Abstract: In hot strip rolling processes， the methods of processing the actual data influence the selflearning precise of the model and further affect the product quality control precise. To solve this problem mentioned above， a multisample strategy of practical data acquisition was established and the high dispersion data were eliminated with the sample variation coefficient method， while the collected valid data synchronization was realized with the samples mapping method through which the selflearning source data with highly reliability were achieved that improved the validity of the selflearning of the model and the predict precise. This multisample strategy was applied to a tandem hot mill in China， the practical application results showed that the rolling force predicted precise was up 233%， which met the requirement of the automation gauge control system， and the product quality was improved.

Key words: tandem hot mill, finishing rolling, selflearning, multisample process, variation coefficient