorithm updates the optimal trajectory for system state in order to minimize operations costs while ensuring instantaneous power balance and physical constraints respect. The proposed framework is implemented on a semi-physical demonstrator including residential load, electric vehicles, solar panels and battery storage to assess viability and optimality of control policies. In order to assess control efficiency and influence of economic parameters, several scenarios are run in parallel on the semi-physical platform. The robustness against forecast error of the proposed control architecture is confirmed through sensitivity analysis with various error distributions.