The twin threats of climate change and cyber-attacks threaten the electrical grid with power outages. Since the electrical grid is a safety-critical system, the consequence of such outages can be great, including risk of injury or death (such as during heat wave outages) and economic loss as many businesses depend on electricity to operate. A resilient grid is required to prevent, withstand and recover from these High Impact Low Probability (HILP) events. However, the simultaneous need to rapidly de-carbonize the energy system is leading to an increased penetration of intermittent renewables, and insufficient storage to compensate for them. Consequently, the challenge of operating the grid to supply power to end users under all circumstances has grown more difficult. In this research project this challenge is addressed in part through data-driven methods, such as Deep Reinforcement Learning (DRL), which can make use of measurements from an increasingly digitized grid to formulate effective control strategies that improve grid resilience. This includes a focus on safety, interpretability and explainability - since data-driven models that behave erroneously in edge cases, are opaque to scrutiny from engineers and do not explain why a decision was made cannot be deployed in a safety-critical system such as the electrical grid.