Multi-response adaptive sampling method for a radar anti-jamming evaluation model

Abstract Adaptive sampling is an efficient computational strategy for iteratively selecting informative samples under constrained budgets by leveraging existing data. To address the challenge of constructing high-fidelity evaluation models for radar anti-jamming performance with limited resources, this study proposes a novel multiple response adaptive sampling method. Our approach employs Bayesian support vector regression to estimate sample variances and gradients across the design space, enabling the design of a multi-response learning function to guide sample selection. In the learning function, sample variance and nearest-neighbor gradient process (NNGP) are combined to balance global exploration and local exploitation. Furthermore, weight factors based on the mean and maximum values of NNGP are assigned to different responses, thus balancing their accuracies. Five function examples, two real-world datasets and one engineering case study validate the effectiveness of the method. Numerical examples show that when the determination coefficients are all 0.9, approximately 50% of the sample points can be saved on average compared to other experimental design methods. Real-world datasets show that the proposed sampling method outperforms other methods in terms of prediction accuracy on the test set throughout the sampling period. In radar anti-jamming performance evaluation, the proposed method can significantly improve the model accuracy by 10%–20% under the same sample budget. The proposed method demonstrates superior performance in minimizing sampling points and experimental expenditures for multi-response modeling, showing considerable promise for practical applications in optimal experimental design.