基于深度学习DBN算法的高速公路危险变道判别模型

Dangerous lane-change detecting model on highway based on deep learning DBN algorithm

针对危险变道过程影响交通安全这一问题,提出一种基于深度学习DBN(deep belief netw orks)算法与分类分析方法的新型危险变道量化判别模型,以解决现存车辆变道过程不可被量化分析及准确判别的问题.招募28名被试者,利用模拟器仿真平台开展实际场景下试验,获取详细的车辆行驶数据及驾驶环境数据作为训练的模型输入.采用SVM算法作为输出层的分类器,建立了DBN-SVM判别模型及基于样本下模型的一般训练方法.该模型的识别精度为93.78%,较朴素贝叶斯模型和BP-ANN神经网络分别提高20.11%和14.45%,并且调整参数后判别结果稳定.DBN-SVM模型可以根据驾驶员历史驾驶数据对即将发生的危险变道做出预测及判别,对驾驶员提出警告,从而减少交通事故的发生.此外,该研究为车联网环境下变道判别及警示的研究提供了理论支持.

Aiming at the problem that the vehicle lane-changing process cannot be quantitatively analyzed and accurately discriminated,a new quantitative discriminant model based on the DBN（ deep belief networks） algorithm and the classification analysis method is presented. Twenty-eight participants were recruited. The participators took part in real-scene simulation experiments using the simulation driving platform. The detailed data of vehicle traveling and driving environment was required and used as the input of the model. With the SVM（ support vector machine） algorithm as the classifier of the output layer,the discriminant model DBN-SVM and corresponding training method are set up. The discriminant accuracy of the DBN-SVM is 93. 78%,increasing by 20. 11% and 14. 45%compared with the Na6 ve Bayes model and BP-ANN（ back propagation-artificial neural networks）,respectively. And,the results are stable with adjusted parameters. The DBN-SVM model can predict and discriminate coming dangerous lane-changing according to drivers＇ driving history data,and warn drivers. As a result,it can reduce the chance of traffic accidents. This study provides theoretical support for lane-changing discrimination and warning under the connected vehicle environment.