为准确评估钢管混凝土拱桥地震损伤，将拱桥系统视为串-并联体系，即将拱桥系统各关键体系进行串联，而不同关键体系的单个构件之间则分别采用串联或并联形式。以某钢管混凝土拱桥为研究对象，通过时程分析和神经网络预测获得拱桥各构件的地震响应值，基于Copula函数分别得到拱桥各关键体系易损性和整体系统易损性，并与基于一阶界限法得到的系统易损性进行对比。研究结果表明：通过神经网络预测可以获得准确的拱桥结构地震响应值，当地震动峰值加速度A _PG为0.4g时，预测准确率超过90%，并且随着A _PG的增大，准确率逐渐提高；拱桥各关键构件体系中，拱上立柱体系失效概率最高，在抗震设计时应对其采取减隔震措施，风撑体系的失效概率最低，进行拱桥系统易损性分析时可忽略风撑体系的影响；基于串-并联体系的拱桥系统易损性介于一阶界限法的上界和下界之间，当A _PG=0.3g时，轻微、中等和严重损伤状态下串-并联体系的系统失效概率分别为98%、94%及25%，与一阶界限法上界的相对偏差分别为-1.4%、-3.1%及-16%，与下界的相对偏差分别为0.6%、3%及11%，钢管混凝土拱桥采用串-并联体系进行系统易损性分析明显更合理。

In order to accurately evaluate the seismic damage of concrete-filled steel tubular （CFST） arch bridge， this paper regards the arch bridge system as a series-parallel system， that is， the key systems of the arch bridge system are connected in series， and the single components of different key systems are connected in series or in parallel. Taking a CFST arch bridge as the research object， the seismic response values of each component of the arch bridge are obtained by time-history analysis and neural network prediction. Based on the Copula function， the vulnerability of each key system and the overall system of the arch bridge are obtained respectively， and compared with the system vulnerability based on the first-order boundary method. The results show that the accurate seismic response value of arch bridge structure can be obtained by neural network prediction. When the peak ground acceleration A _PG=0.4g， the prediction accuracy rate exceeds 90%， and with the increase of A _PG， the accuracy rate gradually increases. Among the key component systems of the arch bridge， the failure probability of the arch column system is the highest， and the seismic isolation measures should be taken in the seismic design. The failure probability of the wind bracing system is the lowest， and the influence of the wind bracing system can be ignored in the vulnerability analysis of the arch bridge system. The vulnerability of the arch bridge system based on the series-parallel system is between the upper and lower bounds of the first-order boundary method. When the A _PG=0.3g， the failure probabilities of the series-parallel system under mild， moderate and severe damage conditions are 98%， 94% and 25%， respectively. The relative deviations from the upper bounds of the first-order boundary method are -1.4%， -3.1% and -16%， respectively， and the relative deviations from the lower bounds are 0.6%， 3% and 11%， respectively. It is obviously more reasonable to use the series-parallel system to analyze the vulnerability of the CFST arch bridge.