Abstract: The randomness and fluctuation of short-term power load are strong, which makes the traditional power load forecasting method difficult to grasp the rule of short-term load variation. In order to increase the accuracy of short-term power load forecasting, the paper puts forward a new short-term power load forecasting method with the combination of Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN), Long Short-Term Memory(LSTM) network and Convolutional Neural Network(CNN). It first extracts original load series from dataset and uses CEEMDAN to decompose them into several Intrinsic Mode Functions(IMF), decreasing their non-stability.Then, LSTM network is adopted to analyze quantified time series characteristics to achieve several forecasting results. Thirdly, after superimposing these forecasting results, CNN and the fully connected layer are used to extract features and learn date features respectively to obtain the final power load forecasting results. The proposed method is compared with the benchmark model and other literature methods by practical examples. The results show that it can accurately grasp the rule of load variation, and the accuracy of the daily load prediction problems reaches 97.32%.

    Key words: power load forecasting; CEEMDAN; LSTM; CNN; Pearson correlation coefficient