Accelerating Transformer Fault Diagnosis Method Based on Data and Mechanism

This paper primarily addresses the exploratory needs of foundational and forward-looking technologies in power artificial intelligence (AI) brain-inspired computing. It proposes a power AI brain-inspired computing architecture aimed at mastering learning methods for power brain-inspired models based on online autonomous collaborative evolution. The goal is to develop a prototype system for power AI brain-inspired computing to enhance the reasoning capabilities of power AI models while reducing their dependence on large computational resources and large parameter sets. To address the aforementioned issues, this paper summarizes the relevant research progress in foundational and forward-looking technologies of power AI brain-inspired computing, including brain-inspired cognitive intelligence, AI command emergence, and autonomous learning. Furthermore, this paper proposes specific implementation plans, including research on the co-design of software and hardware for the power AI brain-inspired computing architecture, research on learning methods for power brain-inspired models based on online autonomous collaborative evolution, and research on key technologies for the power AI brain-inspired computing prototype system.