Study On Short-Term Prediction Methods Of Traffic Parameters On Expressway Based On Data Fusion

Traffic congestion affects greatly the further development of urban and travel of people. ITS can solve the serious traffic problem. Traffic parameters prediction is an important research item in the field of Intelligent Transportation Systems. It can provide real-time, accurate, reliable traffic information for ITS and its subsystems, so that ITS can quickly and accurately find out road network operation state, then adopt according measures to dismiss congestion and increase the efficiency of limited road network.This dissertation based on Great Technique Start-up Foundation Item "The Study on Technique of the Foundational Transportation Information Collection and Fusion" and National Science Foundation Item "Study on crucial Theoretical and method of intelligent highway traffic incident surveillance system". The purpose of this dissertation is to increase the precision of traffic parameters prediction from the accuracy and reliability of input information, the reasonability of model-building period and performance of prediction method. This dissertation proposed the method for identifying and modifying dynamic traffic data malfunctions which assure the qualities of input information, and build new prediction method by using statistics analysis, artificial neural network and data fusion.This dissertation contains seven chapters.Chapter 1 introduces the purpose and meaning of research, and presented the objective and content of this dissertation.Chapter 2 is the study on traditional prediction methods. Firstly, research background and status are summarized, and then several general prediction methods are described and testing by using the same data sets. Finally the factors was analyzed which can affect prediction precision, and the technique route and important research contents are proposed. Chapter 3 is studied to assure the reliability and accuracy of input information. The paper proposes a new data processing method and the rule for updating the historic database, named dynamic data malfunction identifying and modifying. The method plays an important role in increasing prediction precision due to it can improve the qualities of input information. Chapter 4 proposed Adaptive Weight Exponential Smoothing (AWES) which is from simple exponential smoothing prediction method. AWES adjust the smoothing factor continuously according to anterior prediction errors. Demonstration dedicates the prediction precision of AWES is better than fixed weight exponential smoothing (FWES). Chapter 5 adapts ANN to build traffic parameters prediction model. According to the different of input variables, prediction models are divided into Single-spot Based ANN (SB_ANN) and Multi-spots Based ANN (MB_ANN). Training, calibrating and testing of models are performing by using SNN which is a statistic analysis software developed by American. Finally, SB_ANN and MB_ANN are verified with actual data and get satisfied prediction results.Chapter 6 designs a new prediction method based on data fusion, named Multi-Model Fusion Algorithm (MMFA). In this chapter, the dynamic errors are firstly defined and two weight-choosing methods are designed based on dynamic errors. In the end, MMFA are verified by RTMS data in an expressway and compared with other methods. The results show that MMFA can significantly increase the prediction precision.Chapter 7 concludes the whole dissertation and presents the further research directions of traffic parameters short-time prediction and data fusion.