近期，我院本科生林子軒（第一作者），教師范偉（通訊作者）等的研究成果Robust Probabilistic Sparse Identification of nonlinear dynamics for industrial anomaly detection在《Journal of the Taiwan Institute of Chemical Engineers》（IF=6.3）上發(fā)表
論文簡(jiǎn)介如下：
Industrial processes often exhibit nonlinear and time-varying behaviors, which makes reliable monitoring essential for safety and efficiency. Traditional statistical methods rely on Gaussian noise and stationary assumptions, leading to poor robustness under disturbances and outliers.To address this issue, a Robust Probabilistic Sparse Identification of Nonlinear Dynamics (RPSINDy) is proposed in this work. It combines sparse regression with probabilistic state-space modeling, introduces a Gaussian–Student’s t mixture distribution to capture heavy-tailed noise, and employs EM with particle filtering for parameter estimation and inference. Three monitoring indices are designed to evaluate abnormal operating conditions and dynamic deviations.Case studies on a three-phase flow facility and a marine diesel engine show that RPSINDy achieves earlier and more accurate fault detection than traditional methods. The results highlight its practicality as a robust and interpretable tool for monitoring complex industrial systems.
