Modeling and interpretability analysis method for anti-jamming response of radar seeker

The anti-jamming performance of radar seekers is influenced by comprehensive factors such as interference, background, and targets. How to construct a key anti-jamming performance indicator model, namely an anti-jamming response model, in combination with the requirements of practical combat evaluations, in order to evaluate the anti-jamming boundary performance of seekers, is a challenging issue of great concern to missile development and testing units. Currently, research on modeling the anti-jamming response of radar seekers is still very limited, and related research achievements are very rare. Some mechanism-based studies focus only on single-factor, single-response scenarios, which severely restrict research on designing anti-jamming test schemes for seekers that combine numerical simulations with practical tests based on models.

Based on Gaussian process regression, an anti-jamming response model for radar seekers is constructed, and the model is subjected to interpretability analysis using the LIME and SHAP methods. Through the construction and analysis of an angle tracking accuracy model for a certain active radar seeker against shipborne active intermittent suppression jamming, the feasibility and effectiveness of the proposed model and methods are thoroughly verified.

A framework for constructing an anti-jamming response model for seekers based on Gaussian Process Regression and an interpretability analysis has been proposed. A model for the angle tracking accuracy of radar seekers against shipborne active intermittent suppression jamming has been established. Upon comparison, it has been found that the Gaussian process regression model exhibits higher accuracy compared to support vector machine regression (SVMR) and kernel regression models (including SVM kernels and least squares kernels). The linear regression coefficient between the predicted and actual values is 1.027, with an R-squared value of 0.93415, a root mean square error of 0.0728, and a sum of squared errors of 0.5307. When the ratio of training samples is less than 0.7, the model validation error decreases rapidly. However, when the ratio exceeds 0.7, the decrease in model validation error becomes very slow, indicating that further increases in the sample size have limited impact on improving model accuracy at this point. Compared to the LIME method, the SHAP method allows for analysis of both individual test samples and the entire test sample set, yielding richer and more effective conclusions.

Gaussian process regression can be utilized to establish a high-precision anti-jamming response model for radar seekers. Based on this model, verification and analysis can save experimental samples to a certain extent. Furthermore, through interpretability analysis, the degree of influence of various experimental factors on the response and the correlation between these factors can be effectively analyzed, which can then guide the conduct of mechanism analysis or validate the results of mechanism analysis.