Multi-Objective Antenna Design Based on $l_{1}$-Optimization BPNN Surrogate Model

In order to solve the problem of large computational cost and long time consumption of traditional multi-objective antenna design method, this paper proposes a surrogate model based on sparsely connected back propagation neural network (SC-BPNN). In this model, l₁ optimization is used to automatically determine the network structure and connection parameters. The l₁ formulation of modeling can force some weights of the mapped neural network to zero while leaving other weights as non-zero. We utilize this feature to allow l₁ optimization to automatically determine which mapping is necessary and which mapping is unnecessary. The proposed method is applied to the fast multi-objective optimization design of the multi-parameter antenna structure. The design results show that the proposed surrogate model of SC-BPNN can be integrated into multi-objective evolutionary algorithm for processing complex antenna designs with high dimensional parameter space.