基于模糊C均值和神经网络的驾驶行为评价研究  

Study on Driving Behavior Evaluation Based on Fuzzy C-Means and Neural Network

作　　者：吴紫恒 吴仲城[1] 张俊 陈松 陈杰[1,2] 

机构地区：[1]中国科学院合肥物质科学研究院,强磁场中心,合肥230031 [2]中国科学技术大学,合肥230031

出　　处：《计算机系统应用》2018年第3期263-267,共5页Computer Systems & Applications

基　　金：国家自然科学基金(61273323)

摘　　要：交通拥挤正成为一个日益严重的问题,一些不安全的驾驶行为所导致的交通事故是造成拥堵的主要原因之一.因此,如何准确评价驾驶员的驾驶行为成为研究的热点.本文提出了一种基于模糊C均值聚类(Fuzzy C-Means,FCM)和BP神经网络相结合的驾驶行为评价方法,首先利用FCM对驾驶行为进行初始聚类,基于FCM聚类结果,为了提高BP神经网络分类精度,本文提出了一种自动挑选训练样本即典型样本的方法,利用BP网络进行学习,最终用训练得到的BP神经网络分类器对驾驶行为进行实时分类,研究结果表明该算法摒弃了人为主观因素,实现了驾驶行为准确、客观、高效的评价.Traffic congestion is becoming an increasingly serious problem.Traffic accidents caused by risky driving behaviors are one important cause.Therefore,the accurate evaluation of driving behaviors has become a research hotspot.This study puts forward an evaluation algorithm of driving behaviors based on the combination of FCM and BP neural network.Firstly,FCM is used to make initial clusters of driving behaviors.Secondly,in accordance with the results of clusters,an algorithm that the typical samples are automatically selected as the training samples for BP neural network classifier is proposed.Finally,the trained BP neural network is used to classify the driving behaviors.The research result shows that the algorithm can eliminate subjective factors and make accurate,objective and efficient driving behavior evaluation.

关 键 词：驾驶行为 模糊C均值聚类 神经网络 典型样本 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] U491[自动化与计算机技术—控制科学与工程]