In the context of a national strategy of ‘emission peak and carbon neutrality’, the penetration of distributed generators (DG) is increasing dramatically, bringing significant social and environmental benefits. However, DG features such as high intermittency and uncertainty lead to problems such as high peak-valley difference and frequent voltage violation, causing negative impacts on the safe and stable operation of distribution networks. As one of the most effective solutions, the coordination of a battery energy storage system (BESS) and DG has been attracting a lot of attention. However, existing studies of BESS and DG coordination unreasonably assume distribution networks are three-phase balanced, and are less capable to deal with uncertainties because of the limitations of BESS characteristics. Thus, a BESS and DG-based two-stage control strategy of unbalanced distribution networks is proposed. Specifically, based on a three-phase network model and predictions of DG and load power, stage I performs BESS control on a long-time scale (24 hours) to improve the operational economy. To deal with the voltage violation risk in stage I, stage II conducts distributionally robust voltage optimization of the unbalanced distribution network over a short time scale (15 min), based on source-load uncertainty modeling and the reactive capability of a DG inverter, thus improving the operational safety. Convex optimization and dual reconstruction are used to solve the above two-stage optimization problem, and simulations are carried out on a real distribution network. The results show that the proposed two-stage optimization strategy can effectively take into account the impacts of three-phase unbalance and source-load uncertainty, and also improve the economy and safety of distribution network operation. This work is supported by the General Program of National Natural Science Foundation of China (No. 61873159).