Peak shaving is one of the most important applications of battery energy storage system（BESS）. In order to fully make use of the battery as well as prolong the battery life， in the load shift optimization， many factors should be considered such as battery capacity， times of charge and discharge， load curve and the coordination with other applications of battery energy storage system. A real-time optimization strategy based on the dynamic programming is presented. Constraints like times of charge-discharge and depth of discharge are added to the optimization model. Dynamic programming is employed to solve discontinuous constraints problem when discrete BESS power adjustment is considered. BESS may also respond in other control modes or functions to the power system disturbances or fluctuations， which may result in unpredictable capacity variations. The real-time optimization strategy can be used to correct the battery capacity online. The strategy mentioned above is successfully applied to a MW-level BESS demonstration project in Biling substation， China Southern Power Grid. The results of the simulations and field tests based on the actual load data of Biling substation demonstrate the effectiveness of the optimization strategy.