Abstract: To reduce the influence of complicated environmental factors on the results of the ultra-short-term electrical load forecasting and improve the prediction precision and operation efficiency of the algorithms, this paper proposes a CNN-LSTM hybrid prediction algorithm based on the cluster empirical mode decomposition (CEMD). Firstly, this empirical mode decomposition (EMD) is used to decompose the load data into several intrinsic mode functions (IMF) and residual (Res) with the excellent smoothness and regularity. Secondly, for the sake of simplify the calculation volume of the subsequent model, the k-means clustering method is used to group and integrate the decomposed components. In the mean time, by analyzing the prediction effect of different clustering numbers, the optimal clustering tag is selected to construct the input data of the neural network. After that, the data of each group are input into the CNN-LSTM hybrid neural network, where the CNN mines the features between the data to form the feature vectors which are input into LSTM for prediction. Finally, all the forecasting results are added linearly to get the complete forecast load. Compared with the existing models, the proposed method in this paper has higher prediction accuracy through the real load test.