With the increasing penetration of wind power and gradual retirement of conventional generating units (CGUs), wind turbines (WTs) become promising resources to provide steady-state energy reserve (ER) and frequency support for the grid to facilitate supply-demand balance and frequency security. In this regard, a novel frequency constrained dispatch framework with ER and virtual inertia from WTs is proposed. Firstly, this paper establishes the WT model with both ER and virtual inertia, whose energy sources are WT's deloading and rotor kinetic energy, respectively. Secondly, the system frequency response and CGUs' power response are derived while considering WTs exiting inertia response at frequency nadir. Then, this paper develops a stochastic-optimization-based frequency constrained dispatch model, where both WTs' frequency regulation parameters and rotor speeds are decision variables, so that the coupling between WT's mechanical and electrical parts and the coupling between system's transient dynamics and steady-state operation can be fully reflected. Finally, convex hull relaxation, convex hull approximation and deep neural networks are used to transform the original nonlinear model into a mixed-integer second-order cone programming model. Case studies on the 118-bus system verify the effectiveness of the proposed models and methods.