关键词: 制烟丝间歇过程, 烟丝含水率预测, 递推式建模, TimesNet, 动态加权集成

Abstract: To address the challenges of time-delay transmission and dynamic disturbances caused by process-to-process material buffering,multiple batches,and environmental temperature and humidity in the intermittent production of cigarette shreds,a quality prediction method integrating time-delay coupling and dynamic environmental disturbances was proposed.Process parameters and environmental temperature and humidity data were temporally and spatially aligned through batch-based data preprocessing.A dual-layer modeling framework of process and environment was then constructed.In the process modeling layer,a serial prediction architecture based on key process node decomposition was adopted,and a Convolutional Neural Network—Bidirectional Long Short Term Memory—Multi-Head Attention(CNN-BiLSTM-MHA)hybrid network was used to deeply explore the complex time-delay relationships between cross-process parameters,enabling recursive propagation of historical process quality indicators.In the environmental modeling layer,the TimesNet time-series analysis framework was introduced to capture the nonlinear time-varying effects of temperature and humidity fluctuations on moisture content through multi-period spatiotemporal tensor transformations.A dynamic weighted integration strategy was further proposed,where the fusion weights of the process and environmental models were adaptively optimized using multiple linear regression.Experiments showed that the proposed method significantly outperformed traditional models in terms of RMSE(0.0113),MAE(0.0091),and R(0.9815),demonstrating its ability to characterize the dynamic evolution of quality indicators in intermittent production scenarios.

Key words: intermittent production of cigarette shreds, moisture content prediction of cigarette shreds, recursive modeling, TimesNet, dynamic weighted integration