Aiming at the error accumulation in the traditional iterative training of process quality prediction models

a prediction model combining temporal attention mechanism

long and short term time series network

improved complete ensemble EMD

random forest (TAM-LSTNet-CEEMDAN-RF) error correction was proposed. Firstly

by introducing mutual information and stacking sparse autoencoder

the effective features are screened from process data

and the effective dimensions are constructed. Then

TAM-LSTNet model was used to mine the complex correlation between the effective dimension and the process time series data

get the first predicted value and subtracted from the test value to calculate the error sequence. The error sequence was corrected by using CEEMDAN-RF model to obtain the second predicted value. Finally

the two predicted values are added to obtain the predicted value of the quality index. The results show that the fitting degree of the coupling model is 0.036 and 0.029 higher than that of TAM-LSTM model and TAM-LSTNet-RF model

respectively

which verifies the effectiveness and applicability of the present method. The error correction model can accurately predict the production quality of the process.