| ITS (Intelligent Transportation System) is an advanced road transportationsystem which including many technical fields, such as electronic information,communication, intelligent control technology and etc. Traffic congestion affectsgreatly the further development of urban and travel of people. ITS can solve theserious traffic problem.Traffic Flow Guidance System (TFGS) take part in an important role in ITS.Traffic parameters prediction is the core problem in the TFGS. How to predict trafficparameters amount online is the key solution to the TFGS. It can provide real-time,accurate, reliable traffic information for ITS and its subsystems so that ITS canquickly and accurately find out road network operation state, then adopt accordingmeasures to dismiss congestion and increase the efficiency of limited road network.Depending on the data source used in the prediction process, short-term trafficparameters prediction models can be mainly categorized into three groups:1) those using only historical data,2) those using only real-time data,3) those using both historical and real-time data.A large body of past work is available on the short-term prediction algorithms,such as time-series model, Kalman filtering theory, simulation models, dynamictraffic assignment models, neural network models, and nonparametric methods. Theneural-network models have been employed in many research studies.The thesis first introduces the significance of this subject, development statusand some basic theory knowledge. The important part and mostly task of this thesisare the prediction of short-term traffic parameters. The necessity and feasibility areanalyzed in this thesis. And then the particular contents are confirmed.The dissertation mainly introduces the algorithms of traffic parameters prediction,includes four parts: Firstly, introduction of the traditional methods. Secondly thedissertation introduces the algorithm of traffic parameters prediction based on BPnetwork and optimization of astringency based on L-M algorithm. And then itpresents the algorithm of traffic parameters prediction based on RBF network. And atthe end a new prediction method based on data fusion, named Multi-Model FusionAlgorithm (MMFA)is designed. Part one introduces the traditional methods of traffic parameters prediction.Firstly, research background and status are summarized, and then several generalprediction methods are described. And then, as an example, the exponentialsmoothing prediction method is verified by data in an expressway. The defection ofthis method is that the coefficient is a constant value. It couldn't follow the change ofthe traffic parameters. So this method may be affected by randomicity of the trafficparameters. Part two introduces the basic knowledge of neural networks. In this part themathematical model, networks structure and algorithms are presented. And then thealgorithm of traffic parameters prediction based on BP network and optimization ofastringency based on L-M algorithm is realized, which solves problems of slowastringency and local minimum. According to the different of input variablesprediction models are divided into based on single-spot ANN (SB_ANN) and basedon multi-sports ANN (MB_ANN). Finally, SB_ANN and MB_ANN are verified withactual data and get satisfied prediction results. From the results we can see that BPnetwork can well describe the complicated traffic parameters. On the whole, BP network can successfully solve the prediction of nonlinearsystem such as traffic parameters. Part three presents the algorithm of traffic parameters prediction based on RBFnetwork. Prediction models are divided into based on single-spot ANN (SB_ANN)and based on multi-sports ANN (MB_ANN). Finally, SB_ANN and MB_ANN areverified with actual data and get satisfied prediction results. The results proves thatthe RBF algorithm gain advantage in training speed over BP algorithm. It is moreflexible than traditional algorithms. Part four designs a new prediction method based on data fusion, named V... |
PDF Full Text Request