• Proposed a sense-to-act full-process artificial intelligence-driven smart energy management system • Elaborated on key artificial intelligence technologies in three typical extreme scenarios • Identified major deployment challenges for artificial intelligence-driven smart energy management system • Offered forward-looking perspectives on the evolution of artificial intelligence-driven smart energy management system Energy applications in extreme environments, such as polar and deep-sea exploration, impose unprecedented demands on reliability and resilience. Traditional energy management systems often face formidable challenges in these scenarios, including extreme temperature fluctuations, scarce communication links, and highly nonlinear dynamics, rendering them inadequate for sustained operation. In such critical domains where reliability equates to survival, conventional deterministic approaches fall short in addressing these complex and uncertain conditions. Consequently, developing a self-evolving and adaptive smart energy management system (SEMS) capable of handling complex uncertainties has emerged as a pivotal frontier for bridging the gap between extreme scenario requirements and technological limitations. Accordingly, this perspective innovatively proposes the holistic framework of SEMS towards extreme application scenarios, integrating perception, learning, prediction, optimization, and control. Following this, the advancements of applying artificial intelligence (AI) in energy management across extreme applications, including unmanned underwater vehicles, deep-sea exploration equipment, high-power industrial energy systems, and research stations in remote environments, are systematically reviewed for the first time. Then, this perspective provides a forward-looking analysis of the key challenges in the field and outlines targeted future research directions, such as enhancing model interpretability via physics-informed machine learning, developing hardware-aware edge intelligence for communication-constrained environments, and resolving data privacy dilemmas through federated learning etc. In summary, through an in-depth analysis of framework, research progresses, and future perspectives, this paper aims to provide systematic theoretical support and practical guidance for the SEMS towards extreme scenarios.
Jiang et al. (Sun,) studied this question.