PROPOSE A HYBRID TRAINING METHOD FOR SPIKING NEURAL NETWORKS TO IMPROVE THE ACCURACY IN IDENTIFYING THE AERODYNAMIC COEFFICIENTS OF AN AIRCRAFT

Abstract

This paper proposes a hybrid training method for spiking neural networks to identify the aerodynamic coefficients of an aircraft in the attitude channel using a nonlinear model. The proposed training method combines the backpropagation algorithm with adaptive decay time and the normalized spiking error backpropagation algorithm. This combination leverages the strengths of both algorithms in updating decay times and synaptic weights. As a result, it reduces training time and enhances stability during error reduction, thereby improving the accuracy and reliability of the identified aerodynamic parameters. Simulation results show that the aerodynamic coefficients in the aircraft’s pitch channel identified using the proposed method are more accurate than those obtained with traditional methods, and the network achieves faster convergence compared to using each training method individually. Additionally, the bootstrapping technique is used to determine the confidence interval for the aerodynamic parameters. The confidence interval indicates that the proposed method yields more reliable results compared to the algorithm before the combination.