(183s) Process Data Analytics Using Deep-Learning Based Methods
AIChE Annual Meeting
2018
2018 AIChE Annual Meeting
Computing and Systems Technology Division
Interactive Session: Data and Information Systems
Monday, October 29, 2018 - 3:30pm to 5:00pm
In recent years, various deep learning methods have been introduced for knowledge extraction. Deep learning is a class of machine learning techniques developed based on learning data representations1, as opposed to task-specific feature engineering. Effective deep learning methods can extract features automatically from data, which results in much better performance.
In this paper, we introduce our recent investigation on the prediction of catalyst age of a Liquid Phase Methanol (LPMEOH) process using a recurrent neural network (RNN) technique. Facing the existence of massive time series data involving exogenous parameters, we adopt a powerful technique to construct a dual-stage attention-based recurrent neural network (DA-RNN)2. This method consists of an encoder with an input attention mechanism, which can adaptively select the most relevant input features, and a decoder with a temporal attention mechanism that can capture the long-term temporal dependencies of a time series. The case study will demonstrate the methodological efficacy, which shows the outstanding prediction performance as compared with traditional learning methods.
[1] Bengio Y, Courville A, Vincent P (2013). Representation learning: A review and new perspectives. Pattern Analysis and Machine Intelligence, IEEE Transactions on 35(8):1798â1828. doi:10.1109/TPAMI.2013.50.
[2] Yao Qin, Dongjin Song, Haifeng Chen, Wei Cheng, Geoff Jiang, Garrison Cottrell (2017). A Dual-Stage Attention-Based Recurrent Neural Network for Time Series Prediction. Proc. 27th Intl. Joint Conf. on Artificial Intelligence (IJCAI).
_____________________________________________________________________________
* All correspondence should be addressed to Prof. Yinlun Huang (Phone: 313-577-3771; E-mail: yhuang@wayne.edu).