相关向量机在车辆行驶状态估计中的应用

Application of relevance vector machine in vehicle state estimation

针对车辆运动的非线性特性,利用比支持向量机(SVM)测试时间短、多样本时具有计算量小的相关向量机(RVM)对车辆行驶状态进行估计。为了能够较为准确地估计车辆行驶状态,采集实车试验数据,利用Kalman滤波器对采集到的车速和横摆角速度数据进行滤波,将滤波后的数据作为RVM的输入。依据贝叶斯理论建立最大似然函数,考虑到横摆角速度和车速变化的差异性,依据不同迭代次数下最大似然估计值、伽马值以及值的差异性确定最佳的迭代次数,保证模型具有较短的测试时间和较高的击中概率。有效性验证结果表明:该模型能够较为准确地逼近待估计样本的真值,其中波动性较大的横摆角速度所需要的迭代次数更多,伽马值和值的变化更为迅速,收敛速度较快。

Because vehicle movement exists nonlinear characteristics,this paper adopted relevance vector machine（RVM） to estimate the vehicle state.The test time of relevance vector machine is shorter than support vector machine（SVM） and the amount of computation is smaller than SVM.In order to estimate vehicle state accurately,Kalman filter was used to filter yaw rate and velocity data after collecting the test data.The filtered data were regarded as the input of relevance vector machine.Then the maximum likelihood function was established according to Bayesian theory.The difference between the change of yaw rate and velocity data was considered.The optimal number of iterations could be determined by gamma values,values difference and the maximum likelihood values in different iterations so as to guarantee the shorter testing time and higher hit probability of this model.Finally,the validation of the model was proved.The results show that this model can approximate to the true values of the estimated samples more accurately.The yaw rates with greater volatility need more number of iterations,gamma values and values change more rapidly,and the convergence speed is faster.2 tabs,9 figs,11 refs.