MIDPhyNet: Memorized infusion of decomposed physics in neural networks to model dynamic systems

Integrating simplified or partial physics models with data-driven machine learning is an emerging concept targeted at facilitating generalizability and extrapolability of complex system behavior predictions. In this paper, we introduce a novel based fusion model MIDPhyNet that decomposes, memorizes, integrates first principle physics-based information models. the output decomposed into Intrinsic Mode Functions (IMFs), which are then infused to Memorization Unit generate embedded vectors. A Prediction synthesizes all data prediction results. We test performance on modeling dynamic systems such as inverted pendulum under wind drag. The results clearly demonstrate benefits our hybrid architecture over both purely state-of-art in terms extrapolability. architecture’s superiority most significant when trained sparse sets general, provides generic way explore how physical can be