Deep Transfer Learning for Approximate Model Predictive Control

Transfer learning is a machine technique that takes pre-trained model has already been trained on related task, and adapts it for use new, task. This particularly useful in the context of predictive control (MPC), where deep transfer used to improve training MPC by leveraging knowledge gained from controllers. One way which applied using MPC, then fine-tuning controller new process automation similar how an equipment operator quickly learns manually processing unit because skills learned controlling prior unit. reduces amount data required train approximate controller, also improves performance target system. Additionally, actions alleviates computational burden online optimization calculations, although this approach limited systems developed. The paper reviews formulations with case study illustrates neural networks create multiple-input multiple-output (MIMO) MPC. resulting existing but requires less than quarter system training. main contributions are summary survey motivating includes discussion future development work area. presents example MIMO discusses potential further research Overall, goal provide overview current state as well inspire guide learning.