Brain-inspired memory architecture for condition monitoring based on hippocampal-neocortical complementary learning

This paper tackles the critical challenge of catastrophic forgetting and inefficient learning in artificial intelligence models processing continuous, non-stationary data streams. Inspired by neurobiological mechanisms, specifically the Complementary Learning Systems (CLSs) theory involving the hippocampus and neocortex, we propose a novel brain-inspired biomimetic memory system. The core innovation integrates dimensionality reduction techniques—covariance decomposition and low-dimensional mapping—for efficient feature extraction from long spatiotemporal-scale information flows, with a biomimetic learning/forgetting mechanism grounded in CLS principles. Evaluated on real-world power plant operational data, the proposed system demonstrates robust performance against noise and fluctuations, validating the effectiveness of the bionic learning/forgetting mechanism.