Review of Concept Drift Detection Method for Industrial Process Modeling
Zijian Sun, Jian Tang, Junfei Qiao, Chengyu Cui · 2020
With the advent of big data era in industry, data-driven modeling methods have been applied widely. The concept drift problem in industrial process modeling has also attracted widespread attention. However, the current research on concept drift focuses on classification tasks and computer fields, with less work on regression modeling of industrial processes. Aiming at the above problems, this paper summarizes the concept drift detection methods for industrial process modeling, and guide for solving this problem. First, the general definition of concept drift and its existence in industrial processes are introduced. Then, the existing drift detection technologies based on process variables, based on prediction error of difficulty-to-measure parameter and based on combine multiple factors are addressed. Thirdly, these methods are discussed, and some research difficulties are given out. Finally, the conclusion and the future research directions for the existing concept detection difficulties are presented.