Support Vector-Enhanced Batch Adaptive Sampling for Global Surrogate Modeling and its Application on Remote Sensing Satellite Effectiveness Evaluation

Global surrogate model is used to replace complex spatial systems for reliability analysis, performance evaluation and optimization, which requires sufficient samples. To reduce the overhead of expensive simulation experiments with high fidelity, a batch adaptive sampling method that synthesizes the informativeness and diversity of samples is proposed. Support vector-enhanced batch adaptive sampling (SV-BAS) method utilizes the properties of support vector machines to add sampling points in batches using a hybrid sampling strategy that combines local exploration and global exploitation. Utilizing the pipeline principle of support vectors, the support vectors are first filtered using a low-precision support vector machine. Then, a high-precision support vector machine is used to compute the predicted errors of support vectors with no computational cost, which is used for local exploration. The sparsity of support vectors and the sampling center method based on the distance criterion are used to ensure the diversity of sample points in the batch sampling. The performance of the proposed method is verified by the simulations using six numerical functions with varying features and dimensions. The results show that compared with other single-sample adaptive sampling methods, the proposed algorithm performs better in terms of model accuracy as well as sampling efficiency. Further, the proposed algorithm is used for remote sensing satellite effectiveness assessment and has good application prospects.