Transfer Learning-Based Modeling and Predictive Control of Nonlinear Processes

Abstract

This work develops a transfer learning (TL) framework for modeling nonlinear dynamic systems using recurrent neural networks (RNNs). The TL-based RNN models are then incorporated into the design of model predictive control (MPC) systems. Specifically, transfer learning uses a pre-trained model developed based on a source domain as the starting point, and adapts the model to a target domain with similar data distribution. The generalization error for TLbased RNNs (TL-RNNs) that depends on model capacity and discrepancy between source and target domains is first derived to demonstrate the generalization capability on target process. Subsequently, the TL-RNN model is utilized as the prediction model in MPC for the target process. Finally, a chemical process example is used to demonstrate the benefits of transfer learning.