小样本规模船型优化策略的选择研究

Study on Selection of Hull Optimization Strategy with Small Sample Size

选择合理高效的优化策略进行小样本规模下的船型优化,以KCS船的阻力性能为优化目标,对比分析不同优化算法的优化效率和阻力性能提升结果。首先,利用CAESES软件参数化变形技术实现船型的自动变换,分别对球鼻艏、船体入流段与去流段、艉封板以及船体其他位置进行组合变形,同时考虑排水体积和浮心纵向位置的约束;其次,采用船舶水动力专用CFD软件SHIPFLOW对船型的总阻力性能进行评估,与模型试验数据对比验证预报方法的精度;另外,分别采用CAESES软件集成的多种主流优化算法和自研优化算法Silverbullet的不同优化策略进行最优值的探索,对比分析不同样本规模下的优化效果。结果表明,在样本规模百次左右的情况下,Silverbullet算法的优化效果优于CAESES软件中集成的主流算法,当Silverbullet算法开启智能边界突破技术后,能够进一步获得总阻力性能的提升。

For the hull optimization task,considering the numerical simulation cost of single solution and the limitation of the design period,obtaining the better optimization result with fewer design solution iterations is the goal sought by designers.The hull optimization problem is usually a high-dimensional problem with multiple design parameters,and the number of samples required for optimization is usually exponentially related to the number of design parameters,so it is particularly important to choose a reasonable and efficient optimization strategy for small sample size.This paper takes the KCS ship as the research object,takes its resistance performance as the optimization target,and compares the optimization efficiency and resistance performance improvement of different optimization algorithms of CAESES software commonly used in the industry and the in-house optimization algorithm Silverbullet.Firstly,CAESES software parametric deformation technology is used to realize the automatic transformation of the hull shape.A total of 8 design parameters are combined to deform the bulbous bow,the in-flow and out-flow sections of the hull,the transom and other positions of the hull,and the constraints of the displacement and the longitudinal position of the floating center are considered at the same time of the transformation.Secondly,SHIPFLOW,a special CFD software for ship hydrodynamics,is used to evaluate the total drag performance of the ship model and verify the accuracy of its forecasting method by comparing it with the data from model tests.On this basis,different optimization strategies of various mainstream optimization algorithms integrated with CAESES software and the in-house optimization algorithm Silverbullet are used to explore the optimal values and compare and analyze the optimization results under different sample sizes.The results show that the optimization effect of the Silverbullet algorithm is better than the algorithms integrated in CAESES software for a sample size of about one hundred,and the total resistance performance can be further improved when the Silverbullet algorithm is enabled with the intelligent boundary break technique.