Modeling of Small Unmanned Helicopter Using a Self-Constructed Kernel Function APSO-LSSVM

The complex nonlinear and strongly coupled dynamics of small unmanned helicopters make mathematical modeling challenging. Traditional approaches often rely on least squares support vector machine (LSSVM) algorithms using standard kernel functions, which are limited in their learning and generalization capabilities, resulting in insufficient accuracy for flight control systems. This paper proposes an adaptive particle swarm optimization-based LSSVM (APSO-LSSVM) method, incorporating a self-constructed kernel function. Using Mercer’s theorem, the custom kernel addresses the limitations of conventional kernels and is integrated into the LSSVM framework. An adaptive particle swarm optimization algorithm, capable of dynamically adjusting inertia weights and learning factors, optimizes the model parameters, overcoming the standard particle swarm optimization’s tendency to get trapped in local optima. The model is trained using flight test data from a self-developed small unmanned helicopter, and its identification performance is cross-validated in the time domain against traditional models. Experimental results show that the proposed method significantly improves the modeling accuracy of small unmanned helicopters.