Short-term Traffic Flow Prediction Based On Data Mining

Along with the rapid development of social economy,the continuous expansion of city size,car ownership also experiences rapid growth.While bringing many conveniences,Cars also bring many problems such as environmental pollution,traffic congestion and frequent traffic accidents,energy shortages.Intelligent transportation system is now globally recognized as a useful measure to enhance the efficiency and management of road traffic congestion problems.As one of its key components,the traffic flow prediction plays an important role in real-time control and intelligent management of urban roads.To get good results for real-time control of urban road network,the accurate,real-time traffic flow prediction is needed.This thesis makes a detailed analysis on urban road traffic flow data,taking into consideration of the the characteristics of data mining models and transport stream data at the same time,prediting the representative data of transport flow and speed through the linear regression model and M5 P model tree.First,Python is used to obtain original information,and parameters such as traffic flow,speed,occupation and headway that are frequently used in the predication of intellectual transport system are selected here based on the characteristics and principles of transportation.Dirty data involved are dealt with in accordance to data processing principles,during which outlier data are picked out through threshold method,absent data and redundant data are processed based on the data of previous traffic condition,data of nearby roads and linear interpolation method.Reasonable traffic flow data preprocessing provides reliable data support for subsequent prediction.Based on the strong time correlation of traffic data,TimeseriesForecasting originated from weka is proposed here.Then data are used to formulate M5 P model tree and Linear Regression Model in the algorithm which are further used to predict the traffic flow and speed in a short period.Linear regression of optimized Ridge parameters is used in the set of Linear Regression Model which enables CV Parameter Selection automatically adjust the parameter of Ridge so as to find the best R and finally improve the generalization ability of the model.Data used in the prediction process are previous data of typical roads and those of relative roads.Those data are used to predict the traffic flow and speed.The prediction process includes data preprocessing,sample defining,characteristics exploring,sample generation,simulative training as well as presentation and evaluation of the results.90%sample data are used in the process of simulative training and evaluation.The analysis on the absolute average error of the model prediction shows good prediction performance of the M5 P model.In addition,a higher accurate traffic flow prediction can be obtained through theprevious data of the roads,while a higher accurate sped prediction can be obtained based the relative roads.Finally,the rest 10% sample are used to verify the M5 P model tree through the evaluation program.The conclusion shows that M5 P model tree does well in the prediction of both speed and traffic flow.