面向工业过程软测量建模的概念漂移检测综述

基于数据驱动的软测量模型广泛用于工业过程中产品质量与环保指标等难测参数的在线测量,该过程中存在的概念漂移问题易导致模型精度下降.如何有效识别过程概念变化并精准检测漂移样本是提高模型测量性能的关键.本文总结并分析目前漂移检测的研究思路与进展,为面向工业过程软测量的漂移检测算法提供设计指导.首先,介绍了概念漂移的通常定义与其在工业过程中的表现形式;然后,从检测依据与检测对象两个视角分析了目前具有代表性的检测方法;接着,讨论了这些算法的技术特点和当前工业领域的研究难点;最后,展望了未来的研究方向.

Overview of concept drift detection for industrial process soft sensor modeling

Data-driven soft sensor models are widely used for online measurement of difficult-to-measure parameters such as product quality and environmental protection indicators in industrial processes, and the concept drift in this process will lead to a decrease in model accuracy. Effective recognition of process concept changes and accurate detection of drift samples are the keys to improving model measure performance. This paper summarizes and analyzes the current research ideas and progress of drift detection, and provides design guidance for drift detection algorithms for industrial soft sensor modeling. First, the general definition of concept drift and its manifestation in the industrial process are introduced. Then, the current representative research methods are analyzed from the perspective of detection object and detection basis. Next, the technical characteristics of different algorithm strategies and the current research difficulties in the industrial field according to the literature are discussed. Finally, suggestions for future research directions are given.