Abstract: Aiming at the characteristics of large load fluctuations and complex energy coupling in the user-level integrated energy system, we proposed a combined load forecasting method for integrated energy systems based on wide ＆ deep and Residual Network（ResNet） framework, which adopted Complete Ensemble Empirical Mode Decomposition with Adaptive Noise（CEEMDAN） and Principal Component Analysis（PCA）. The proposed model consists of two parts: width and depth. The data in the depth part were decomposed using CEEMDAN before input, and principal component analysis was used to extract and sort the main influencing factors of the decomposition results. The depth part of the model referred to the idea of ResNet, stacked multiple LSTM sub-layers to build a depth prediction network, and realized the cascade processing of data with different information densities; the width part of the model adopted a simple model and improved the input of the Wide part of the traditional Wide＆Deep-LSTM model, which effectively reduced the training difficulty of the model. The analysis of practical examples shows that the proposed model has good prediction accuracy and convergence speed. Compared with conventional models, the proposed model has certain advantages.