Abstract: In order to efficiently predict the stress responses of interconnected electrical equipment in substations under seismic action and to achieve rapid seismic vulnerability assessment, this study proposes a deep learning-based framework for seismic vulnerability analysis of such equipment, with an in-depth investigation into the impact of bus coupling effects on the aseismic performance of different equipment types. A typical 220 kV/110 kV step-down substation was selected as the research object. Numerical simulation models of five representative multi-electrical-equipment connection configurations were established, and nonlinear dynamic time-history analyses were conducted under seismic excitation. An equivalent mapping model driven by deep learning was developed to calculate the seismic vulnerability of each equipment type under varying connection configurations. The research results demonstrate that the deep learning model proposed can rapidly and accurately predict stress responses at structural weak points of substation equipment. When accounting for bus coupling effects, the median seismic capacity of transformers increased by approximately 54.5%, whereas that of disconnectors decreased by about 75.6%, which unequivocally confirms the significant influence of bus coupling on equipment aseismic performance.