Novel deep deterministic policy gradient technique for automated micro-grid energy management in rural and islanded areas

Abstract

The microgrid enhances power grid reliability, resiliency, and sustainability, particularly in rural and islanded areas with limited manual network management. However, microgrid energy management systems (EMS), especially in islanded mode, require precise and reliable techniques to prevent severe blackouts/brownouts. This paper presents a novel deep deterministic policy gradient (DDPG) algorithm to schedule EMS for the autonomous microgrid in real-time. Our solution utilizes deep reinforcement learning (DRL) to converge model-free, sequential, random, and continuous characteristics of the microgrid. Additionally, we use reward shaping and transfer learning attachment to DDPG to support microgrid performance restrictions and minimize load shedding during peak hours. This solution offers an efficient training process comparable to other DRL techniques in simplicity, less computation, and supporting future system extension. Residential Gasa Island microgrid profile characteristics have been selected and tested to examine the proposed approach. Results demonstrate the high efficiency and accuracy of the proposed technique compared to existing methods.