A prediction model of hard landing based on RBF neural network with K-means clustering algorithm

This paper proposes a prediction model for forecasting the hard landing problem. The landing phase has been demonstrated the most dangerous phase in flight cycle for fatal accidents. The landing safety problem has become one of the hot research problems in engineering management field. The study concentrates more on the prediction and advanced warning of hard landing. Firstly, flight data is preprocessed with data slicing method based on flight height and dimension reduction. Subsequently, the radial basis function (RBF) neural network model is established to predict the hard landing. Then, the structure parameters of the model are determined by the K-means clustering algorithm. In the end, compared with Support Vector Machine and BP neural network, the RBF neural network based on K-means clustering algorithm model is adopted and the prediction accuracy of hard landing is better than traditional ways.

I. Introduction

The fatal accidents data demonstrates there are about 23 percent of all fatal accidents take place in the landing phase from 2003 to 2012, though this phase just accounts for about 1% proportion in flight cycle [1]. One of the most dangerous potential flight accidents factors is hard landing. According to the servicing manual of Boeing Company, hard landing is defined as an event such that the vertical acceleration or speed exceeds the prescribed threshold. It may cause the damage of the wings, landing gear or plane structure [2]. To analyze the dynamic complex land process, the Quick Access Recorder (QAR) is widely used for aircraft health status monitoring. The QAR data contains variety of flight information. It is used in many methods to analyze landing status and aid decision-making in land process.