基于因果分析的交通流组合预测模型  

Traffic flow combination prediction model based on causal analysis

作　　者：林蒙蒙 覃锡忠[1] 贾振红[1] 祁欣学 LIN Meng-meng;QIN Xi-zhong;JIA Zhen-hong;QI Xin-xue(College of Information Science and Engineering,Xinjiang University,Urumqi 830046,China)

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046

出　　处：《计算机工程与设计》2021年第7期2030-2036,共7页Computer Engineering and Design

基　　金：新疆维吾尔自治区自然科学基金项目(2019D01C058)。

摘　　要：针对目前基于聚类方法的交通流预测模型,在聚类时,未考虑到不同因素对交通流影响程度不同的问题,引入因果分析方法来量化各因素的重要程度,同时提出一种预测框架,基于因果分析的套索回归(LASSO)和极限学习机(ELM)组合预测模型。采用占用率和车速两种因素,引入符号转移熵分别对各因素与交通流进行因果分析;根据分析结果为每种因素加权,利用K-Means算法对数据进行聚类;通过LASSO捕捉线性关系,ELM学习非线性关系,为每一类交通流建立特有的预测模型。通过对洛杉矶地区的实验,验证了组合模型对预测精度的提升具有很大帮助,在引入因果分析后,预测精度得到了更进一步提升。In view of the present traffic flow prediction model based on clustering method,in clustering,without considering the different influence degree of different factors on traffic flow,the causal analysis method was introduced to quantify the importance degree of each factor,and a kind of prediction framework,based on the analysis of the causal LASSO regression and extreme learning machine(ELM)combination forecast model,was proposed.The occupancy rate and speed were used to analyze the causal relationship between the factors and traffic flow respectively by introducing sign transfer entropy.Each factor was weighted according to the analysis results,and K-Means algorithm was used to cluster the data.The linear relationship was captured through LASSO,and the nonlinear relationship was learned by ELM,so as to establish a unique prediction model for each type of traffic flow.Through experiments in Los Angeles,it is verified that the combined model is of great help to improve the prediction accuracy,and the prediction accuracy is further improved after introducing causal analysis.

关 键 词：交通流预测 因果分析 聚类 线性关系 非线性关系 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]