Sparse Array Optimization Method Based on Genetic Algorithm Improved by Stacking Ensemble Learning

To address the issues of conventional genetic algorithms in sparse array optimization—such as excessive dependence on manual design of fitness functions, a tendency toward premature convergence during the search process, and difficulties in balancing multiple constraints—this paper proposes an improved genetic algorithm based on Stacking ensemble learning for sparse array optimization. First, a mathematical model of the sparse array is established. High-discrimination multi-domain features with 11 dimensions are extracted from the time domain, frequency domain, wavelet domain, and bispectrum domain, respectively, to construct the mapping relationship among array element topology, beam performance, and constraint conditions. Next, a Stacking ensemble architecture is designed, which adopts decision trees, support vector machines, and naive Bayes as base learners and logistic regression as the meta-learner, thereby realizing a data-driven fitness evaluation method. The above model is integrated into the iterative process of the genetic algorithm, forming a closed-loop optimization mechanism that combines ensemble prediction and intelligent search. Multiple sets of simulation results demonstrate that the proposed algorithm exhibits favorable noise-robust performance within the jam-to-noise ratio (JNR) range of -10 dB to 20 dB, with significant improvements in both sidelobe suppression effect and iterative convergence speed. Compared with the traditional genetic algorithm, the method presented in this paper shows outstanding advantages in optimization accuracy and optimization stability under multi-constraint conditions, and can provide an efficient and reliable technical approach for the design of radar sparse arrays.