Multi-Objective Physics-Guided Recurrent Neural Networks for Identifying Non-Autonomous Dynamical Systems*

While trade-offs between modeling effort and model accuracy remain a major concern with system identification, resorting to data-driven methods often leads complete disregard for physical plausibility. To address this issue, we propose physics-guided hybrid approach non-autonomous systems under control. Starting from traditional physics-based model, is extended by recurrent neural network trained using sophisticated multi-objective strategy yielding physically plausible models. purely fail produce satisfying results, experiments conducted on real data reveal substantial improvements our compared model.