There are a large number of lowvoltage distributed smallcapacity photovoltaic power supply in the distribution network. Due to its large scale and limited realtime communication, its economy and operability of online adjustment can not be guaranteed. It is difficult to effectively use these reactive resources for auxiliary adjustment of the distribution network. Under the background, this paper proposes a lowvoltage photovoltaic participating in the reactive power operation mode optimization strategy of mediumvoltage grids based on cluster mode presets and online autonomous operation to make full use of the potential of lowvoltage distributed PVs (LVDPVs). Firstly, LVDPVs linked with the same transformer are assigned to a cluster and managed by a terminal platform. Secondly, the dayahead reactive power mode optimization is carried out aiming at compensating load reactive power demand within a feeder. Then, five cluster reactive power operation modes are designed and the dayahead prediction operation scenario is constructed. In addition, a cluster reactive power operation mode and parameter optimization model is established considering the optimal feeder reactive power balance and the minimum reactive power flow between clusters based on the scene characteristics. The improved particle swarm optimization algorithm is employed to solve the problem. Finally, examples verify the effectiveness of the proposed method.