APPLICATION OF MACHINE LEARNING ALGORITHM IN OPTIMIZATION OF PSV FOR 110000DWT OIL TANKER

摘要

The pre-shrouded vane (PSV) in front of propeller is a kind of energy-saving device which can change the inflow to improve the received power of the propeller. The device needs to be optimized according to the flow field of the stern. Most of the existing design methods rely on the experience of the designer. In order to improve the design efficiency of PSV and obtain a design scheme with higher energy-saving effect, this paper presents an optimization and analysis method for PSV in front of propeller based on agent model. Aiming at an 110000 dwt oil tanker, 11 design parameters such as stator angle and duct radius are determined by means of parameterization. The design parameters are sampled by Latin hypercube sampling method (LH), and the sample space with 300 samples is generated. The energy-saving effect of each sample is analyzed by CFD method. The data set is formed and next divided into training set and test set. Then, machine learning methods are used to build the agent model of sample space. The error of each model in the test set is analyzed. To obtain the best model, the performance of several models in the test set and training set is considered. The applicability of different models is also highly considered. On this basis, the sensitivity analysis method is used to analyze the sensitivity of each design parameter. Then, the main influencing parameters are found. Finally, particle swarm optimization and genetic algorithm are compared to optimize the design parameters of PSV for 110000 dwt oil tanker. The optimization results are verified by CFD method. The results show that the artificial neural network model is better on this dataset, and the model error on test set is less than 1% compared with the CFD result. The optimal solution by genetic algorithm method is better than all the sample points, and a better design scheme of PSV is obtained.