Attention-based recurrent neural network for multistep-ahead prediction of process performance. (2nd September 2020)

• Record Type: Journal Article
• Authors: Aliabadi, Majid Moradi; Emami, Hajar; Dong, Ming; Huang, Yinlun
• Is Part Of: Computers & chemical engineering. Volume 140(2020)
• Publication Date: 2020-09-02
• Subjects: Recurrent neural network -- Deep learning -- Attention mechanism -- Process performance prediction; Chemical engineering -- Data processing -- Periodicals; 660.0285
• DOI: 10.1016/j.compchemeng.2020.106931
• Languages: English
• ISSNs: 0098-1354

Abstract

Abstract: Attention-based RNN modeling technique could be potentially used for investigating a variety of process engineering problems that require multiple step predictions. This type of model consists of an RNN that encodes a sequence of time series data into a new representation form, an another RNN that decodes the representation into a target sequence, as well as an attention model added in between that allows the model to focus on part of the input sequence that are critical to predicting the target sequence. The model with this deep architecture for high-level representations can learn very complex dynamic systems. To demonstrate the effectiveness of the modeling approach, a comparative study on the problem of catalyst activity prediction is illustrated.