Combination Forecasting Of Short-term Traffic Flow By Robust Statistics

Along with the progress of human civilization, urban transport service isdeveloping rapidly, but it brings a series of traffic accidents at the same time. The ITSplays an important part in reducing traffic accidents and improving the smooth flow ofthe road. Short-term traffic flow prediction theory is at the core of the ITS, whichplays an important role in the traffic information service, traffic control, and trafficguidance.Currently, the combination forecasting is the mainstream of short-term trafficflow prediction. In this article, Robust Statistics theory and Robust scale estimatorsare applied to forecast the traffic flow better, which will enhance the accuracy andstability of short-term traffic flow prediction.The main contents and results are:(1) Given Mathematical Statistics theory，Robust scale estimators are applied tocalculate the weight value in the combination forecasting, different Weight valuesreflect the influence of the different Monomer models. Appropriate Weight values areconducive to enhance the short-term traffic flow prediction.(2) There are five classic prediction models: AR(3), ES1, ES2, MA1, MA2.Based on the five prediction models, we do numerical simulations to forecastaccording to the traffic flow data, calculate and analyze prediction error informationof the various models.(3) In the mathematical statistics, in order to close approximately to thecumulative distribution function of standard normal distribution, the Fisher ztransformation is the widely employed explicit elementary function, and is used toestimate the confidence interval of Pearson product moment correlation coefficientand to estimate the confidence intervals of the Dickinson best weights for linearcombination of forecasts. A new Sigmoid-like function is suggested to replace theFisher z transformation, and can be4.677times more accurate than the Fisher ztransformation.