The rapid penetration of distributed energy intensifies the problem of voltage fluctuation in distribution network, and the traditional static prediction model and compensation strategy are difficult to cope with the dynamic topology change and rapid voltage fluctuation. Therefore, this paper proposes a solution that combines graph neural network (GNN) and dynamic compensation algorithm. In the aspect of node voltage prediction, a weighted graph is constructed to describe the electrical connection between nodes in the distribution network, and the voltage-reactive sensitivity matrix is embedded into the GNN edge weight initialization as a priori knowledge. The GNN-LSTM joint architecture is used to capture the temporal and spatial evolution law and realize multi-step voltage prediction. In the aspect of dynamic compensation, the dynamic threshold adjustment mechanism based on prediction error distribution is designed, combined with multi-objective optimization model (minimizing voltage deviation and control cost), and solved by improved particle swarm optimization algorithm. The experiment is based on the improved IEEE 33-node distribution network. The results show that the average absolute error (MAE) of the proposed forecasting model in different scenarios is significantly lower than that of LSTM and static GCN. The total time of voltage exceeding the limit is greatly reduced by the dynamic compensation strategy, and the total amount of reactive power compensation, OLTC operation times and voltage volatility are better than the traditional methods, which effectively improves the stability and economy of the distribution network.
Zongxi Xie (Sun,) studied this question.