Decision-making trial and evaluation laboratory (DEMATEL) is widely used because of its ability to effectively analyze nonlinear relationships between factors in complex systems. With the increasing complexity of decision-making problems, large-scale group decision-making (LSGDM) has become the norm. Most existing DEMATEL methods are only suitable for small-scale groups and simple systems. This study, therefore, proposes a large-scale group hierarchical DEMATEL method that considers consensus reaching. The DEMATEL method for LSGDM faces three challenges: large differences in knowledge structures, difficulty coordinating expert opinions, and slow group-consensus convergence. To address these challenges, first, we use hierarchical decomposition to decompose the complex system into simple systems with different levels to reduce the difficulty of decision-making in complex systems. Second, considering the limitations of expert knowledge and experience, we use the basic probability assignment function to extract the opinions of experts at different levels of subsystems and factors. Third, we divide experts into different clusters using K-means clustering to solve the problem of difficult expert-opinion coordination. Fourth, we design two types of consensuses (intrasubgroup and intersubgroup consensus) and an efficient new type of opinion autocorrection mechanism to solve the problem of the slow convergence of intragroup consensus and improve the efficiency of consensus reaching. Finally, we demonstrate the superiority of the proposed method through data analysis and method comparison.

中文翻译：

---

自动达成共识的大规模群体分层DEMATEL方法

决策试验与评估实验室（DEMATEL）因其能够有效分析复杂系统中因素之间的非线性关系而得到广泛应用。随着决策问题的日益复杂，大规模群体决策（LSGDM）已成为常态。大多数现有的 DEMATEL 方法仅适用于小规模群体和简单系统。因此，本研究提出了一种考虑达成共识的大规模群体分层DEMATEL方法。 LSGDM的DEMATEL方法面临三个挑战：知识结构差异大、专家意见难以协调、群体共识收敛缓慢。为了应对这些挑战，首先，我们使用层次分解的方法将复杂系统分解为不同层次的简单系统，以降低复杂系统中的决策难度。其次，考虑到专家知识和经验的局限性，我们使用基本的概率分配函数来提取不同级别的子系统和因素的专家的意见。第三，我们利用K-means聚类将专家分为不同的簇，以解决专家意见协调困难的问题。第四，我们设计了两种类型的共识（子群内和子群间共识）和高效的新型意见自动纠错机制，解决群内共识收敛速度慢的问题，提高共识达成效率。最后，我们通过数据分析和方法比较证明了所提出方法的优越性。