摘要
针对汽车尾气的非线性数据聚类问题,提出一种在自组织特征映射(Self-Organizing Map,SOM)下的聚类方法来评估汽车的排放水平。根据汽车在城区真实环境中的行驶速度设置SOM神经网络中的神经元个数,通过神经元之间拓扑相关的学习方式,自动形成具有数据原始属性的有序映射,实现不同排放水平的尾气数据聚类。为避免网络训练过程中出现训练死区的现象,竞争学习采用弹性邻域半径代替固定邻域半径,自适应地缩放学习区域。以某轻型车的THC和CO2排放数据为对象的数值试验结果表明,采用弹性邻域半径的SOM神经网络的聚类准确性优于采用固定邻域半径的SOM神经网络,能有效评估汽车尾气排放水平。
This paper provides an assessment method of automobile exhaust emission levels using the self- organizing map （SOM） neural network. The neuron number of the SOM neural network is selected based on the driving speed in urban areas. Through the topology learning of the neurons, an ordered mapping with the original properties of the exhaust data has been automatically formed to achieve the data clustering. To avoid the dead zone phenomenon occurring in training process of the network, the learning areas are adaptively scaled by the competitive learning with the flexible neighborhood radius instead of the fixed neighborhood radius. The numerical experiments, using the THC and COz emissions data, demonstrate that the SOM neural network based on the flexible neighborhood radius oftbrs a higher accuracy of the clustering than the network based on the fixed neighborhood radius.
出处
《汽车工程学报》
2013年第5期354-360,共7页
Chinese Journal of Automotive Engineering
基金
车辆排放与节能重庆市市级重点实验室资助项目(34.8-9.1)
