Aiming at the problems of many processes

serious coupling between processes

and complexity of multivariable processing data

a high-dimensional and multi-scale process quality prediction method based on Seq2Seq temporal attention mechanism is proposed. Based on the analysis of the characteristics of multi-process process data and the problems encountered in the process of encoding and decoding by using Seq2Seq model

the sequential attention mechanism was introduced to construct the time-domain information matrix of long-distance variation

and the convolutional neural network and BiLSTM were designed as the encoder components. At the same time

the potential depth features such as process parameter correlation and bidirectional sequence relationship of process timing data were learned

and key information was extracted by using sequence attention mechanism

so as to realize adaptive learning of nonlinear correlation characteristics and sequence dependence of process parameter timing data related to process quality. Finally

the practicability and effectiveness of the proposed method were verified by prediction experiments on the quality of silk manufacturing process. The method provides the implementation approach for accurate quality prediction of multi-process coupling process.