Research On Travel Time Prediction Algorithm Based On Online Sequence Extreme Learning Machine

The Traffic Flow Guidance and the line Guidance are the two typical models of the modern ground transportation management system in the 21 st century.One of the most important content of them is path route travel time.The traffic flow data is analyzed in order to get the travel time.However,the traffic flow is uncertain,random and real-time,so it caused the path route travel time is very complex to predicate.The traditional predicated methods have too many shortcomings to obtain good prediction result,such as The Kalman filtering algorithm's applicability is not good enough,the BP neural network can describe the trend of travel time,but its prediction accuracy is not very good.This paper describes the significance of travel time prediction methods aims at the characteristics of urban road network,and analyzes the principle and characteristics of the existing vehicle travel time prediction methods,applies the online sequence Extreme Learning Machine to travel time prediction,puts forward the online sequence Extreme Learning Machine Travel Time Prediction(OSELM-TTP).Using characteristics of the data of Curing samples of Hidden layer node output matrix to handle the real-time request of travel time effectively.Adopting Data Sliding Window Mode to add data according to the size ofwindow in online selection sequence learning stage,to update parameter of algorithm constantly to get the prediction effect.In this paper,based on the MATLAB toolbox and the measured traffic flow data,the travel time prediction and simulation of real sections is discussed.The online learning sequence limit local and online support vector machine regression(Online Support Vector for Regression,OSVR)and BP neural network(Back propagation Neural Net,BPNN)are comparably analyzed.Results show that the OSELM-TTP has good adaptability,timeliness and accuracy of online learning sequence.It has changed he online support vector machine regression algorithm process complexity,high time complexity and BP neural network poor forecast accuracy limitations.Its overall prediction performance is better than the online support vector machine regression algorithm and BP neural network,which further proved the ultimate learning machine prediction link travel time based on the feasibility of online sequence.