Distribution Network Topology Identification Based on CNN and Improved Conformal Prediction

New power system state changes frequently, which needs identifying the possible operating state of the distribution network accurately and quickly. However, the accuracy of the topology recognition using machine learning method is difficult to reach 100%. Therefore, a method of distribution network topology identification based on CNN and improved conformal prediction is proposed in this paper. It gives a variety of possible alternative topologies and their probabilities. Based on multi-mapping relationship between the switching state of the distribution network and measured voltage data, this paper uses CNN to build a multi-output classifier. Improved conformal prediction method is used to integrate and output the possible operating topologies. The IEEE-33-node distribution network is used for verify the proposed method. The simulation experimental results indicate that this model would achieve a high accuracy of topology identification when the average number of outputs is small.