Predicting excavation performance in dense populated urban areas is critically important. Inverse analyses are powerful tools that are used to learn from local experience and predict soil response in new excavations with similar soil stratigraphy. This paper demonstrates the performance of a recently developed inverse analysis approach, SelfSim, with a special focus on its ability to provide soil models based on field measurements that can predict excavation performance in similar ground conditions for a case study in Shanghai. In Shanghai metro station excavation, the soil behavior is extracted by learning from a set of measured wall deflections and surface settlements at a selected section. The extracted soil models provide a reasonable prediction of wall deflections and surface settlements elsewhere.