一种平稳化短时交通流预测方法  

Stationary Short-Term Traffic Flow Prediction Method

作　　者：康军[1] 段宗涛[1] 唐蕾[1] 温兴超 KANG Jun;DUAN Zong-tao;TANG Lei;WEN Xing-chao(School of Information Engineering, Chang, an University, Xi, an 710064, China)

机构地区：[1]长安大学信息工程学院,陕西西安710064

出　　处：《测控技术》2018年第2期33-37,共5页Measurement & Control Technology

基　　金：国家自然科学基金资助项目(61303041);交通运输部应用基础研究项目(2014319812150);陕西省科技厅工业科技攻关项目(2014K05-28;2015GY002;2016GY-078)

摘　　要：支持向量机回归模型是以预测噪声具有对称性概率分布为假设条件,而实际的短时交通流数据序列具有非平稳特征,这就使得在采用支持向量机回归模型进行短时交通流预测时,难以保证预测噪声的对称性概率分布,从而会影响到预测精度。针对上述问题,在证明支持向量机回归模型对平稳时间序列的预测噪声具有对称性概率分布的基础上,分别针对平稳化和未平稳化的短时交通流观测序列进行了仿真预测,并对预测结果进行了比对分析。分析结果表明,采用平稳化短时交通流预测方法可将预测的均方根误差降低约21.6%,绝对值误差降低约21.3%,相对误差降低约17.3%,仿真结果验证了所提方法的有效性。Prediction noise that has the symmetric probability distribution is an assumption condition for thesupport vector machine （ SVM） regression model. However, actual short-term traffic flow data sequence hasnon-stationary characteristic, which makes it difficult to guarantee the symmetry probability distribution of predictionnoise when SVM regression model is used in short term traffic flow prediction, and the prediction precisionwill deteriorate. In order to solve the above problems, on the basis of proving that the SVM regression modelhas the symmetry probability distribution for prediction noise of stationary time series, the short-term trafficflow observation sequences with stationary and unstationary are simulated and predicted respectively, and theprediction results are compared and analyzed. The results indicate that the proposed method can reduce the rootmean square error of prediction by about 2 1 .6% , reduce the absolute value error by about 2 1 .3 % , and reducethe relative error by about 17. 3 % . The simulation results validate the proposed method.

关 键 词：短时交通流预测 统计学习 平稳化方法 支持向量机 季节性差分 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]