Multilayer heterogeneous network-enabled hierarchical consensus optimization for group decision-making in complex product research and development

Complex product research and development (R&D) involves multiple domains with heterogeneous information, making consensus among decision makers (DMs) difficult to achieve. Moreover, most studies overlook the inherent organizational hierarchy of DMs in practical R&D, severing the connection between DMs’ social influence and their evaluations, leading to significant semantic information loss during preference fusion. Therefore, this paper presents a hierarchical consensus-optimized group decision-making (GDM) framework based on multilayer heterogeneous networks. First, we construct a multilayer heterogeneous network model, encompassing requirement, research, production, and test layers. The trust relationships and heterogeneous rating information are mapped to low-dimensional vectors using graph embedding, enabling the quantitative integration of heterogeneous information. To overcome direct interaction sparsity, a weighted trust propagation mechanism is developed to capture indirect relationships, and node embedding vectors are learned using the skip-gram model to analyze the importance of the DM nodes. Then, to address decision consensus conflicts between intragroup and intergroup levels, consensus measures are proposed at the attribute, scheme, intragroup, and intergroup levels. On this basis, to balance individual adjustment costs with collective consensus efficiency, a bilevel programming consensus model is designed for the consensus-reaching process (CRP), and the adjustment strategies are optimized using the double deep Q-network (DDQN) algorithm, improving optimization stability by reducing value overestimation. Collective consensus is enhanced through the collaboration of the upper and lower deep Q-networks (DQNs). Finally, an aero-engine R&D case study demonstrates that the proposed method significantly outperforms traditional methods in CRP, exhibiting good engineering applicability in large-scale group decision-making scenarios.