Abstract: Prediction of passenger traffic volume on route is one of the most important technologies of route network optimization. Traditional prediction method is based on passenger booking data. Generally used models include regression method, time series method and so on. However, these models consider less about the randomness of route flow data and the continuous growth of passenger volume. In order to solve these problems, combined forecast model is proposed based on regression method, which rely on two different reference periods.Construction of the model is divided into four stages: a. using load factor data to forecast and the visiting rate data of first-order accumulative smoothing makes the target curve becomes smooth and monotonous; b. using DOWmethod to predict the target year data; c. using the fitting curve of the adjacent point value to simulate the annual growth amount and build a forecasting model; d. taking average weighted value based on forecasting results from the above two stages, and establishing a new combined forecasting model. XMNPEK segment guest rate data of an airline in 2011-2015 is used to predict the load factor data for the first half of 2016. Comparedmwith traditional regression method and time series method, mean absolute error of the current method is reduced from 4.76 and 4.21 to 3.77 and the prediction accuracy is improved obviously.

Key words: passenger traffic volume on route, combined forecasting, linear regression, time series