Loading [a11y]/accessibility-menu.js
Attention Weighted SIQAE Neural Network for Chemical Process Quality Prediction | IEEE Conference Publication | IEEE Xplore

Attention Weighted SIQAE Neural Network for Chemical Process Quality Prediction


Abstract:

In this paper, an attention weighted stacked input isomorphism and quality driven autoencoder (AWSIQAE) network is developed for product quality prediction of chemical pr...Show More

Abstract:

In this paper, an attention weighted stacked input isomorphism and quality driven autoencoder (AWSIQAE) network is developed for product quality prediction of chemical process. Firstly, in order to eliminate the input reconstruction error accumulation of the traditional SAE model, the stacked input isomorphism and quality driven AE (SIQAE) model is proposed to solve the error accumulation issue while extracting quality-related features. Secondly, consider that the SIQAE model only predicts product quality by the deepest hidden layer feature, and loses the quality-related information in other hidden layer features. To this end, an attention-weighted SIQAE model is further proposed for final product quality prediction. By introducing attention mechanism to each hidden layer feature, a larger attention value is given to the hidden layer feature which is more important for quality variable prediction. Thus, the proposed model focuses on key information and removes the inhibition of quality prediction results by non-critical features. Finally, the proposed AWSIQAE network is used for the prediction of sulfur dioxide concentration in SUR. The application results indicate the AWSIQAE method increases the precision of the original SAE method by 11%.
Date of Conference: 22-24 September 2023
Date Added to IEEE Xplore: 03 November 2023
ISBN Information:
Conference Location: Yibin, China

References

References is not available for this document.