摘要
以常用的交通数据———交通量时间序列的实测数据为例,给出多个噪声识别及消噪预处理的实验结果.为提高建模的准确度,采用模糊减法聚类对交叉口实测交通量进行噪声识别.对实测交通量采用奇异值分解的滤波方法进行除噪预处理,并通过训练径向基函数网络进行预测.与数据未经滤波直接训练网络相比,预测结果的平均绝对相对误差降低了3.31%.用小波变换对交通量数据进行消噪处理,即通过多分辨率的小波分解和重构来实现消噪.实验表明,若对原始交通量时间序列消噪后再建立预测模型,将获得更好的预测结果,这说明研究的噪声识别和消噪方法的可行性和有效性.
Taking noise recognition and noise reduction of traffic volume time series which are commonly used traffic data as example, several experimental results are illustrated. In order to improve the accuracy of modeling, fuzzy subtraction clustering is employed to recognize the noise data hidden in traffic volume time series gathered in intersection; De-noise filter method based on single value decomposition is applied to preprocess traffic volume time series, and a radical basic function neural network is trained for prediction. The mean absolute relative error of the prediction is reduced by 3.31% compared to that of network trained with raw data without filter. Wavelet transform, i. e. multi-resolution decomposition and reconstruction is also used to reduce noise. These experiments indicate that the prediction model built with traffic volume time series after noise reduction can yield better results. It oroves the feasibility and validity of above mentioned approaches.
出处
《东南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2006年第2期322-325,共4页
Journal of Southeast University：Natural Science Edition
基金
国家自然科学基金资助项目(50378016)
江苏省教委自然科学基金资助项目(05KJB520056)
关键词
噪声识别
消噪
交通数据
小波分析
免疫算法
noise recognition
noise reduction
traffic measure data
wavelet analysis
immune algorithm
