Research On Short-term Power Load Forecasting Based On Deep Learning Technology

Short-term power load forecasting plays a key role in the production planning and actual operation scheduling of the power system,which always been a hot research issue in the academic field.Due to the randomness of the power system,the historical load data is a random non-stationary time series,and there are many challenges in load forecasting.This thesis studies deep learning methods,exploits the advantages of deep learning,and conducts research on short-term power load forecasting.The main research contents include:Aiming at the problem of low prediction accuracy of existing short-term load prediction algorithms,an attention mechanism and proposes an Attention-CNNBi GRU short-term load forecasting model based Convolutional Neural Network(CNN)and Bidirectional Gated Recurrent Unit(Bi GRU)is developed in this thesis.Firstly,feature extraction is performed on the input vector using CNN and Bi GRU.Secondly,the feature extraction results are input into the Bi GRU layer containing the attention mechanism,and the influence of key features is enhanced through the attention mechanism so that it can be assigned to a higher weight,improving the accuracy of load prediction.Experimental results show that the proposed Attention-CNN-Bi GRU model can effectively improve the accuracy of short-term load prediction.The Res Bi GRU short-term load forecasting model based on the improved Res Net network to further improve the accuracy of load forecasting is proposed in this thesis.The proposed model is optimized on the Res Net network structure,including designing the residual block structure of Bi GRU and a large number of jumper connection structures.Through the design of the Bi GRU network in the residual block,the prediction model can better learn the temporal correlation of features.At the same time,a large number of jumper connection structures increase the network density which further improves the expressive ability and back-propagation efficiency.The experimental results show that the proposed Res Bi GRU model effectively improves the accuracy of short-term load forecasting.Variational Mode Decomposition(VMD)method is introduced in this thesis to improves the Res Bi GRU model,and a VMD-Res Bi GRU short-term load forecasting model is proposed.The development of VMD method can fully exploit the advantages of deeper information of historical loads,and improve the ability of prediction model feature extraction and learning.The experimental results show that compared with some existing models,the performance and prediction accuracy is further proposed for the proposed VMD-Res Bi GRU model.