<p>Global warming is causing an increasing number of severe weather events, and the current electrical infrastructures are unable to handle these rare but highly impactful disasters. To address highly unpredictable events with extensive state spaces during an extreme occurrence, standard model-based methods create optimization models to represent the effects on the power system. Nevertheless, the advanced models exhibit a significant level of computational complexity and lack the capacity for learning. This paper introduces a novel methodology for enhancing the resiliency of distribution systems by employing ensembled deep reinforcement learning. The proposed methodology reconfigures the system during outages to supply critical loads while maximizing the utilization of distributed energy resources (DER). This would enable the system to efficiently supply the maximum critical load by creating microgrids inside the system. Combine Resilient Microgrid Ensembled Deep Reinforcement Learning (RMG-EDRL) approaches with DERs to reduce penalties and DER running costs. The results show that our EDRL framework outperforms previous methods and may improve distribution system resiliency in unpredictable and dynamic operating situations. Ultimately, the suggested model was effectively tested on both a 33-bus and 123-bus distribution networks. The effectiveness of the proposed strategy was confirmed by the numerical findings.</p>