A Novel Soft Sensing Method for Transient Processes Regression Utilizing Locally Weighted PLS

This paper develops a novel soft sensing method using locally weighted partial least squares (PLS) for transient processes regression. Industrial transient processes cannot be described using merely one model and therefore the regression model should be updated according to the online system condition. Different from previous just-in-time (JIT) methods using Euclidean distance, a supervised approach is proposed involving both process data X and quality data Y to finish sample selection tasks. The locally weighted PLS is adopted to depict the relation between X and Y. The performance of the novel soft sensing structure is validated by an industrial process.