Online Data-Driven Model Predictive Control in Variable Noise Environment Based on Neural Network and Gaussian Process Regression

This paper investigates model predictive control in variable noise environment. A modeling method based on neural network and online Gaussian process regression is proposed, and the corresponding model predictive control algorithm is designed. Compared with the common data-driven model predictive control, this paper uses neural network to replace the traditional mathematical model, which reduces the workload of measuring parameters in modeling. In addition, Gaussian process regression is used for online learning to improve the performance of agents under noise interference. A simulation experiment is designed to demonstrate the effectiveness of the method. Finally, the experiments on a real quadrotor are conducted and the results show that the proposed algorithm remains feasible in the real world.