Data Driven Approaches for Nonlinear System Identification and Control of Hardware in the Loop Systems

Abstract

System identification can play in important role in the engineering design process. Generating a high-fidelity model of a system can lead to improved performance and reduced costs. There are many ways to create these models but this work will focus on data driven approaches, specifically machine learning based methods. The major classes of machine learning techniques viewed are long short-term memory networks, nonlinear autoregressive models, and feed-forward networks with lagged inputs. These modeling techniques are explored in use for creating models of a hardware in the loop testing (HWIL) system and a vehicle with the focus being the reduction of mean squared error between the estimated outputs and the true outputs. The feed-forward network for the HWIL system is then used to create a trajectory optimization network to improve the performance of the system by reducing the tracking error to a reference input. The flexibility and generality of these machine learning based data driven modeling techniques provides a viable option for the creation of models for various unknown complex systems.