Long-Term Traffic Volume Prediction Based On Type-2 Fuzzy Sets

Long-term traffic volume prediction plays a vital role on urban road planning,design and management.It allows users to have a global insight of the traffic patterns and traffic state information.Traffic volume has character of randomness and uncertainty of the typical giant complex system.Therefore,most of long-term prediction models are often difficult to obtain high-precision prediction results.This thesis introduces fuzzy set theory to deal with long-term traffic volume prediction problem,and proposes three long-term traffic volume prediction models with type-2 fuzzy sets.The main work contents are as follows:Firstly,this thesis combines K-means clustering with Gaussian interval type-2 fuzzy sets theory to get a long-term traffic volume prediction model.K-means clustering algorithm is used to process traffic volume into interval data that can represent the fluctuation range of traffic volume,and the parameters of embedded fuzzy sets are obtained by clustering center,which obtains the type-2 fuzzy sets model through the embedded Type-1 Fuzzy sets.The simulation results show significant improvement.Secondly,based on the transfer law between traffic volume states,this thesis proposes a combined type-2 fuzzy sets model based on Markov model to extract fuzzy rules.This model uses Markov model on the same forecast time horizon one day and the same sampling time horizon multi-day.Looking for the hopping law between traffic states,the fuzzy rules are generated by the Markov model,and the combined type-2 fuzzy sets model based on Markov model is gotten to make more accurate predictions.Finally,the combined type-2 fuzzy sets model based on fuzzy-fusion algorithm is proposed.This model constructs type-2 fuzzy sets by data driven way.Each time segment constructs a different number of type-2 fuzzy sets.According to the simulation results,the combined type-2 fuzzy sets model based on fuzzy-fusion algorithm's prediction results are the most accurate,and its fluctuations are small.This thesis uses the actual traffic data of a large city in China to carry out model test.The methods proposed in this thesis can predict the long-term traffic volume well.