Abstract: Grid-connected distributed photovoltaics alter the power flow characteristics of distribution networks. Fluctuations in photovoltaic output lead to voltage limit violations and rising network losses. To enhance the operational performance of power grids, this paper conducts research on reactive power optimization. A multi-objective optimization model that takes both network loss and voltage deviation into account is constructed. An improved particle swarm optimization (PSO) algorithm integrated with adaptive inertia weights and dynamic learning factors is proposed. Simulative comparisons under multiple operating conditions are carried out on the IEEE 33-node test system. Compared with the standard particle swarm optimization algorithm, the improved algorithm can escape local optima, cut active power losses significantly, and keep the voltages of the entire network steadily within the allowable range. Conclusion: The proposed algorithm features superior convergence accuracy, which can offer a feasible solution for on-site reactive power regulation of distribution networks with high photovoltaic penetration.