System Identification of an Aerial Delivery System with a Ram-Air Parachute Using a NARX Network

Abstract

Neural networks are one of the methods used in system identification problems. In this study, a NARX network with a serial-parallel structure was used to identify an unknown aerial delivery system with a ram-air parachute. The dataset was created using the software-in-the-loop method (Software in the loop). Gazebo was used as the simulator and PX4 was used as the autopilot software. The performance of the NARX network differed according to parameters used, such as the selected training algorithm, input and output delays, the hidden layer, and the number of neurons. Within the scope of this study, each parameter was examined independently. Models were trained using MATLAB 2020a. The results demonstrated that the model with one hidden layer and five neurons, which was trained using the Bayesian regularization algorithm, was sufficient for this problem.

Keywords

• Bayesian regularization
• Levenberg–Marquardt
• NARX network
• scaled conjugate gradient
• software in the loop