A virtual battery (VB) provides a succinct interface for aggregating distributed storage-like resources (SLR) to interact with a utility-level system. To overcome the drawbacks of existing VB models

including conservatism and neglecting network constraints

this paper optimizes the power and energy parameters of VB to enlarge its flexibility region. An optimal VB is identified by a robust optimization problem with decision-dependent uncertainty. An algorithm based on the Benders decomposition is developed to solve this problem. The proposed method yields the largest VB satisfying constraints of both network and SLRs. Case studies verify the superiority of the optimal VB in terms of security guarantee and less conservatism.