BP神经网络跟车模型优化  

Optimization of BP neural network based model for car-following

作　　者：郭艳君[1,2] 鲁玉萍 张运虎[1] 王畅[1] 

机构地区：[1]长安大学汽车学院,陕西西安710064 [2]辽宁省交通高等专科学校物流管理系,辽宁沈阳110122

出　　处：《中国安全科学学报》2016年第3期103-108,共6页China Safety Science Journal

基　　金：国家自然科学基金资助(61374196;61473046);长江学者和创新团队发展计划项目(IRT1286);中央高校基本科研业务费专项资金项目(310822151028)

摘　　要：为提高跟车预警系统给出的结果的准确性和可靠性,利用毫米波雷达开展实际驾驶试验,获取驾驶人在跟车过程中的稳定、加速和减速状态表征数据。以自车速度、自车与前车相对速度、自车与前车相对距离等3类参数的不同组合为输入变量,以自车的加减速特性为输出变量,建立BP神经网络模型。用遗传算法(GA)优化该模型。结果表明,单纯的BP神经网络模型预测准确率较低,利用GA优化模型后可有效提高模型的准确率。当输入参数为自车速度、相对距离与相对速度时,模型的有效率达到94.17%。In order to improve the accuracy and reliability of the warning system, driving tests were carried out using a millimeter-wave radar. Data were obtained for characteristics of the driver＇s stability,acceleration and deceleration. A BP neural network based model was built for car-following, for which different combinations of the vehicle speed, the relative velocity, the relative distance were taken as input variables,the car＇s acceleration and deceleration characteristics as output variables. GA method was used to optimize the model.The analysis results show that the optimiztd model is much better than the BP neural network based model in prediction accuracy, and that effective rate of the former can reache 94.17%.

关 键 词：BP神经网络 跟车模型 遗传算法(GA) 优化 预测 

分 类 号：X913.4[环境科学与工程—安全科学]