Traffic Flow Prediction Based On Interval Type-2 Fuzzy Logic System

With the continuous progress of society,the field of intelligent transportation is also developing,and people’s demand for intelligent urban transportation is increasing.In order to improve transportation efficiency and reduce congestion,it is necessary for traffic managers to formulate reasonable management plans and strategies,so as to accurately describe the current state and predict the future state.Traffic flow has the characteristics of strong randomness,nonlinearity and complexity,which will bring some difficulties to the prediction process.As a strong nonlinear modeling tool,Type-2 Fuzzy Logic Systems have been successfully applied to chaotic time series prediction,short-term wind speed prediction and chemical identification process,and can effectively reduce the error impact caused by strong nonlinearity.Therefore,the application of type-2 fuzzy logic system to traffic flow prediction has a good development prospect and potential.The research of this paper will include the following aspects:(1)Expounding the basic concept of type-2 fuzzy sets,introducing the composition and differences of fuzzy systems of Mamdani type-1,Mamdani type-2,TSK type-1 and TSK type-2,so as to lead to the principle of fuzzy algorithm and its implementation steps.(2)In this paper,five kinds of Mamdani FLS are studied: Type-1 SFLS,Type-1 NSFLS,Type-2 SFLS,Type-2 NSFLS-1 and Type-2 NSFLS-2,and they are applied to the prediction of chaotic time series.Under the same conditions,the experimental results show that the prediction effect of Mamdani FLS is better than that of SVM,and interval Type-2 NSFLS-1has the best prediction accuracy.Secondly,the interval Mamdani FLS is applied to the traffic flow forecast at home and abroad.During data preprocessing,the number of rules set will increase sharply due to more feature information,so it is necessary to use PCA algorithm to reduce the dimension of the initial data,and then use FCM to determine the clustering center.For Mamdani FLS,the Back Propagation algorithm is used to adjust the antecedent and consequent of fuzzy rules.Under the same conditions,it is compared with other methods such as SVM and neural network.The experimental results show that type-2 Mamdani FLS is superior to type-1 Mamdani FLS,SVM,neural network and other methods.Among them,Type-2 NSFLS-1 has the highest accuracy.(3)Four kinds of TSK FLS: A1-C0,A1-C1,A2-C0,A2-C1 are studied and applied to the prediction of chaotic time series.Under the same conditions,compared with Type-2 NSFLS-1and SVM,A2-C1 TSK FLS has the highest prediction accuracy.Secondly,the interval TSK FLS is applied to the traffic flow forecast at home and abroad.Compared with Type-2 NSFLS-1,SVM and neural network methods under the same conditions.The experimental results show that type-2 TSK FLS is superior to type-1 TSK FLS,SVM and neural network,among which A2-C1 TSK FLS has the highest accuracy.In order to further improve the prediction effect,the extended Kalman filter is used to adjust the parameters of A2-C1 TSK FLS.The experimental results show that EKF-A2C1 TSK FLS has the highest prediction accuracy and the best prediction effect.