Energy storage aggregation is an essential mean for effectively managing the participation of large-scale storage in power system dispatch. However, existing energy storage aggregation methods primarily focus on numerical aggregation of energy storage physical parameters, resulting in high computational complexity. Moreover, after aggregation, discrepancies often arise between the aggregated and actual dispatch outcomes, making these methods difficult to apply directly in practical power system dispatching decision. To address these issues, this paper proposes a large-scale energy storage aggregation method for suppressing economic dispatch deviation. The core idea is to minimize the overall operating cost difference between dispatch schemes before and after storage aggregation. By optimizing the aggregation parameters, a virtual storage model is constructed such that its participation does not interfere with the optimal dispatch of other resources. A bi-level optimization model is established to aggregate large-scale storage units into a virtual storage system. Based on the KKT conditions, the model is then transformed into a single-level model to determine the virtual storage parameters. To ensure adaptability of the virtual storage parameters in system dispatch, a multi-scenario aggregation approach is further developed to accommodate evolving system conditions. Finally, case studies on the PJM5, IEEE118, and a provincial 661-node system demonstrate the effectiveness and adaptability of the proposed method.