| With the rapid development of economy nowadays, the travel mode of resident has gradually diversified. Factors such as the large population in China, increasing number of vehicles every year and expansion of city scale has made the traffic congestion become a main issue in majority of cities. The key to solve traffic problems is to strengthen the construction of urban rail transit which has advantages like big capacity, fast speed and low occupation area, although it has features including high construction cost and complexity and it is unchangeable after finishing. Moreover, the network planning and programming in urban rail transit depends on volume of passenger flow, hence it is crucial to predict passenger flow scientifically.In the aspect of short-term passenger flow forecast, the article gives an review about time series method, gray model and neural network model, then it analyzes deficiency about traditional gray model in initial value setting. That is, the initial value in traditional gray is set on the premise that the fitting curve passes through the first point of given data sequence.While any point could be the one having high fitting precision in the given data sequence,therefore the article respectively uses Matlab and SPSS software to fit all points in the given data sequence into gray model fitting curve, determining prediction value of future annual passenger flow by the two highest fitting precision points. And finally it establishes the model which combines time series model, improved gray model and Elman neural network model.The numerical example in the article is based on historic data of passenger flow in Chongqing Rail Transit Line 2, which verifies effectiveness of the improved gray model and composite model. The results, in comparison with traditional gray model, shows that the mean absolute error between fitting value in improved gray model and actual one is reduced by 9%, and the error between forecasting model in composite model and actual one has reduced 61.4%,76.6% lower than the single use of time series model. Meanwhile, the composite model is also used in the article to predict future passenger flow in Chongqing Rail Transit Line2.In the aspect of long term demand forecasting of passenger flow, the article focuses on common models used in four-stage method. In the process of trip generation, the most commonly used one is linear function model. And in the course of modeling, the article makes improvements in variable selection, combining SPSS software to analyze relativity of independent variables, to exclude high correlative variables and to introduce accessibility as independent variable. In the stage of trip distribution, forecast results obtained only by gravity model cannot guarantee the consistence between generation volume, attraction volume and sum of traffic volume in every traffic zones, so the article corrects attraction volume of every zone by the combination of gravity model and growth rate method, introducing logarithmfunction and composite impedance function into gravity model. In the stage of traffic modal split, the article makes some improvements for utility function of Logit model, which bases on the defection that selection probability of Logit model is related only with utility difference.And validity of improvements are proved by numerical examples.Finally, the article combines theory of passenger transfer and cubic attraction method to establish the forecast model of transfer passenger flow and then it has forecasting analysis about transfer passenger flow for the new constructed Lanzhou Rail Transit. |
