The rapid fluctuations in renewable energy output cause frequent voltage oscillations in power grids, severely threatening the safe and economical operation of power grids. To address this issue, a real-time reactive power optimization method based on graph convolutional networks （GCN） is proposed. Firstly, a multi-objective reactive power optimization model considering the importance of nodes is constructed. Based on this, the graph representation of the reactive power optimization model is established, and the adjacency matrix is restructured in conjunction with the optimization problem. Then, the optimal solution set is mapped using the GCN algorithm, and the improved CRITIC-AHP-TOPSIS combined weighting algorithm is employed to select the optimal solution, forming a real-time optimization strategy. Finally, using the modified IEEE 39-node system and an actual power grid as examples, the proposed method is verified. The results show that the method not only has the advantages of fast solving speed and avoiding local optima but also achieves more favorable voltage deviation and active power loss, ensuring the safety and economy of new power system operation.