Data-driven adaptive improved just-in-time learning for industrial process soft sensor

作者：Zuliang Shen, Yuanyu Cao, Wei Cheng · 年份：2021 · DOI：10.1109/ichci54629.2021.00065 · 被引用次数：4 · 研究领域：Fault Detection and Control Systems、Advanced Control Systems Optimization、Advanced Algorithms and Applications

In the industrial process, data-driven soft sensor technology can obtain the information of key variables in real time. However, real industrial processes often show multimodal characteristics due to the influence of switching conditions. The traditional methods are all aimed at single-modal data modeling. Applying these methods directly to multi-modal data will result in a decrease in model accuracy. In this paper, improved just-in-time learning for the soft sensor is proposed to deal with multimodal characteristics. First, we select similar samples as training data sets from the historical database based on the new sample. Then, a neighbor similarity coefficient is designed to judge the similarity between samples, determining the number of samples in the training set adaptively. Finally, a partial least squares (PLS) model is established based on the selected training set data, which is used to measure the key variables for the new sample. The effectiveness of the proposed method is verified by an industrial example.