Abstract: The "double-high and double-peak" characteristics of power loads are becoming increasingly prominent with the advancement of new power system, necessitating reliable and accurate load forecasting for power system operation planning. To predict the power load with better accuracy, a short-term power load combination prediction model based on MIC-VNCMD-TCN-TPA-LSTM is proposed. The maximal information coefficient (MIC) theory is utilized to analyze the nonlinear coupling of load and meteorological information and identify the crucial information. Variational nonlinear chirp mode decomposition (VNCMD) is introduced to process the nonlinear non-stationary load data and decompose them into corresponding components. On this basis, a combined TCN-TPA-LSTM prediction model is constructed, and the corresponding prediction model is selected according to the prediction evaluation index of each element. Subsequently, the overall prediction results are reorganized. The actual electric load data from certain place is used as the dataset for comparison experiments, which demonstrates superior accuracy and generalization capability compared to other prediction models, thus verifying the effectiveness and superiority of the proposed method.