Heavy Load and Overload Pre-warning for Distribution Transformer with PV Access Based on Graph Neural Network

The scale of distributed PV access exceeds the hosting capacity of the distribution network, which will cause heavy load and overload of the distribution transformer. To solve the problem of inaccurate load prediction caused by insufficient dataset and multiple influencing factors when using the existing models for the distribution transformer overload pre-warning with PV access, this paper proposes a load rate prediction method based on Graph Sample and Aggregate (GraphSAGE) and Long Short Term Memory (LSTM). The method uses Euclidean distance and empirical knowledge to analyze the correlation of the time series data of each distribution transformer, and then constructs the data association graph of the distribution transformer; Then input the dataset and association graph into the GraphSAGE-LSTM network for offline training, and finally input the real-time operation data of the distribution transformers into the model to predict the load rate of a certain distribution transformer and make a heavy load and overload pre-warning. Combined with the IEEE33-bus distribution network and operation data, the case shows that the method can achieve high prediction accuracy in the high-dimensional and small dataset with PV access.