A real-time topology identification method of distribution networks based on CNN-LSTM-Attention

Abstract

Accurate identification of the topology in a distribution network is crucial for its operation and control. Addressing the dynamic changes in the actual topology of distribution networks， an intelligent deep learning model capable of recognizing distribution network topologies was developed. Firstly， measurement data for distribution networks under different topologies were generated， followed by data preprocessing. Subsequently， an intelligent topology identification model was constructed， integrating convolutional neural network （CNN）， long short-term memory network （LSTM）， and Attention mechanism. The model was trained and tested using historical measurement data. Finally， in simulation scenarios using the IEEE 33-node and PG&E69-node distribution systems， the superiority of this CNN-LSTM-Attention-based topology identification method over traditional approaches in terms of identification accuracy was validated， and online application of the model was achieved.

Keywords

• distribution networks
• topology identification
• convolutional neural network
• long short-term memory network
• attention mechanism