Differential Recurrent Neural Network Based Model Predictive Control for the Control of MAV Attitude

An efficient differential recurrent neural network is developed in this paper, and the trained network can be used in the nonlinear model predictive control, and also predict the future dynamic behavior of the nonlinear process in real time. In the new training network, use Taylor series expansion and automatic differentiation techniques. The effectiveness of the differential recurrent neural network predictive model training and predictive controller demonstrated through the MAV attitude control. The differential recurrent neural network-based NMPC approach results in good control performance.