This paper presents a novel gradient-based Recurrent Neural Network (RNN) for solving fuzzy quadratic optimization problems (FQOPs) through a direct primal-solving approach–a key departure from traditional indirect dual-solving methods. The proposed framework first reformulates FQOPs into bi-objective and weighted problems, then derives the Karush-Kuhn-Tucker (KKT) optimality conditions to construct the RNN model. Crucially, our direct approach offers superior computational efficiency, simpler implementation, and enhanced numerical stability compared to dual-formulation alternatives. We rigorously prove the Lyapunov stability and global convergence of the proposed RNN. Comprehensive numerical experiments, including performance comparisons against dual-solving methods (demonstrating faster convergence and reduced CPU time), validate the advantages of our approach. Finally, a real-world case study further illustrates the method’s practical effectiveness.

Keywords