Online average-based system modelling method for batch process

Online system identification provides a powerful tool to process control engineers for controller synthesis, process dynamics monitoring, real-time optimization and other purposes at a low computational cost. Instead of processing data in time, this paper intends to propose processing data in a different “direction” – in iteration/batch to improve the estimates tracking performance. However, directly changing the data processing direction gives rise to severe fluctuations on parameter estimates within a batch. To overcome this problem, two online identification methods with simple implementations are devised based on average. One method is applying average in the dual space, while the other in the primal space. The convergence of both approaches has been analyzed. An adaptive average strategy based on moving-window is also developed to track inter-batch dynamics drift. Finally, the simulation results on injection molding, two-tank system and continuous stirred tank reactor (CSTR) testify the improved performance of the methods proposed in this paper.