| The original traffic flow time series are usually polluted by various types of noise,separating the noises from the original traffic flow and improving the accuracy of traffic flow prediction is the key point of this study.The main work and innovations of this paper are as follows:(1)Using the method of square root transformation to transform the non-stationary time series into a stationary series.For a non-stationary time series containing a white Gaussian noise,the square root transformation will bring a systematic error which should not be ignored,while the systematic error has been neglected for more than 70 years.In this paper,the Taylor series expansion of the square root transformation is used,we deduce the mathematical expectation and variance of the white Gaussian noise from the preceding four terms of the Taylor expansion.Removing this systematic error can reduce the root mean square error by 0.1%–1% in the forecasting or prediction of the traffic flow time series.(2)Based on partial least-squares regression method to establish a prediction model,using the PLS method to analysis the correlation coefficient between the various models,and establish the regression equation.The validity of the model is verified by the traffic flow forecasting,the mean error and the root mean squared error are reduced,and the model shows good prediction effect.(3)The historical data inputted to a prediction model,as well as the prediction model itself,are the two factors to cause prediction errors.According to the Shannon sampling theorem,increasing the sampling frequency of traffic flow series,can improve the effect of prediction of traffic flow.According to the central limit theorem,as the decrease of statistical intervals of traffic flow time series,the fluctuation of traffic flow decreases exponentially.In conclusion,to increase the sampling frequency moderately can improve the effect of short term traffic flow time series prediction reliably. |
PDF Full Text Request