Novel Hybrid Neural Network Modeling Method for Melt Index Industrial Soft Sensor

Polypropylene (PP) is a synthetic resin widely used in various industries and daily activities due to its exceptional physical characteristics. Therefore, improving the quality control of PP products is crucial for enterprises, and the Melt Index (MI) is a vital metric for assessing PP quality. To enable prompt and accurate MI prediction, we propose a pioneering hybrid neural network model. This model incorporates three components: the Causal Convolutional Neural Network (CCNN), Long Short-Term Memory (LSTM), and Temporal Pattern Attention (TPA). CCNN is employed to extract features from time series data, LSTM is used to identify temporal correlations and patterns, and TPA is used to extract implicit temporal pattern information across multiple time intervals. By employing this model in the actual polypropylene production process and comparing it with other different models, it achieves unparalleled performance in realtime MI computation.