Multi-moving-window neural network for modeling of purified terephthalic acid solvent system

To explore the unsteady-state and dynamics of purified terephthalic acid (PTA) solvent system, a multi-moving-window neural network (MMWNN) is proposed for process modeling. The core of this modeling approach is that multi-moving-window concept is incorporated in combination with auto-associative neural network (AANN) and generalized regression neural network (GRNN). The integrated neural network model is developed with different moving windows for process inputs, AANN for data compression and GRNN for model prediction, which can effectively capture the changing process dynamics, reduce the data dimension and reveal the nonlinear relationship between process variables and final output. For comparison, single-moving-window with AANN and GRNN (SMWNN), none-moving-window with AANN and GRNN (NMWNN) are also established for process modeling. Through the actual application in PTA solvent system of a chemical plant, the predicted results show that the proposed MMWNN is supervior to other neural networks with smaller prediction error that is more consistent with actual process. It is considered that MMWNN modeling could provide a useful guideline to explore the complicated dynamics of industry process.