Chemical process modelling using the extracted informative data sets based on attenuating excitation inputs

Background: Operation designs on rapid developed advanced chemical processes require proper models and parameter identification of these models needs input-output data sets with high information content. The optimal parameter identification experiment requires additional persistently exciting input to excite the chemical process for informative data sets, which ensure the process dynamic information. However, the optimal identification experiment with persistently exciting input is a time-consuming, high-cost task which may interfere with the process operation. As the requirement of low-cost identification, these experiments cannot be applied, some limited external excitations or extracted informative segments from historical data can support chemical process modeling.Methods: Attenuating excitation is a typical limited excitation can be used as identification experiment input. This paper proposed the identification experiments using informative data sets composed of attenuating excitation inputs, designed by users or extracted from historical process data. A goal-orientated hybrid algorithm based on Expectation -Maximization method and Exhaustive method is proposed to achieve rapid and effective infor-mative segments selection and obtain required parameter estimates. This method also can be used to verify the theorical attenuating excitation inputs design.Significant findings: A numerical case and an ethylene distillation column process demonstrate the feasibility of identification with attenuating excitation inputs and the robustness of the proposed algorithm. The result is capable for attenuating excitation input design or historical informative data mining for chemical process modeling, which has less or no impact and cost on chemical normal operations, decreases the process infor-mation waste and improves the data utilization.

Automated summary

Proper models and parameter identification are required for operation designs on rapid developed advanced chemical processes, and these models need input-output data sets with high information content. The optimal parameter identification experiment, which requires additional persistently exciting input, can ensure the process dynamic information. However, this experiment is time-consuming and high-cost, and may interfere with the process operation. In low-cost identification, limited external excitations or extracted informative segments from historical data can support chemical process modeling.