A Novel Adaptive Just-in-Time Learning Soft Sensor Modeling Method With a Density-Based Data Update Strategy

With the rapid development of information technology, data-driven soft sensors are often used in industrial process quality control. In recent years, soft sensor modeling method based on just-in-time learning (JITL) has been widely studied and applied due to its strong adaptability to time-varying characteristics of processes, simple modeling principles, and easy implementation. However, when the working conditions of the production process migrate due to various reasons such as sensor drift, catalyst deterioration, and raw material changes, the historical data used for modeling may not cover the current working condition, and the performance of the JITL model will also sharply decline even fail. Therefore, this article proposes a density based modeling data adaptive update management mechanism under the JITL framework to effectively enhance the adaptive tracking ability of soft sensors for process condition migration, thereby maintaining the longterm effectiveness of the model in practical use. Finally, the advantages of the proposed method were verified through numerical simulation and comparative analysis.