Compactness prediction of airport soil field based on artificial neural network

In order to improve the dynamic prediction and management capabilities of soil field safety in airfield, a new artificial neural network (ANN) model was established based on the evaluation data of six years in an airport. By factors analysis of the actual soil field of the airport, natural density (ND), actual water content (AW), optimal water content (OW), rainfall condition (RC) and compaction condition (CC) were chosen as the input data, and hyperbolic tangent sigmoid function was set as the transfer function. The network was trained by 400 sets of data and validated for its accuracy by 100 sets of data selected randomly from the database. The prediction capability of achieved ANN model was analyzed by Nash-Sutcliffe efficiency coefficient (NSE) method. And engineering application had been done in another airport. The results show that soil field compactness can be effectively predicted by well-trained ANN model with R-Squared of 0.98 and NSE of 0.89. The outcomes of validation test in another airport prove that the errors of most sample zones are between -5% and 5%, with only one exception of 15%, with calculated NSE of 0.86, which satisfies the requirements of engineering application. By optimization of ANN model with factor analysis method, it indicates that ND and AW are the controlling factors of model compactness prediction, and the best ways of improving the safety of soil field in airport are soil gradation and drainage control strictly.