Tsianikas, Stamatis. Microgrid expansion planning using simulation-based optimization and reinforcement learning. Retrieved from https://doi.org/doi:10.7282/t3-037x-6774
DescriptionThis dissertation provides an analytical framework for tackling the long-term microgrid expansion planning problem. In the wake of the highly electrified future that is ahead of us, the need for reliable and economical power supply will become more urgent than ever. The role of microgrids in fulfilling this need is expected to be highly crucial. While there is a lot of active research going on related to developing optimization models for such systems, the current work innovates by considering both economic and reliability aspects, as well as the stochastic nature of various components in the energy industry. Furthermore, the fact that the microgrids will be placed at the core of the future energy systems will naturally give birth to another important problem from the planning perspective; this problem concerns the derivation of optimal strategies when expanding the microgrids, both in storage and power capacity. The criticality of formulating systematic, analytical and novel methodologies to tackle this problem can be easily justified by considering the steady growth of load demand, the technological advancements continuously being made, and the high operating costs incurred in these processes. The research work that can be found currently in the literature lacks in considering several peculiarities of microgrids. Moreover, many proposed approaches fail to provide realistic and complex-enough formulations due to the incompetence of traditional solution techniques in handling problems of that scale. The current research work serves as the first attempt to formulate a highly detailed long-term expansion planning problem in microgrid setting and solve it using advanced artificial intelligence techniques. Towards this direction, a simulation-based approach is developed to determine cost-optimal battery sizing under preset reliability constraints, and a unified dynamic optimization framework is built and used to derive holistic optimal expansion strategies. Overall, the goal of the present research work is to provide novel baseline models that give a well-shaped structure to the stochastic problem of long-term expansion planning, while utilizing advanced machine learning tools and techniques.