The high proportion of intermittent distributed photovoltaic access brings challenges to the operation of an AC-DC hybrid distribution network. The current two-stage stochastic optimization dispatch model leads to a large deviation between real-time and day-ahead decision-making. This paper presents a multi-stage stochastic optimization model based on a scenario tree. The proposed model aims at minimizing the sum of day-ahead purchase cost, day-ahead adjustment cost and real-time balance cost. The model also realizes photovoltaic absorption and peak load filling through flexible adjustment measures such as energy storage device adjustment, converter station output power adjustment and demand response. The discrete decision variables of the multi-stage stochastic optimization model can be adjusted adaptively with the change of uncertain information in the intra-day and real-time stage. This can meet the operational requirements of an AC-DC hybrid distribution network with a high proportion of PV penetration and provide technical support for its economical and flexible operation. This work is supported by the National Natural Science Foundation of China (No. U1966205).