Incorporating Structural Process Knowledge in Recurrent Neural Network Modeling of Nonlinear Processes

This work proposes two methods for incorporating structural process knowledge in recurrent neural network (RNN) modeling for a general class of nonlinear dynamic process systems. Specifically, based on the structural a priori knowledge of the relationship between the process state variables, the fir...

Full description

Saved in:
Bibliographic Details
Main Authors: Wu, Zhe, Rincon, David, Christofides, Panagiotis D.
Format: Conference Proceeding
Language:English
Subjects:
Biological neural networks
Biological system modeling
Data models
Nonlinear systems
Predictive models
Recurrent neural networks
Training
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This work proposes two methods for incorporating structural process knowledge in recurrent neural network (RNN) modeling for a general class of nonlinear dynamic process systems. Specifically, based on the structural a priori knowledge of the relationship between the process state variables, the first approach is to design a partially-connected structure for RNN models. Additionally, a weight-constrained RNN optimization problem equipped with a constraint on weight parameters and a regularization term in loss function is proposed to develop an RNN model that satisfies the assumption on input-output relationship. The proposed partially-connected RNN modeling method is applied to a chemical process example to demonstrate its better approximation performance compared with the fully-connected RNN model that does not incorporate any structural process knowledge.
ISSN:2378-5861
DOI:10.23919/ACC45564.2020.9147519