Multi-layer moving-window hierarchical neural network for modeling of high-density polyethylene cascade reaction process

With the growing scale of industry production, process modeling has been paid more and more attention, which could effectively explore the dynamics of the process and provide guidelines to production operation. High-density polyethylene (HDPE) cascade reaction process is such a complex and nonlinear industry process. To enhance the performance of process modeling, a multi-layer moving-window hierarchical neural network (MMHNN) is proposed, which is developed with the incorporation of multi-layer moving-window concept and hierarchical neural network (HNN). Multi-layer moving-window is used to ensure the continuity and time-variation, HNN is used for input compression and 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-layer moving-window HNN (SMHNN) and HNN are also established for the process modeling. Through the actual application in HDPE cascade reaction process of a chemical plant, the prediction results show that MMHNN is obviously better than SMHNN and HNN with higher accuracy, thus exploits a new and efficient way to simulate and guide the industry process.