Abstract: Stimulated by China’s dual carbon strategy, the new energy vehicle industry has undergone rapid development, leading to the large-scale expansion of supporting charging infrastructure. During this accelerated construction phase, safety incidents resulting from the superposition of equipment failures and human error have become increasingly prevalent. However, existing safety resilience assessment systems exhibit significant limitations: many rely excessively on subjective experience, others determine weights based solely on empirical data, and most fail to account for the inherent fuzziness and randomness of the assessment process. Consequently, these models are inadequate for dynamic safety management across the infrastructure lifecycle. Establishing a robust safety resilience assessment methodology is therefore imperative. This study defines the safety resilience of charging stations as a comprehensive operational capability that guarantees the long-term stable functioning of charging service systems amid multiple disturbances, including electrical system faults, extreme meteorological conditions, and human operational errors. Based on this definition, an analytical framework integrating four stages and three dimensions was constructed. The four stages resistance, response, recovery, and adaptation encompass the complete risk lifecycle from prevention to optimization, while the three dimensions physical, social, and information correspond to hardware facility protection, personnel management collaboration, and technical data support, respectively. Through the synthesis of national standards, integration of industry development reports, and expert validation, 26 core indicators were identified. To resolve the imbalance between subjective and objective weights in traditional weighting methods, this study employed the order relation analysis (G1) method to streamline the conventional analytic hierarchy process. This was integrated with the criteria importance through inter-criteria correlation (CRITIC) method, which determines objective weights by calculating the coefficients of variation and correlation; a preference coefficient was introduced to balance these factors. To quantify assessment uncertainties, a cloud model was applied: the reverse cloud generator transformed standardized indicator data into three-dimensional digital features—expectation, entropy, and hyper-entropy—and the forward cloud generator established standard cloud models for five assessment levels—excellent, good, moderate, poor, and very poor. Cosine similarity was utilized to assess the proximity between the comprehensive assessment cloud and the standard cloud model of each level, with the maximum membership principle adopted to determine the safety resilience level. Empirical analysis of nine representative charging stations in Jiaozuo indicated that their safety resilience followed a stepwise distribution: Longyuan lake station is excellent, Railway station is good, seven stations are moderate. Subsequent obstacle analysis revealed that physical protection, fire-extinguishing facilities, and emergency response efficiency were critical issues across all stations. Stations covering diverse scenario types demonstrate differentiated safety and operational weaknesses. On-site disposal and evacuation efficiency serve as the unique shortfalls of station A adjacent to the Longyuan lake urban public space and station E situated in congested old urban roads. Station B, located in the university new district, shows weakened performance in facility maintenance and response timeliness, represented by safety monitoring coverage. Station C in a high-mobility public scenario and small-scale station F are constrained by battery risk early warning accuracy. Both station D and station H share the limiting indicator C18 determined by inherent scenario attributes, while station D requires additional attention to indicator C14. The open-air energy supply layout of station I forms the intuitive manifestation of its deficiency in lightning protection and grounding integrity. The innovations of this study include the construction of the four-stage, three-dimensional analytical framework, the integration of the G1–CRITIC–cloud model to rectify weighting imbalances, and the precise identification of obstacles and scenario-specific defects through empirical research to provide a technical foundation for improvement strategies.