Because of climate change, the ageing power system infrastructure is under threat due to the ever-increasing intensity and frequency of high-impact, low-probability (HILP) events. Although, in most cases, these events are area-specific, the impact of such events, if unaddressed, can lead to cascading failures. Therefore, it is vital for the grid of tomorrow to not only be reliable but also be resilient in view of the broad inter-dependencies. Despite being a widely researched topic, the applicability of the concept of resilience, especially in power systems terms, is not a straightforward task due to the lack of consensus on a consistent definition, or a set of robust metrics. Therefore in this dissertation, an analysis of different definitions, frameworks, metrics, and enhancement techniques related to resilience proposed by multiple researchers and research organizations are discussed. Together, the aforementioned concepts set up the fundamental basis of this dissertation.Strengthening the existing system to withstand the extreme weather events (infrastructure resilience) or improving the operability of the system under emergency conditions (operational resilience) are the two important aspects of power system resilience improvement. While the infrastructure resilience improvement techniques are effective, due to the inherent characteristic of extreme weather events (to be spatio-temporal in nature), the benefits these improvement techniques provide are unevenly distributed which hinders the applicability of such techniques. Therefore, improving the operational resilience aspect of the distribution system under normal and emergency conditions is the primary focus of this dissertation. Microgrids (MGs) have emerged as one the solutions for improving operational resilience. By supplying the critical loads using localized generating resources under emergency operating conditions, MGs can improve the survivability aspect of system's resilience. MGs that are closely located geographically can be interconnected with each other forming a network of MGs called Networked Microgrids (NMGs). These clusters of MGs further enhances the operational resilience by improving critical load pickup through resources sharing under abnormal conditions. With an aim to improve the operability under uncertainties (during normal operating condition) and resilience by supplying critical loads (during emergencies), we introduce a novel Dual Agent-Based framework for optimizing the scheduling of DERs and loads within a NMG that leverages the field of advanced machine learning. A deep reinforcement learning (DRL) framework that aims to minimize operational and environmental costs during normal operations while enhancing critical load supply indices (CSI) under emergency conditions is developed in this dissertation. Modeling a robust reward function that provides a feedback to the agent regarding the best control actions is pivotal in DRL frameworks. Therefore, a multi-temporal dynamic reward shaping structure along with the incorporation of an error coefficient to enhance the learning process of the agents is designed. To appropriately manage loads during emergencies, we propose a load flexibility classification system that categorizes loads based on its criticality index. The scalability of the proposed approach is demonstrated through running multiple case-studies on a modified IEEE 123-node benchmark distribution network. We also test the proposed method with different DRL algorithms to demonstrate its compatibility and ease of application, whereas for validation we compared the results with the metaheuristic algorithms: particle swarm optimization (PSO) and genetic algorithm (GA). To gain a deeper understanding of the developed model, we conducted a sensitivity study. The key findings from this study align with the mathematical foundation of the approach outlined in this dissertation, providing further support.