Abstract：Short-term load forecasting in power systems is challenging due to its strong volatility and high

randomness. A short-term power load prediction method based on multi-model fusion was proposed，specifically

the WTCN-BiGRU-Attention-CatBoost approach. The original load data was decomposed using variational mode

decomposition（VMD）to extract its intrinsic modal components，effectively mitigating data non-stationarity. The

decomposed data was then fed into the WTCN-BiGRU-Attention neural network，the network hyperparameters

were optimized using an improved whale optimization algorithm，followed by parallel integration with the

CatBoost regression model. Subsequently，the MAPE-RW fusion algorithm was introduced to search for fusion

weights，establishing the optimal fusion strategy and generating the final output. Simulation results show that the

proposed fusion model achieves MAPE and RMSE values of 0.625 5% and 72.654 MW，respectively，both

significantly lower than those of single models，indicating its superior short-term generalization capability and

interference resistance.

         Key words：short-term load forecasting；variational mode decomposition（VMD）method；wavelet temporal

convolutional network（WTCN）deep learning architecture；whale optimization algorithm；CatBoost algorithm

     Format Citation:李子乾，张才俊，王殊,等. 基于WTCN⁃BiGRU的CatBoost短期负荷预测方法[J].电气传动,2026,56(10):62-68. LI Ziqian，ZHANG Caijun，WANG Shu,et al. Short-term electricity load forecasting method based on multi-model fusion WTCN-BiGRU-Attention-CatBoost [J].Electric Drive, 2026,56(10):62-68