Research On Short-term Power Load Forecasting Based On Deep Learnin

Power load forecasting is an important part of power system operation management.Accurate load forecasting helps power supply companies arrange and coordinate the power generation of power plants and formulate corresponding power outage maintenance plans,so as to ensure the safe and economical operation of the power grid,provide users with reliable power quality,and effectively alleviate the sudden power outages.Economic losses provide the necessary conditions for the safe and stable operation of society.However,with the increasing complexity of the power system,the changes in the power load are also more and more complex.Accurate load forecasting brings great challenges to researchers in related fields.In order to improve the accuracy and stability of load forecasting,based on the research status at home and abroad,this paper proposes a combined forecasting model based on the improved GRU model and the quadratic decomposition algorithm based on the deep learning model and decomposition algorithm.In order to improve the data quality of the load series,this paper firstly carries out the processing steps of missing values and outliers for the load series.Then,based on the deep learning model,an improved deep learning model is proposed to predict the model.Since the input at different time steps has different correlation and importance to the output.Therefore,in order to suppress and gate the input adaptively,by introducing the input suppression and gating module into the GRU model,the improved GRU model IGRU can adaptively suppress the input elements that are not related to the output,and the gating is related to the output.the input element.In order to fully exploit the potential characteristics of the load sequence,this paper proposes a quadratic decomposition algorithm to decompose the load sequence,and combines the quadratic decomposition algorithm with the improved GRU model to predict the load.The specific steps are as follows: First,the CEEMDAN decomposition algorithm is used to decompose the original load sequence into several subsequences with different frequency characteristics.Then,the load sequence is divided into high and low frequency components by using the sample entropy of each subsequence.Next,the high-frequency component group is recombined into a high-frequency subsequence by addition,and the high-frequency subsequence is decomposed twice by the VMD decomposition algorithm.Finally,the IGRU model is used to model the high-frequency subsequences obtained by the secondary decomposition of VMD and the low-frequency subsequences obtained by CEEMDAN decomposition,respectively.Add the prediction results of each subsequence to get the final prediction result of the load.Taking the load data sets of two regions in the United States as an example,the effectiveness of the prediction method proposed in this paper is verified through experimental research.