Neural Network-Based Prediction for Lateral Acceleration of Vehicles

Abstract

Lateral acceleration is a key element of vehicle dynamics. It is consumed by several control, stability and comfort functions of the vehicle. In this paper a neural network-based prediction method is demonstrated for predicting the value of lateral acceleration. The inputs of the method are the most accessible signals in any modern vehicle: wheel speed information, longitudinal acceleration and steering wheel angle. For training, validating and testing the neural network, experimental data was used. The hyperparameters of the neural network were tuned by a hybrid approach. The accuracy of the approach was evaluated by comparing the actual measured values to those predicted by the neural network. Evaluation results convincingly demonstrate the usefulness and reliability of the developed model.