State Parameters Estimation Of Vehicle Based On ROC-Kalman Filter

The accurate estimation of vehicle tire cornering stiffness online is one effective way to improve operating stability control system performance.Because the least square method is used to estimate the cornering stiffness online,the front and rear tire cornering stiffness will tend to zero when the side-slip angle is to be zero.This paper proposed to estimate the cornering stiffness and certroid side-slip angle online by the method of lateral acceleration.In this analysis,the parameters involved in estimation are sensitivity to noise,so the relevant parameters(lateral acceleration,lateral velocity,longitudinal velocity,longitudinal acceleration and steering angle)are necessary to be filtered before the estimation,and Kalman estimator is used to estimate tire cornering stiffness and centroid side-slip angle.Because the constant noise variance of Kalman filter is difficult to adapt to the strong nonlinear systems,and tire nonlinear characteristics,this paper proposed a dynamic optimal method of noise variance of Kalman filter for the nonlinear characteristics of running states based on ROC curve(receiver operating characteristic curve).Then the optimized model is used to estimate tire cornering stiffness and centroid side-slip angle.Firstly,the nonlinear vehicle model is established and the observable parameters are extracted from all state parameters.According to the observable parameters,The corresponding Kalman filter model can be established.The characteristics of driver's operator output would be taken into consideration and the time window scope can be determined preliminary.And then,the best fitting of the variables need to be estimated in single time window are achieved through high order.The fitting precision is used to determine the best time window.According to an observation value previous time,the observation error of this time window can be determined and the expect error is set to zero.According to the actual requirements,the error limits can be determined and the labels of class within the error limits are set to zero,the labels of class outside the error limits are set to one.So the questions can be converted to binary classification problems and the corresponding ROC curve was constructed according to the results of classification.The classification accuracy is used to modify the noise variance of Kalman filter,through which,noise variance can be updated dynamically.Eventually,the collaborative optimization of the system noise variance and measurement noise variance can be achieved.It can reduce estimation error caused by random selection of two classes of noise variance.The lateral acceleration method is adopted to estimate tire cornering stiffness and centroid side-slip angle online and the estimation results are compared with the measured values.The results show that compared with fixed constant noise variance of the Kalman filter,the optimized model calculation cost is slightly increased,but the consistency of the model convergence greatly enhanced,it can be close to the measured values accurately in real-time.