大数据背景下城市短时交通流预测

Urban Short-term Traffic Flow Forecasting with Big Data

为了在尽可能短的时间内挖掘和分析海量城市交通流数据,实时准确地预测城市短时交通流状态,建立有效的城市交通诱导系统,改善城市交通管理水平。根据城市交通大数据的来源异同、数据量大、种类繁多等特征,提出大数据背景下的城市短时交流状态预测新方法。新方法综合利用了随机森林算法进行机器学习的优势,克服了决策树算法的一些不足,又保留了决策树算法的优点;同时,新方法在大数据体系下实现了并行运算,提高了新方法各方面的学习性能,能够更快速、更加精确地实现城市短时交通流状态预测,并为城市交通诱导系统提出合理的交通建议。首先,针对城市交通流大数据的特征和城市短时交通流状态的预测需求,采用通用大数据分析处理平台构建城市交通流大数据管理平台,实现城市交通流大数据的整合、分布式存储与管理;然后,结合云计算技术,利用并行化计算模型MapReduce对随机森林算法实现并行化,增强算法的数据分析与处理效率,提高算法对大数据的处理能力;最后,采用并行化的随机森林算法对城市交通流大数据进行计算与处理,实现城市短时交通流状态的高效和实时预测。试验结果表明,并行化的随机森林算法的数据分析与处理效率、对城市短时交通流状态的预测精度,以及在不同数据集上对大数据的处理能力等各方面的性能均优于传统的预测方法。

In order to mine and analyze massive amounts of urban traffic flow data in the shortest possible time,predict the urban short-term traffic flow state accurately in real time,establish an effective urban traffic guidance system,and improve the management level of urban traffic,based on the characteristics of different sources,volume and various kinds of urban traffic big data,a new method for predicting urban short-term traffic flow state in the context of big data is put forward. The new method makes use of the advantages of the random forest algorithm for machine learning,overcomes some disadvantages of the decision tree algorithm,and retains the advantages of the decision tree algorithm. At the same time,the new method realizes parallel operation under the big data system,improves the learning performance of the new method in all aspects,and realizes the prediction of urban short-term traffic flow state more quickly and accurately,and puts forward reasonable traffic suggestions for the urban traffic guidance system. First,according to the characteristics of urban traffic flow big data and the forecast demand of urban short-term traffic flow state,the urban traffic flow big data management platform is built using general big data analysis and processing platform to realize integration,distributed storage and management the urban traffic flow big data. Then,combining with cloud computing technology, the random forest algorithm is paralleled using the parallel computing modelMapReduce to enhance its data analysis and processing efficiency and to improve its processing capacity to big data. Finally,using the parallel random forest algorithm the urban traffic flow big data are calculated and processed to realize the efficient and real-time forecast of urban short-term traffic flow state. The experimental result shows that the performance of paralleled random forest algorithm, including data analysis and processing efficiency,prediction accuracy of the urban short-term traffic flow state,and big data processing ability on different data sets and so on,are superior to the traditional prediction methods.