Short Term Load Forecasting Method Based on Time Series Decomposition

Electricity provides a basic guarantee for social production and life, which puts high demands on the safe and stable operation of the power system. To improve the accuracy of short term power load forecasting, this paper proposes a shortterm load forecasting method based on CEEMDAN and BiGRUMLR models. Firstly, the Spearman rank correlation coefficient is used to quantify the impact of the characteristic sequence of; Secondly, the influencing factors on power load; factors and power load data are subjected to CEEMDAN, and each component is reconstructed into high-frequency and low frequency components by introducing zero crossing rates; Then, BiGRU is used for HF prediction, and MLR model is used for LF prediction; Finally, the predicted values of each model are superimposed to obtain the final result. The experimental results show that the evaluation indexes MAPE and RMSE of the load forecasting model combined with CEEMDAN algorithm and BiGRU-MLR model reach 0.631 and 101.7, respectively. Compared with the comparison model, MAPE is reduced by 1.403, 1.120, 0.840, 0.826 and 0.584, respectively. RMSE decreased by 129.8, 111.5, 77.5, 66.2 and 34.0, respectively, which has strong convergence performance and can effectively improve the prediction accuracy.