حميدرضا مومن

With the growth of natural calamities in recent years, there has been a heightened focus on enhancing resilience of power distribution networks against such incidents. Following a widespread blackout, Distributed Generation Resources (DERs) such as Diesel Generators (DGs), renewable energy sources, Energy Storage Systems (ESSs), and Mobile Energy Resources (MERs) can be coordinated to restore the loads of the distribution network. Dynamic microgrid formation is a promising solution for resilience enhancement by providing a flexible framework to utilize DERs. Due to the large decision space and inherent uncertainties of distribution network, conventional model-based approaches suffer from long response times and heavy computational burdens. This dissertation employs a Deep Reinforcement Learning (DRL) approach to model the dynamic microgrid formation process, routing of MERs, and scheduling of ESSs. DRL offers a significantly faster response and incorporates uncertainties associated with load demand and renewable resources into the proposed data-driven approach through the training process. In this dissertation, DRL algorithms such as DQN and TD3 are applied to manage discrete and continuous decision spaces in the microgrid formation problem, respectively. To eva‎luate the proposed approach, multiple simulations were conducted on the modified IEEE 37-bus and 123-bus test networks. The results demonstrate that the proposed method outperforms conventional mathematical programming-based methods and other DRL-based algorithms in terms of restored load and computational time.

تاب آوري شبكه توزيع , تشكيل پوياي ريزشبكه ها , سيستم هاي ذخيره ساز انرژي , منابع انرژي متحرك , يادگيري تقويتي عميق

Resilience of distribution network , Dynamic microgrid formation , Energy storage systems , Mobile energy resources , Deep reinforcement learning

Hamidreza Momen

Shahram Jadid