Melt flow index prediction for batch polypropylene based on temporal feature matching

Abstract Due to the delay in the laboratory results of the melt flow index (MFI) for polypropylene (PP) batch processes, production personnel are unable to observe MFI changes in a timely manner and guide the production of the next batch. To obtain the MFI values in a timely manner, a prediction method based on time‐series feature matching for MFI is proposed. First, the process mechanism is analyzed to determine the process variables affecting MFI, and the corresponding historical data is collected. Then, in the offline training phase, expert rules based on the process mechanism are constructed to extract historical batch data and integrate batch MFI values. Next, for the extracted long time‐series data, it is transformed into a spatiotemporal matrix, and singular value decomposition (SVD) is used to extract features. In the online prediction phase, considering the peak characteristics of hydrogenation times, a similarity calculation method using a pseudo‐4th‐order central moment (P4CM) is proposed. This is combined with Euclidean distance to compare the similarity between online and historical batches for prediction. Finally, a comparative experiment with Kernel principal component analysis (KPCA) is conducted, demonstrating the feasibility of the proposed prediction method.