基于SVR模型的驾驶简单反应时间预测方法

Predicting Method of Simple Reaction Time of Driver Based on SVR Model

为准确预测驾驶人对突发事件的简单反应时间,并为构建自适应式危险驾驶状态预警系统奠定基础,提出了一种基于实时脑电图(EEG)信号的驾驶人简单反应时间预测方法。首先,通过快速傅里叶变换(FFT)对EEG信号进行特征参数提取,作为驾驶人简单反应时间的客观预测指标。在此基础上,基于支持向量回归(SVR)建立驾驶人对突发事件简单反应时间的预测模型。最后采用20名驾驶人连续驾驶4h的EEG信号与反应时间数据,对该模型予以试算。研究结果表明:3项脑电特征参数(θ,α,β)与反应时间均具有显著相关性,其中脑电特征参数α的相关性最显著,为SVR模型对简单反应时间进行预测提供了客观预测指标;分别采用径向基函数(RBF)、Polynomial函数、Sigmoid函数作为核函数构建SVR模型对简单反应时间进行预测时,所得预测结果中采用RBF函数所产生的各项误差均低于其他2项函数,表明采用RBF函数为核函数的SVR模型预测精度最优,其预测准确率达到80%以上。

In order to accurately predict the simple reaction time of drivers to emergencies and construct the basis of adaptive warning system in dangerous driving state, the forecasting method of the driver＇s simple reaction time based on Electroencephalogram （EEG） in real time was proposed. First, characteristic parameters were extracted from the EEG by the fast Fourier transform （FFT） as objective forecasting indexes of drivers＇simple reaction time. The forecasting model of the simple reaction time of drivers to emergencies based on the support vector regression （SVR） was established. Finally, 4 hours continual driving reaction time and EEG from 20 drivers were used to test the model. The results indicates that three EEG characteristic parameters （θ,α, β） are significantly correlated with the simple reaction time and the correlation of EEG characteristic parameter a is the most significant, which provides the objective prediction index for predicting the reaction time of the SVR model. The SVR model is constructed by the kernel function in terms of radial basis function （RBF）, Polynomial and Sigmoid function and the simple reaction time is predicted as well. than those in consideration of two The various errors of SVR based on the RBF function are lower other functions, which indicates that the prediction precision of SVR model using RBF kernel function is the best, and the accuracy of prediction is above 80 %.