To address voltage over-limit and instability issues caused by high penetration of distributed photovoltaic in distribution  grids,  this  paper  proposes  a  voltage  regulation  strategy  based  on  distributed  photovoltaic  cluster  partition.Firstly,  a  dual-criterion  modularity  function  considering  net  load  and  equivalent  electrical  distance  is  constructed.  A community  detection  algorithm  based  on  modularity  (i.e.  fast-unfolding)  is  applied  to  dynamically  partition  photovoltaic clusters. Secondly, differentiated voltage regulation is designed according to the severity of cluster over-limit conditions:intra-cluster reactive power adjustment for mild over-limits, and multi-device hierarchical collaborative control for severe over-limits, with task allocation based on response speed and economic priority. Finally, an optimization model targeting minimization  of  network  losses  and  voltage  deviation  is  established.  The  improved  multi-organization  particle  swarm optimization (MPSO) algorithm, combined with niche-based techniques, is employed to determine the optimal regulation sequence and device action levels. Simulation results demonstrate that this method effectively controls voltage fluctuations,reduces network losses, and enhances system stability in both the modified IEEE 33-node system and the IEEE 123-node system.