Reinforcement‐learning‐based load shedding and intentional voltage manipulation approach in a microgrid considering load dynamics

Abstract

In the case of a power mismatch in an island microgrid (MG), sharp variations occur in the frequency and voltages due to the low inertia and stochastic behaviour of distributed generation units and loads. Load shedding (LS) is the last corrective measure to achieve frequency regulation. A Q-learning-based under-frequency LS (QLLS) approach is proposed, in which the MG model is considered a grey box. Hence, the dependence of the proposed method on the grid model is minimized. The presented approach considers four main characteristics of the loads, including priority, dynamics, outage cost, and daily variations. In addition, to reduce the amount of the shed load, a supplementary intentional voltage manipulation (IVM) algorithm is presented that exploits the voltage dependency of loads as leverage to help frequency regulation. The proposed QLLS/IVM uses the observable MG states to find a corrective course of action to achieve frequency and voltage stability while keeping the under-frequency load shedding price as low as possible. The performance of the proposed QLLS/IVM approach is tested on the modified IEEE 37-node MG, and the results show feasibility, fast and optimal performance of the proposed algorithm.