Prediction for Sea Surface Temperature of Submarine Thermal Wake: Combining CFD with Neural Network

Abstract

Predicting the maximum temperature of the submarine’s thermal wake rising to sea surface has a high early warning significance. Computational Fluid Dynamics (CFD) simulation can obtain accurate prediction, but is usually time consuming. This paper proposed a prediction method combining neural network with CFD to reduce calculation time. Firstly, five most important variables influencing the temperature of thermal wake were specified and a CFD model was established to calculate maximum temperatures on sea surface under different values of these variables. Then, the variable values and temperatures were constructed as a dataset to train a 5-input and 1-output neural network. Trials were carried out to find the optimal hyper-parameters for network during training. Results show that the prediction of optimal model has R2=1 on training set and R2=0.99 on test set. The calculation time reduces 6 orders of magnitude compared to the CFD model.