Integrated energy stations (IESs) are widely spread in power-electronized active distribution network (ADN) for multi-energy services, and heterogeneous uncertainties from source-loads exacerbate voltage violation risks with increasing penetration of distributed generators (DGs). Existing research has not explicitly revealed the spatio-temporal synergy adaptability of multiple response-otherness resources. This paper proposes a bi-level mixed-scale strategy to leverage multiple resources for voltage improvement under uncertainties. Upper level performs model predictive control (MPC)-based rolling for optimizing large-scale reserve-integrated coordination schemes of soft open points (SOPs) and thermal storage/demand; and reserve margins for building temperature are constrained for the sake of unpredictable environmental factors. In lower level, mixed-scale adjustment is applied in which ADN executes re-generated small-scale schemes and IESs obey large-scale schemes of upper level. SOPs’ rapid response abilities are invoked to tackle real-time errors while building temperatures vary within comfort range for responding to insensitive environmental fluctuations. Via sufficient flexibility assurance and mixed-scale coordination in separated levels, the voltage level is effectively facilitated. The original nonconvex model is relaxed into a second-order cone programming (SOCP) formulation, which can be effectively tackled for timely voltage regulation. Numerical results show the proposed strategy can realize spatio-temporal regulation of multiple resources for adapting to different-scale uncertainties, mitigating voltage deviation significantly.