With the support of an extensive dataset, the spatial distribution and temporal evolution of urban rail transit infrastructure vibration acceleration in the long term is analyzed. The dataset is collected by an infrastructure monitoring system deployed on a metro line, with the coverage of 1 year. An energy spectrum density (ESD) based method of damage -sensitive feature extraction, dedicated to the examination of the dy namic response caused by passing trains is designed. Meanwhile, cumulative sum (CUSUM) and haar filtering are applied to the measurement of long-term variation of feature value. The results show that the vibration monitoring system fits the demand of structural health monitoring well. When operational infrastructure is at a steady health state, its dynamic response excited by passing trains is to some extent randomly distributed whereas its mean value is stable. It shows that the proposed analysis methods can clearly characterize the well-operating infrastructure during a long service time. In addition, the calculated feature value and its CUSUM & the haar filtering value can be referred to, when the assessing of other track structures' health state is demanded in the future.