摘要
通过对汽车行驶数据的处理后找出刻画交通路况的特征值,利用主成分分析对各运动学片段的表征值降维后提取前三个主成分,再利用自组织映射神经网络模型进行训练,将所获得的权值设置为粒子群聚类的初始聚类中心,聚类后根据皮尔逊相关系数选取运动学片段,拟合各类行驶工况图.通过分析合成工况与原始工况的特征参数误差、速度与加速度联合分布图,并与K-means聚类方法进行对比后,得出:本文采用的方法误差较小,收敛速度快,合成工况图可以反映出原始数据所蕴含的交通路况信息.
The characteristic values of the traffic conditions are found after the big data of the car driving.Then,the characteristic value of each kinematic segment is reduced by principal component analysis and then the first three principal components are extracted,and then the self-organizing mapping neural network model is used for training.In this way,the initial cluster center of particle swarm clustering is set by the weight obtained,and kinematic segments are selected according to the Pearson correlation coefficient after clustering to fit various driving conditions graphs.By analyzing the characteristic parameter error,joint speed distribution of velocity and acceleration of the synthetic case and the original case,and comparing with the K-means clustering method.It is concluded that the method in this paper has less error,fast convergence speed,and synthetic the working condition map can reflect the traffic information contained in the original data.
作者
张林平
李风军
ZHANG Linping;LI Fengjun(School of Mathematics and Statistics,Ningxia University,Yinchuan Ningxia 750021)
出处
《宁夏师范学院学报》
2020年第10期76-89,共14页
Journal of Ningxia Normal University
基金
国家自然科学基金(61662060)
宁夏自然科学基金(NZ17011).
关键词
自组织映射神经网络
主成分分析
运动学片段
粒子群聚类
行驶工况
Self-organizing neural network
Principal component analysis
Kinematic segment
Particle swarm optimization clustering
Running condition
