Research On Power Load Forecasting Based On Improved Gated Cyclic Neural Network

Load forecasting is an important guarantee for the design of power system,planning and operation.In order to further improve the accuracy of power load forecasting and facilitate power system management and planning personnel to make reasonable,safe and accurate plans for power energy dispatching.Accurate load forecasting can make the output and consumption of electric energy reach a certain balance point,reduce the waste of electric energy,indirectly reduce the resource cost of electric energy,and play an important role in building a sustainable information society.As a classic forecasting problem,power load forecasting has been paid attention by researchers.In recent years,with the technological progress of computer hardware and software,as well as deep learning is used in various fields of life,it has brought a variety of changes to the society.In the field of load forecasting,there are a series of technical methods that can adapt to modern power data.Based on the analysis of the characteristics of power load data,this paper first analyzes the characteristics of power data,the outliers generated in power load data and the steps of data pretreatment for power load data.Then the classical power load data model is analyzed to study the existing optimization of the model.Finally,based on the gated cyclic neural network,a more accurate Attentional GRU model and a combined prediction model of XGBoost and gated cyclic neural network are proposed after learning experiments.This paper proposes and verifies the double-layer GRU load forecasting model,the GRU model based on attention mechanism.Subsequently,combined with the real power load forecasting value of a certain area in China,according to the experimental results of data set verification,it can be concluded that the gating cycle neural network model with attention mechanism has the best forecasting effect.Compared with LSTM network and Gru neural network,the prediction accuracy of attention GRU network is improved by 18% and 15% respectively.A combined prediction model based on Gated Recurring Neural Network（GRU）and Limit Gradient Boost（XGBoost）is proposed.Firstly,according to the load data and the input structure of GRU and XGBoost models,the data are preprocessed,respectively,the preprocessed data are input into the corresponding model,and then the weighted combination of the model is carried out through the optimal weighting method,and the predicted value of the final combined model is obtained.Finally,the combined prediction model is compared with the common prediction model through the analysis of a numerical example.The results show that the proposed method can effectively combine the advantages of the two models,and simultaneously take into account the continuous time series and the discontinuous characteristic variables,and has higher accuracy than the single model and the common prediction model.