Passenger Flow Forecasting Research For Airport Terminal Based On Data Mining

Dynamic allocation and scheduling of airport terminal passenger serviceresources is one of the effective ways to improve passenger service levels andoperational efficiency within the terminal, while the relatively accurate passengertraffic forecasting is the prerequisite for dynamic allocation and scheduling.Currently, the study of the airport terminal departure passenger traffic forecastingare based on a small sample, which failure to use the mass actual data effectivelystored in the database system. The passenger traffic arrival pattern used forforecasting can only be applied in limited conditions, the method is difficult toensure the forecasting results’ accuracy, and in practice it’s difficult to support therealization of the theoretical approach, passenger traffic forecasting cannot meet therequirements of dynamic allocation and scheduling of airport terminal passengerservice resources. Therefore, this article will study three aspects: a kind of datamining system that integrate the airport information system passenger historical data,the exploratory analysis of passenger traffic pattern and the forecasting model basedon data mining methods.The details are as follows:First, in order to use the large number of passenger traffic actual historicaldata stored in the system effective for data mining research, this paper choosesdata mart integration methods and build the mart demand analysis, presented asystem that can mine the Integrated Airport Information Systems historical data.The system can support users to use multiple data mining methods mininghistorical passenger traffic data and forecasting in different dimensions, differentdimensions of the conceptual level, and show the data mining results quickly andintuitively. The constructed system can effectively achieve the forecasting targetin real airport terminal and show the forecasting processes.Previous studies acquired the sample under the specific conditions and thepattern get from the studies is not universal applicable. Therefore, in this paper,through the exploratory data analysis of the large number of actual data, from theaspects of the number of flights, passenger traffic, passengers arrive in advancebehavior pattern we get that the forecast will be based on passengers arrivebehavior pattern, On this basis, presents and analyzes the factors affecting passengers arrival behavior pattern such as departure time, airline. Through thestudy concluded that under the influence of many factors, passengers arrival donot have some definite patterns, the forecast based on passengers arrival behaviorpattern cannot be according to probability distribution function, to solve thisproblem the application of data mining methods is suitable.Finally, according to the forecast ideas determined by the exploratory dataanalysis, a passenger traffic forecasting model is built that consists of the clusteranalysis, decision trees and K-Nearest Neighbor algorithm. Segment is the basicforecast unit, the passenger arrival rates of each time period in advance is theforecast target, each time period interval is30-minutes. By comparing theforecasting passenger traffic and the actual passenger traffic, Verify the accuracyof the forecasting model that can meet a single flight, a single airline and thewhole airport terminal passenger traffic forecast in30-minute intervals.