Distributed DBN-HSBoost Model for Hot Rolling Process Operating Performance Assessment with Partial Communication

The traditional hot rolling process (HRP) monitoring technology is confronted with monumental challenges in the face of high standard safety and quality requirements. How to evaluate the operating performance in real time and trace the causes of non optimal operating performance has become a new hotpot. Considering the extremely imbalanced and small minority (EISM) data problem in the FMP, a distributed operating performance assessment method with partial communication is proposed by utilizing deep belief network and hybrid sampling boosting (DBN-HSBoost). First, according to the process knowledge, variables of the HRP are divided into several sub-blocks, and the communications of different sub-blocks are also considered. Then, DBN is used to extract the features in each sub-block, which is used as the input of the HSBoost algorithm to establish the operating performance assessment model. When the process is in non-optimal condition, a difference index is calculated to trace the non-optimal variables. Finally, the effectiveness of the proposed method is verified by the process data from a real HRP.