Cooperative Modeling of Inner Dynamic Relationships for Predicting Key Quality Indicators in Process Industry

Accurate modeling of dynamic relationships within high-dimensional data is beneficial in effectively predicting key quality indicators (KQIs) in process industry. However, existing methods often overlook the internal dynamic relationships of process variables, focusing solely on their interactions with KQIs. This limitation may result in the extraction of suboptimal dynamic latent variables (DLVs), thereby decreasing modeling performance. To address this issue, a novel cooperative inner dynamic latent variable regression (CiDLVR) algorithm is developed, which integrates the inner dynamic relationships (IDRs) of process variables with their interactions with quality variables. By cooperatively modeling these IDRs, the CiDLVR method strengthens the generalizability of the derived DLVs. Finally, experimental results on the debutanizer column demonstrate that the proposed approach outperforms traditional methods, underscoring its potential and superiority.