Experimental Research on Vehicle Sideslip Angle Estimation Based on Improved RBF Neural Networks

Abstract

The theory and algorithm of the improved radial basis function neural network(IRBF NN) are applied in estimation of vehicle body and wheel sideslip angles,based on the configuration sensor signal of electronic stability program(ESP) system.In order to simplify the RBF NN construction and improve its generalization and real-time calculation performance,three methods are put forward to modify the radial basis function network orthogonal least squares(OLS) learning algorithm.A road test system is designed to mainly acquire the signals of the body and wheel sideslip angles,and of the kinemics parameters of ESP configuration sensors.Several typical vehicle manipulability tests are conducted on high μ adhesion road,and the test data are used to train and configure the NN construction.The final NN construction and learning parameters are determined by cross-verification method,such as the network construction being 4-12-2,the expansion constant 9,the goal learning error and its gradient 0.025 and 0.05 respectively.The estimation accuracies of body and wheel sideslip angles are 0.5° and 0.8° proved by the verification test data.Finally,the NN real-time prediction performance is tested on the PC machine.The study shows that the constructed RBF NN with its good accuracy and real-time calculation performance can meet the requirements of the ESP controller for monitoring the sideslip angles.