Hybrid Feature Selection Approach to Finding Optimal Feature Subsets for Vehicle Lateral Velocity Estimation

This study develops a methodology that estimates the lateral velocity of vehicles by integrating an artificial neural network (ANN) with physics for optimal feature selection. In vehicle active control systems, information about its velocity is crucial for the vehicle to make real-time decisions and adjustments and ensure the vehicle operates safely and efficiently in various conditions. However, no sensor is available for the direct measurement of lateral velocity, and hence, indirect estimation methods are utilized. Given the limitations of model-driven methods and advancements in artificial intelligence, we employed ANNs instead of traditional model-driven methods for lateral velocity estimation. As ANNs are data-driven methods, a preprocessing step to find suitable data is essential to achieve good performance. We propose a hybrid feature selection method that combines data-driven feature selection with physics-based methods to select relevant features for lateral velocity estimation. Thus, the optimal feature subset for ANNs was identified. This approach was validated by designing and training nonlinear autoregressive with exogenous inputs and one-dimensional convolutional neural network models, demonstrating enhanced estimation performance with the selected feature subset. The findings suggest that the proposed method enhances the accuracy of lateral velocity estimation, contributing to the advancement of vehicle safety technologies.