Abstract: The real-time data structure of each device in the new energy cluster is complex, and the interaction between historical data and different variables is easily overlooked, resulting in low warning accuracy. Therefore, a regional new energy cluster protection warning method based on temporal convolutional networks (TCN) and transfer learning is studied. Based on cluster scenarios, construct a temporal convolutional network model to extract temporal features. Using transfer learning strategies to learn temporal features related to the protection and early warning of regional new energy clusters, transfer the pre trained TCN model parameters of similar regions to the current model. Based on this temporal feature, combined with rule algorithms, evaluate the cluster status and classify the regional new energy cluster. The experimental results show that this method has high accuracy in power warning, can capture fluctuation trends, and has significantly better AUC values in its ROC curve, which has practical application value.