Short-Term Power Load Forecasting Based on ICEEMDAN-GRA-SVDE-BiGRU and Error Correction Model

Abstract

The significance of short-term power load forecasting extends to grid dispatching and future planning. To address the issues of nonlinear characteristics and poor prediction accuracy of original power load, a hybrid short-term power load forecasting method was proposed based on Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN), Grey Relation Analysis (GRA), Improved Secondary Variation Differential Evolution Algorithm (SVDE), Bidirectional Gated Recurrent Unit (BiGRU) and Error Correction Model. Firstly, ICEEMDAN decomposition is used to divide the sequence into Intrinsic Mode Functions (IMF) and a residual component (Res), and GRA is used to reconstruct the partial component sequences to improve the model operation efficiency and anti-interference ability. Then, an Improved Secondary Variation Differential Evolution Algorithm (SVDE) is proposed to perform hyperparameter optimization on BiGRU neural networks to predict the processed component sequences. Finally, an Error Correction Model based on SVDE-BiGRU is established by the processed mode components and factors such as temperature and holiday weekends to further increase the accuracy of its load prediction. The experimental results show that the RMSE, MSE, and MAPE of the prediction method are 89.72, 60.56, and 0.55% on average, respectively. Compared with the common BiGRU prediction method its MAE value is reduced by 79.02%. Compared with several mainstream methods, its MAE value is reduced by 70.88% at maximum and 40.62% at minimum, which proves the effectiveness and accuracy of the proposed method in short-term power load forecasting.

Keywords

• Differential evolutionary algorithm
• short-term power load forecasting
• error correction
• bidirectional gated recurrent unit (BiGRU)
• ICEEMDAN
• GRA