Abstract: With the integration of a large number of distributed generators（DGs） into the network, the uncertainty and randomness of the output of the supply and load in the distribution network gradually increase. The application of flexible multi-state switches（FMSs） enable the interconnection between supply areas without restriction of voltage levels and the phases. Therefore, the joint operation optimization of multi-supply areas in active distribution network（ADN） will become a new normal situation, which provides the foundation for the improvement of the load unbalance condition and the voltage level. By using flexible interconnection characteristics of the FMS, taking the weighted minimum of the supply area and feeder load balance indices as the objective function, the bi-layer embedded load balancing model for ADN is established, and a hybrid optimization algorithm consisting of improved particle swarm optimization（PSO） and second-order cone programming is used to solve the problem of network reconfiguration topology and FMS output. First, the closed-loop distribution network is simplified and equivalent to decrease the particle scale during the network reconfiguration of PSO algorithm in the outer layer of the whole method to speed up the calculation. In the inner layer, the objective function is linearized using optimal equidistant piecewise linear approximation algorithm（OEPLAA）, by which the problem of output optimization of FMS can be transformed into a second-order cone programming problem. Finally, the proposed double-layer load balancing method is analyzed and verified through an actual distribution network.