To address the limitations of traditional power supply chain security risk early warning methods constrained by insufficient risk identification accuracy, this paper proposes a self-adaptive convergence factor matrix-iterative extreme learning machine (ELM) model for enhanced security risk prediction. Building upon conventional ELM architecture, the developed algorithm incorporates matrix iteration with adaptive convergence factors to compute weight matrices. It intelligently selects optimal convergence factors through structural analysis of iterative equations, achieving efficient and precise solutions for linear systems while guaranteeing equation convergence. Comparative experiments with benchmark algorithms validate the method’s effectiveness in security risk early warning. Experimental results demonstrate that the proposed matrix-iterative ELM model with adaptive convergence factors achieves breakthrough improvements, showing significant advantages in critical metrics including higher risk prediction accuracy and faster computational efficiency compared to traditional approaches.