An Optimization Research of Evaporation Duct Prediction Models Based on a Deep Learning Method

Abstract

Evaporation duct is a particular atmospheric layer that is crucial for marine vessel communication. Evaporation duct height (EDH) is a vital index to measure evaporation duct architecture and intensity. The most common method to calculate EDH is model diagnosis, and the most prevalent model is Paulus-J eseke (P-J) model. This paper raises a new method to optimize the P-J model based on multilayer perceptron (MLP), a typical neural network in deep learning. To assess the performance of this method, some observation experiment data-sets are introduced here, which were conducted in some sea areas during March 2013 to May 2013. Root-mean-squared error (RMSE) is used as the evaluation index, and comparison between MLP P-J and classic P-J model indicates that the estimation precision of EDH based on this method is significantly higher than that of classic P-J model over all observation experiment areas, which implies that this method with deep learning is a better approach for evaporation duct diagnose model research.