Immersive learning (IL) technologies, including virtual, augmented, and mixed reality, have been increasingly adopted for training in energy systems, where high-risk, low-frequency events and hazardous operating environments limit the feasibility of real-world training. Over the past two decades, such approaches have been explored across nuclear, power grid, oil and gas, and renewable energy domains, yet the literature remains fragmented across applications and training objectives. This survey reviews IL for energy training from 2000 to 2025, focusing on four core priorities: (i) operator training, (ii) maintenance and field services, (iii) safety and emergency response, and (iv) education and workforce development. We introduce a cross-domain taxonomy that organizes prior work by energy domain, immersive modality, pedagogical approach, and evaluation maturity; enabling comparative analysis across sectors. The survey highlights dominant design patterns as well as persistent gaps in evaluation rigor, particularly with respect to transfer of training to real-world performance. We conclude by outlining research directions to support evidence-based, scalable, and trusted immersive training in energy systems.