A LHSRF-based Method of DoE for GNC Systems

For the factor screening of complex systems such as Guidance and Navigator Control (GNC) systems, the number of factors is large and the scale of experimental samples is small. Moreover, the high cost and long experiment period increase the difficulty of factor screening in complex systems. In this paper, we propose a method named as LHSRF to evaluate the importance of a factor based on partial two-level experimental data of a complex system where the factor has only two high and low levels of values. In our method, Latin Hypercube Sampling (LHS) is used to generate samples where the value space of factors is well filled, and then the random forest (RF) model is used to predict the output of sampling points. The original data and sampling point data are combined and the importance of factors is judged by another RF model to achieve the purpose of factor screening. This method is applied to two numerical examples and performs well compared with other methods. Eventually, we apply this method to missile guidance simulation system and successfully screen out important factors.