Abstract: Due to the features of relatively greater randomness and volatility of multiple loads in the user-level integrated energy system, the existing prediction methods cannot get good prediction effects. To this end, a multiple load prediction model based on kernel principal component analysis（KPCA）, quadratic modal decomposition, deep bidirectional long short-term memory（DBiLSTM） neural network and multiple linear regression（MLR） is proposed. First, the complete ensemble empirical mode decomposition with noise adaptability is used to perform eigenmode decomposition of the electric, cooling, and heating loads, and the obtained strong non-steady components after decomposition are decomposed again by using variational modal decomposition.Then, KPCA is used to extract principal components from feature sets of weather and calendar rules to achieve data dimension reduction. The decomposed non-steady and steady components, combined with the principal components of the feature set are respectively predicted by DBiLSTM neural network and MLR. Finally, the prediction results are reconstructed to obtain the final prediction results. Through the analysis of actual calculation examples and compared with other models, the proposed model has higher prediction accuracy.