In recent years, the installed capacity of photovoltaic (PV) in medium-voltage distribution networks (MDNs) and low-voltage distribution networks (LDNs) has increased rapidly, which brings about problems such as power flow direction change and overvoltage. For MDNs and LDNs with a high proportion of PV, a multi-time scale optimal dispatch method involving two stages is proposed: the day-ahead stage and the intra-day stage. Firstly, a centralized day-ahead stochastic optimization model of MDNs and LDNs is established with the objective of minimizing the network loss, the penalty cost of PV abandonment, and the charging and discharging cost of the battery energy storage system (BESS). Then, the model is transformed into a mixed-integer second-order cone programming (MISOCP) problem and solved. Secondly, in view of the different adjustment time scales of different facilities and the difficult acquisition of accurate network parameters of LDNs, a two-layer rolling optimization method for MDNs and LDNs is constructed at the intra-day stage. The upper centralized optimization model provides the benchmark strategy for the operation of MDNs, and the lower distributed control model sequentially adjusts the output of reactive power of PV inverters, reactive power and active power of BESS, and active power of PV in LDNs according to overvoltage degrees. Finally, based on the modified IEEE 33-bus MDN and 21-bus LDN, the effectiveness of the proposed method is verified.