Theoretical line loss calculation method based on feature clustering and residual fully connected network for low-voltage station area containing PV

In view of the inaccurate calculation of theoretical line loss in low-voltage areas containing PV due to the change of tidal direction caused by the access of distributed photovoltaic (PV) users in low-voltage (LV) areas and the complexity of urban low-voltage distribution network lines, the theoretical line loss calculation method based on Feature Aggregation (FA) and Residual Fully Connected Network (RFCNN) for low voltage stations containing PV is proposed. First, a feature clustering model based on hierarchical clustering is constructed to reorganize the similar features to achieve feature extraction and dimensionality reduction of the input data. Then, the best feature clustering number is selected according to the model performance advantages and disadvantages of different clustering feature numbers. Finally, the clustering features are fed into the residual fully connected neural network regression model for training to obtain the daily line loss value calculation model. Experimental results show that the proposed model has better generalization ability and accuracy compared with models such as multilayer perceptron (MLP), extreme gradient enhancement (XGBoost), shallow and deep residual fully connected networks.