Voltage optimization is a key management issue in active distribution systems that can be tackled by acting on the reactive powers provided by distributed generators. Due to the large number of nodes, distributed algorithms are preferable to solve the optimization problem with limited measurements and information exchange. Starting from a centralized formulation of the voltage optimization problem and adopting a network partitioning in voltage control zones, the problem is reformulated in a decentralized form. Then, the Auxiliary Problem Principle is applied to obtain a distributed iterative algorithm that solves the decentralized problem with limited data exchange. The effectiveness of the distributed algorithm is tested on a 24-nodes LV distribution grid with six DGs.