一种归一化的代理模型精度指标

衡量代理模型精度的量化指标种类繁多,但是要横向对比不同种类数据的模型精度,常见的精度指标存在一些不足之处。分析现有代理模型精度指标的不足,提出一种“归一化绝对误差均值暠指标;以某战术导弹模型为例,采用参考文献中的试验设计加点策略与交叉验证策略,建立11种气动参数的Kriging代理模型;通过对“归一化绝对误差均值暠指标与相关性图进行比较,验证所提指标的有效性。结果表明:提出的指标不仅能够有效表征代理模型精度,而且能对不同种类数据进行横向对比,具有一定的应用价值。

A Normalized Accuracy Index for Surrogate Model

There have been a considerable number of accuracy indexes for surrogate models in references. However, most of these indexes are not capable of comparing surrogate models of different data types. In this paper, the drawbacks of the existing indexes were firstly discussed. Then a new index named "Mean Normalized Absolute Error (MNEA)" was introduced. Kriging models of 11 different types of aerodynamic parameters from a tactical missile model were built, with the adding-points design of experiment method and cross-validation verification methods. The effectiveness of MNEA was verified by comparing the MNEA value and the correlation plots of the example, which had proved that MNEA can be used to compare the accuracy of surrogate models of different data types.