Virtual Support Vector Augmentation and Bayesian Optimization for Small Unmanned Helicopter Modeling

Accurate modeling of flight dynamics in small unmanned helicopters is critical for enabling reliable autonomous control. While support vector regression (SVR) performs well in small-sample, nonlinear regression tasks, its generalization capability is often limited by data sparsity. To address this challenge, this study proposes an enhanced least squares support vector regression (LSSVR) method—termed BO-VSV-LSSVR—which integrates virtual support vector data augmentation (VSV-DA) and Bayesian optimization (BO). In the proposed framework, high-confidence virtual training samples are generated around support vectors using Gaussian perturbation and weighted neighborhood interpolation, effectively alleviating data sparsity. Additionally, BO is employed to adaptively tune model hyperparameters, further improving predictive performance. Experimental evaluations using multiple real-flight datasets from a small unmanned helicopter validate that the BO-VSV-LSSVR model outperforms conventional methods in both prediction accuracy and generalization capability. This work offers a high-precision modeling framework for small unmanned helicopter control and contributes to the broader development of data augmentation and intelligent optimization techniques in regression-based modeling.