Predicting Peak Service Rate Based On Weather Impacts Using Machine Learning Techniques

Abstract

As the air traffic congestion and large-scale flight delays become more and more serious, it is particularly important to predict the highest sustainable throughput grades of terminal which improved the effect of TFM. Current research has focused on predicting the impact of runway configurations on airport capacity. However, the selection of runway configuration does not take into account all the weather conditions that affect the terminal zone operation, and the transition from runway configuration to airport capacity is also a complex study. This article describes a methodology for predicting peak service rate based on the meteorological conditions directly. Two machine learning algorithms were introduced and evaluated for use in the process of developing the model. K-means algorithm is one of the unsupervised learning algorithms in machine learning filed which can be used for clustering. Random forest is a non-traditional machine learning algorithm, composed of many decision trees, which can be used for classification and regression. First, a k-means algorithm is applied to all days in 2016 of Guangzhou airport, resulting in the identification of 3 clusters that represent unique classifications of peak service rate that were historically implemented. Second, a forecast model based on the weather is developed by the application of random forest algorithm. The model provided 7 input features that describe the weather and evaluated the importance of each. Finally, after calculating the 5301 data from Guangzhou airport, the predict accuracy of model indicates that this methodology is feasible but still needs some improvements.