The invention, which belongs to the technical field of reactive power regulation in a photovoltaic power station, relates to an improved-particle-swarm-algorithm-based multi-target reactive power optimization method in a photovoltaic power station. According to the method, on the basis of establishment of a model of a large photovoltaic power station, multi-target reactive power optimization based on an improved particle swarm algorithm is carried out inside the photovoltaic power station, so that objectives of balancing of node voltages in the photovoltaic power station, improvement of stability margin, and reduction of active power loss can be achieved. The improved particle swarm algorithm is characterized in that a particle criterion is added into a standard particle swarm algorithm; a particle close to an optimal particle is optimized according to the standard particle swarm algorithm and a particle far away from the optimal particle is processed by chaotic mapping, so that particle diversity of the algorithm is enhanced and thus a global optimum objective is achieved by the algorithm. According to the invention, the voltage at a grid-connected point can be stabilized; the voltages inside the photovoltaic power station can be balanced; and the active loss inside the station can be reduced and the stability margin of the station can be improved.