基于单类分类方法的道路高排放源识别算法

Algorithm of on-road high-emitters identification based onone-class classification method

为了提高对道路高排放源的识别效率,降低其造成的大气污染,提出了一种基于随机傅里叶特征和非常稀疏映射的单类分类(OCC)宽度学习系统(BLS)的道路高排放源识别方法,即OCC-FS-BLS。首先,将道路高排放源数据进行非线性的随机傅里叶特征映射得到BLS的特征节点,再通过非常稀疏随机映射生成增强节点,拼接所有节点作为BLS输出层的输入;然后,通过岭回归求解改进BLS的输出权重;最后,根据OCC-BLS构建单类分类算法的策略,实现OCC-FS-BLS算法。实验结果表明:OC-FS-BLS在高排放源识别任务中相比OCC-BLS等其他模型表现出更好的识别性能。

In order to improve the identification efficiency of the on-road high-emitter and reduce the air pollution caused by them.An on-road high-emitter identification method,that is OCC-FS-BLS,based on one-class classification(OCC),random Fourier feature and very sparse random projection(FS)and broad learning system(BLS)is proposed.Firstly,the on-road high-emitter data are mapped by nonlinear random Fourier feature and get the feature nodes of broad learning system(BLS),the enhancement nodes are generated by very sparse random projections.Concatenate all nodes as input of BLS output layer.Then,the output weights of BLS are improved through ridge regression solution.Finally,OCC-FS-BLS algorithm is realized according to the strategy of one-class classification algorithm constructing by OCC-BLS.The experimental results shows that OCC-FS-BLS has better performance in the on-road high-emitter identification task compared with other models such as OCC-BLS.