•  
  •  
 

Keywords

distributed generation, generalized load, self-organizing map neural network, grey wolf optimizationalgorithm, temporal convolutional network

Abstract

To address the complication of load characteristics caused by the extensive integration of distributed generation into distribution networks, a high-precision generalized load modeling method is proposed. Firstly, the self-organizing map (SOM) neural network is utilized to perform feature extraction and dimensionality reduction on grid node load data, and nodes with similar dynamic characteristics are divided into typical subnetwork systems through unsupervised clustering. Secondly, the grey wolf optimization (GWO) algorithm is adopted to perform global optimization on the hyperparameters of the temporal convolutional network (TCN), establishing a generalized load model for the subnetwork systems. Finally, simulation experiments based on the IEEE 33-node distribution system show that the proposed method outperforms comparison methods in clustering metrics (DBI and silhouette coefficient) and modeling metrics (MAE, RRMSE, MAPE, and R2). Specifically, DBI is reduced by an average of 28.2%; the silhouette coefficient is increased by an average of 56.2%; MAE is reduced by an average of 32.6%; RRMSE is reduced by an average of 37.1%; MAPE is reduced by an average of 33.1%, and R2 is increased by an average of 3.1%. The generalized load modeling method based on the SOM-GWO-TCN integrated algorithm can effectively reduce model complexity and improve the accuracy of the established model.

DOI

10.19781/j.issn.1673-9140.2026.03.012

First Page

120

Last Page

130

Download

Share

COinS