Multiobjective evolutionary optimization for green supply chain network architecture design

Abstract

This study addresses the issues of multi-dimensional objective mutual exclusion and weak topological response in extreme scenarios within green supply chain network architectures. An improved multi-objective evolutionary algorithm incorporating a zero-return neural network tracking mechanism is proposed. A mixed-integer nonlinear programming model integrating operating costs, lifecycle carbon footprint measurement, and topological resilience is constructed. Experimental data validation shows that the proposed algorithm architecture improves the multi-objective comprehensive hypervolume index (HV) by 21.7% in an A100 accelerated cluster environment, and the inverted generation distance (IGD) converges to 0.0112. In blocking frequency domain response and time-varying load experiments, it effectively constrains the transmission deviation rate caused by link failure impacts, and the node connectivity survival rate under extreme scenarios increases to 61.4%. This establishes a computational and measurement mechanism based on multi-dimensional Pareto equilibrium for global low-carbon disaster-resistant physical network planning under uncertain environments.