Product composition control based on backpropagation neural network in pressure-swing distillation processes

In this paper, an intelligent control strategy based on back propagation neural network (BPNN) is proposed for product composition control in pressure-swing distillation (PSD) processes. A data-driven intelligent controller based on BPNN was combined with PID control instead of traditional composition controllers to avoid the problem that composition is difficult to measure online in real-time. The intelligent controllers are used to predict temperature set point in composition-temperature cascade control by using the process variables easy to measure, e.g., reboiler duty, thus avoiding composition measurement. The critical variables for output prediction are analyzed by correlation analysis to present the relationship between the output variables and input variables, then to train highly correlated variables by BPNN. Two typical triple-columns PSD processes, i.e., Ethanol/THF/ Water and ACN/IPA/Water, were used to verify the reliability and accuracy of the intelligent controllers under +/- 20% of feed flow and composition disturbances. Results demonstrated that the proposed intelligent control strategy presents good dynamic performance without the composition analyzer. This study is significant in improving dynamic performance and solving practical application problems by combining the traditional PID control and data-driven intelligent control.