Power load combination prediction based on EEMD decomposition with K-means clustering and frequency separation

Short-term power load forecasting helps to adjust the imbalance in the power market, while reducing physical and human losses. To investigate the performance of short-term load forecasting models, a new short-term forecasting model based on EEMD (Ensemble Empirical Mode Decomposition) decomposition of load data and K-means clustering of subseries to classify categories is proposed. The subseries of high-frequency categories are empirically learned using LSTM (Long Short-Term Memory Networks), while GRU (Gated Recurrent Unit Networks) is fitted to medium-frequency subseries, and linear regression (LR) model is fitted to low-frequency subseries.), and the final prediction results are obtained by combining the three sets of models. The empirical results show that the proposed model (HM2) has an improvement of 1.0226% and 1.2520% in MAPE and RMSE, respectively, compared with the hybrid model (HM1) with the sample entropy threshold dividing the categories. And compared with EEMD-LSTM, the proposed model has 3.0281% and 5.4323% lift, respectively. This combined model further improves the prediction of the model in terms of new way combination.