Accurate long-term prediction of key parameters is of great significance for intelligent perception and optimized decision-making in industrial processes characterized by time-delay dynamics. However, industrial time-series data often exhibit complex characteristics such as nonlinearity, information sparsity, random noise, and non-stationarity. These factors make it challenging for existing predictive modeling methods to effectively decouple confounding factors and adapt to unknown uncertainties.

Addressing these challenges, a research team from the Digital Factory Department at the Shenyang Institute of Automation(SIA), Chinese Academy of Sciences(CAS), has proposed a novel long-term industrial time-series forecasting method. This approach integrates a multifrequency decoupling network with dual-stage shift segmental modeling to tackle the difficulties in long-term prediction of key industrial parameters, optimizing aspects including data decomposition, dynamic modeling, and distribution adaptation.

Long-Term Industrial Time-Series Prediction Modeling Based on Multifrequency Decoupling Network and Dual-Stage Shift Segmental Modeling(Image by the research group)

The methodology employs a multifrequency decoupling network to separate noise and high-frequency interference from the time-series data. For the intermediate-frequency components, the researchers developed a dual-stage shift segmental modeling technique, effectively mitigating the impacts of information sparsity and random noise. Furthermore, to enhance the network's learning capability and its ability to handle non-stationary data, the team designed a weighted position-direction alignment loss and a reversible dual-band conditional adaptive normalization mechanism. This work provides a new pathway for intelligent perception and optimized decision-making in industrial processes.

The research findings have been published in the IEEE Transactions on Industrial Informatics, a leading international journal in the industrial informatics field, under the title Multifrequency Decoupling Network With Dual-Stage Shift Segmental Modeling for Long-Term Prediction of Industrial Parameters. The first author of the paper is LYU Changyao, a master's student at the SIA. The corresponding authors are Professor ZHOU Xiaofeng and Associate Professor LI Shuai. This study was supported by the Key Project of the Major Research Plan of the National Natural Science Foundation of China, the Young Scientists Fund of the National Natural Science Foundation of China, and the Fundamental Research Program of the SIA.