| With the development of urbanization,urban traffic congestion has become acute.The problem not only reduces the citizens' travel efficiency,but also brings environmental pollution,causing serious economic losses.On the other hand,with the development of the Internet of things,the various equipment installed in the road collect a steady stream of traffic flow data.The collected data has huge quantity and strong timeliness,so it is a typical big stream data.We can use the technology of big data to dig out the value behind the data to ease the traffic congestion issue.So this thesis tries to build appropriate congestion prediction model based on machine learning algorithms,using streaming computing technology to analysis the real-time traffic data,so as to establish an online traffic prediction system.The major works of this thesis are as follows:According to the characteristics that the traffic flow data changes quickly,we present small data set in this thesis,using the sliding window technology and random forest algorithm to construct SW-RF prediction model.The model not only has high accuracy,but also reduces the modeling time overhead through the algorithm parallelization.We also design the adaptive sliding window random forest forecast model to problems of concept drift in data flow.The experiment shows that the improved model can better adapt to the data flow of mutation.We study the mechanisms of the various stream computing framework which are popular nowadays.On this basis,we design the traffic congestion forecast system in real time whose core is Storm.This System can handle and process the collected data of traffic flow,and then use the handled data for training predict model.Finally we show the predicted results to the user by the client in real time.In design,we try to build a low latency,scalable,highly available distributed streaming computing system according to the message queue Kafka and the character of the Redis database of memory. |
PDF Full Text Request