Aiming at the capacity optimization of grid-connected user side PV and energy storage system（PV - ESS）, K -means clustering algorithm and proper clustering number judgment indexes are used to select typical PV day output scenarios for planning. A mixed integer linear programming（MILP）model is presented to minimize the multi- scenario comprehensive adaptation costs due to energy storage capacity shortage or redundancy and optimize the storage capacity and operation strategy in different PV output scenarios with analyzing the impact of low valley electricity tariff and PV feed-in price. Case study shows the model can optimize the capacity configuration of ESS considering the uncertainty of PV output. The model provides fine-grained costanalysis of user side ESS based on different PV scenarios for better decision supporting.