Day-Ahead Prediction of Harmonic Voltage Based on TCN-TimeGAN Hybrid Neural Network

In a distribution network with a high proportion of renewable energy, the access points of renewable energy power generation equipment are numerous and diverse, which makes it difficult to predict the harmonic content in advance, especially on a relatively large time scale such as the short term and the medium to long term. To solve short-term prediction difficulties of harmonic voltages in distribution networks, a day-ahead harmonic voltage prediction method based on the TCN-TimeGAN hybrid network is proposed in this work. The framework employs a two-stage approach: First, temporal convolutional networks (TCNs) process historical power quality data from critical nodes using dilated causal convolutions and residual blocks to predict individual harmonic components while maintaining computational efficiency. Then, the TimeGAN network was applied to enhances total harmonic distortion voltage (THDV) accuracy through adversarial training and reversible embedding mappings, addressing error accumulation in conventional calculation processes. Finally, the effectiveness of this method was verified by using historical monitoring datasets of an actual distribution network.