Optimal Data-Driven Control of Embedded Micro-Grids in Developing Countries

Over a billion people worldwide have no access to electricity; and many that do, have sporadic access marred by frequent grid outages, particularly in Asia, Africa, and Latin America. The relatively affluent in these countries circumvent the problem by installing grid-connected backup systems, which provide uninterrupted supply. However, these backup systems increase losses and burden the local electricity grid. For instance, estimates show that in Pakistan, between 2%-3% of annual energy generation is lost through battery charging. In addition, due to these elite backup systems, over 10% of all distribution transformers are overloaded. This overload causes needless expenditure that developing countries can not afford, as well as greenhouse gas emissions which accelerate climate change. This paper investigates various data-driven optimal control strategies to improve backup systems' operational efficiency. Preliminary results show that by employing a smart controller, household energy consumption can be reduced by 3%-7%, while grid overload costs can be reduced by almost a quarter. Furthermore, model-Assisted reinforcement learning controller is able to approach the performance of data-driven model predictive control at a fraction of the computational cost while far outperforming model-free reinforcement learning algorithm. \textcopyright 2019 IEEE.