This study proposes a rapid assessment framework for loop systems in substations after earthquakes, in which multiple one-to-one machine learning (ML) models are established targeting a series of electrical equipment in a loop system and predicting the seismic response levels using ground motion parameters. A multi-input-multi-output assessment model for the loop system is then developed by multi-objective optimization for the performance of various ML models, utilizing swarm intelligence evolution technology to tune the model’s internal architecture. A case analysis of an 800 kV filter loop system demonstrates the efficacy of this method. In this case, the integration of gradient boosting regression tree and non-dominated sorting genetic algorithm with an elite strategy is proposed. This approach not only simplifies the number of objects by addressing coupling loop systems as units but also requires only ground motion signals, thereby avoiding sensor installations on equipment and electromagnetic interference, making it particularly suitable for assisting post-earthquake emergencies in substations.