基于变保真度模型的AUV流体动力参数预测  

Predictions of AUV's Hydrodynamic Parameters Based on Variable-fidelity Modeling

作　　者：宋保维[1] 王新晶 王鹏[1] 

机构地区：[1]西北工业大学航海学院,西安710072

出　　处：《机械工程学报》2017年第18期176-182,共7页Journal of Mechanical Engineering

基　　金：国家自然科学基金资助项目(51375389)

摘　　要：变保真度模型(Variable-fidelity modeling,VFM)作为一种有效的代理模型方法被广泛用于工程设计,其中低保真度模型(Low-fidelity modeling,LF)用于获取目标模型的整体趋势,高保真度模型(High-fidelity modeling,HF)用于校正低保真度模型的准确度,二者通过桥函数(Bridge function,BF)融合,完成对目标模型的近似。将VFM用于预测自主水下航行器(Autonomous underwater vehicle,AUV)在外界工况(速度和攻角)下的流体动力参数(阻力系数、升力系数、力矩系数),其中高低保真度模型使用径向基函数构建,桥函数使用克里金函数构建,从遗传算法交叉算子产生的候选样本集中选择变保真度模型与高保真度模型误差最大的样本进行更新,最后建立变保真度代理模型。结果显示VFM能够准确描述流体动力参数在设计空间的变化;在相同的计算条件和时间下,VFM比高保真度模型精度高;随着高保真度样本的增加,模型的精度稳步提升。Variable fidelity modeling（VFM） which is regarded as an effective surrogate model is widely used in engineering design. In VFM, low-fidelity modeling＇s（LF） role is to capture the trend of objectives while the high-fidelity modeling＇s（HF） role is to revise the accuracy of LF. LF and HF are fused through bridge function（BF） and then finish the approximation for true model. This paper adopts VFM to predict autonomous underwater vehicle＇s（AUV） hydrodynamic parameters（drag, lift and torque coefficients） on certain velocity and angle of attack, in which LF, HF are built using RBF and BF is built using Kriging. Crossover operator of genetic algorithm produces candidate sample set and the sample with maximum error between VFM and HF will be chosen as newly added sample. Finally, VFM is established with all samples. The results show that compared to HF, VFM can describe precisely the change of hydrodynamic parameters in design space. Under same computing condition and time, VFM can gain better precision than HF and with the increases of HF samples in VFM, the accuracy of VFM becomes higher.

关 键 词：代理模型 变保真度模型 桥函数 AUV流体动力参数 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]