Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2020-09-28 Cooperative journals: 《计算机应用研究》
Abstract: A novel and effective technique for short-term traffic flow forecasting is presented. The main contribution is an extension of Kalman filter, such that it becomes to be able to identify the outlier and then filter out it; the present technique is named as Outlier-identified Kalman filter. The fluctuations of the traffic flow, which leads the system uncertainty, is often filtered out by the classic Kalman filter, and the subtle clues indicating the sudden change of the traffic flow may lose in this operation. To achieve better forecasting accuracy, discrete wavelet transform is used to process the original signal and the useful signals are preserved while de-noising. In addition, historical reference values are applied to correct predicted values. Extensive experiments on four benchmark datasets demonstrate an average improvement of 2.919% in MAPE and 79.58 in RMSE.