Research On Short-Term Electricity Load Forecasting Method Based On Feature Information Extraction With Bi-directional GRU

Electric load forecasting is the foundation for ensuring the balance of power supply and demand,and provides information and basis for the planning and construction of power grids and power sources,as well as the operational decisions of power grid enterprises and users.In recent years,the rapid development of new power systems dominated by new energy has led to clean and low-carbon power sources becoming the main source of electricity supply,while the controllability of power sources has decreased.At the same time,the widespread access of nonlinear dynamic load equipment has made load data characteristics more complex,posing severe challenges to load forecasting.In order to improve the accuracy and efficiency of electric load forecasting,this article proposes a short-term electric load forecasting method based on feature information extraction,using deep learning prediction methods,such as bidirectional gated recurrent neural networks,encoder-decoder,attention mechanisms,and ensemble empirical mode decomposition algorithms.The effectiveness of the proposed method is validated using electric load datasets from a province in northwestern China and Singapore.The main research contents of the paper are as follows:(1)Aiming at the multi-dimensional characteristics of power load data,a Bi GRU short-term power load prediction model(AE-Bi GRU)integrating AE autoencoder is proposed.The deep features of the load data are extracted by the self-encoder,and the data are represented by dimensionality reduction to accelerate the model training,and the training and prediction of Singapore’s electricity load data are performed with the help of bidirectional gating cycle units.The experimental results verify the effectiveness of the AE-Bi GRU model,and the model further improves the accuracy of electricity load prediction.(2)To address the non-linear characteristics of electricity load data,this paper proposes an EEMD-Bi GRU short-term electricity load forecasting model based on ensemble empirical mode decomposition.The model decomposes data using EEMD and selects strongly correlated feature components using the Pearson correlation coefficient method,reducing prediction errors caused by load data sequence fluctuations and instability.Bi GRU is used to train and predict the electricity load data from a province in northwestern China,and experimental results validate the effectiveness of the EEMD-Bi GRU model.(3)To further improve the accuracy of load forecasting,this paper proposes an EEMD-EDA-Bi GRU short-term electricity load forecasting model that combines AE autoencoders and ensemble empirical mode decomposition algorithms.The load data is decomposed using EEMD,and the influential feature factors are obtained using AE autoencoders.The output of both is then combined into a new training matrix input for Bi GRU.Finally,the attention mechanism is used to obtain the correlation weights,which improves the accuracy of the prediction.The predictive performance of the model is validated using load datasets from a province in northwestern China and Singapore.Experimental results show that the multi-advantage fusion EEMD-EDABi GRU model can effectively improve the accuracy of load forecasting in load forecasting tasks.