An Rapid Prediction Method for Propeller Hydrodynamic Performance Based on Deep Learning

Abstract

In order to achieve rapid and accurate prediction of the hydrodynamic performance of propellers, a set of propeller hydrodynamic performance prediction model was established based on the improved residual connection network. The residual connection method greatly improves the depth of the model. In combination with the Inception structure to simultaneously extract data features from different scales, the depthwise separable convolution reduces the model parameters. The sample space for training the deep neural network is built based on the propeller geometric parameters and model test results. An improved beetle swarm antennae search algorithm is proposed to optimize the initial weights and thresholds of the model to further improve the prediction accuracy of the model. The research results indicate that the improved beetles swarm antennae algorithm significantly improves the accuracy of the model and solves the problem of overfitting of it. The prediction results of the model are in good agreement with the experimental values, and its prediction performance for the propellers which are not in the dataset is basically the same as that of the CFD method. The model has an excellent universality and its calculation period is extremely short, which can meet the requirements of real-time and accurate prediction of propeller open water performance.

Keywords

• propeller
• hydrodynamic performance
• residual neural network
• inception structure
• the improved beetle swarm antennae search algorithm