A novel iterated multi-step prediction method of traffic flow

Multi-step prediction of traffic flow is very useful for dynamic navigation systems and intelligent transportation systems. A general iterative multi-step prediction is normally based on an iterative use of some one-step prediction and it usually results in error accumulation. To obtain a better method of iterative multi-step prediction and its availability, firstly based on Kalman filtering model and support vector machine model, a new combination model is designed for one-step prediction; and then a method of iterated multi-step prediction combined with the use of historical similar sequence is presented. The use of historical similar sequence in the method can significantly reduce its error accumulation. Experiment results show that its maximum prediction error growth rate within 12-step predictions (about an hour with 5 minutes as a time-interval) does not exceed 5% for the whole day and its relative error is maintained at less than 10% in peak hours. Copyright © 2014 Binary Information Press.