关键词: 复杂流程生产, 产品质量预测, 时间卷积神经网络, 双重注意力机制

Abstract: The quality of products in complex production processes is related to process parameters,equipment operating parameters,real-time operating conditions,etc.There is a complex nonlinear relationship between product quality and process parameters.To improve the accuracy of product quality prediction,a method for accurate prediction of complex production process quality that integrated dual attention mechanism and time convolutional neural network was proposed.A multi-variate temporal dataset was constructed based on real-time data collected by high-frequency sensors during the thin plate drying process,and the correlation coupling and temporal characteristics of the production process data were analyzed.The basic Temporal Convolutional Network(TCN)combined with the PReLU activation function was used to capture the long-term dependencies between process quality and feature variables,achieving feature extraction for all sequences in the multivariate time series dataset,effectively improving the accuracy of model prediction.On this basis,a dual attention mechanism was introduced to determine the multi-variate key time series features that affect product quality.Finally,multiple models were trained and tested based on 20 107 sets of experimental data from the thin plate drying process of a certain silk production line.The results showed that the goodness of fit(R2) of the model was 0.988,and the Mean Square Error(MSE),Root Mean Square Error(RMSE) and Mean Absolute Error(MAE) were 0.000 4,0.021 2,and 0.016 8 respectively.Compared with the basic TCN,the DA-TCN model had a reasonable structural design,high prediction accuracy and strong generalization,providing a reference for accurate prediction of product quality in complex production processes.

Key words: complex process production, product quality prediction, temporal convolutional network, dual attention mechanism