Power Load Forecasting Based On Deep Sparse Autoencoder Dimension Reduction And Improved Deep Learning Model GRU

Short-term power load forecasting plays a very important role in the planning and design of distribution network.Power load forecasting can provide relevant basis and support for the internal production plan,dispatching plan and maintenance plan of power grid company.In addition,the accuracy of power load forecasting affects the safe and stable operation of power system.However,with the development of smart grid and the large number of installation of smart electricity meters,the power load data collected increases exponentially,which brings great challenges to the processing of power load data.Numerous load data influencing factors reduce the accuracy of power load prediction model.In order to solve the problem that the growth of power load data leads to the reduction of the efficiency of the model in processing data and the low prediction accuracy of the traditional load forecasting model,The paper proposed a power load prediction based on Improved Deep Sparse auto-encoder(IDSAE)dimension reduction and a Gated Recyclable Unit Networks(GRU)model.Firstly analyzes the research situation of load forecasting methods at home and abroad and the classification of power load,through periodic analysis of the power load changes the power load is obtained with daily periodicity,weeks in cyclical and periodic characteristics,through analyzing the influence factors of electric power load data,the meteorological factors is obtained,date factors,economic factors,the electricity price influence factors on the accuracy of load forecasting,therefore in the process of power load forecasting need to consider the impact of these factors.Secondly considering in the process of power load forecasting,the introduction of meteorological factors will lead to excessive load forecasting model of modeling difficulty increase,affect the running speed of the model and the prediction precision,so the paper studied the improvement depth sparse IDSAE reduced-order model,since the encoder for high-dimensional data dimension reduction,extracting feature information effectively.By comparing with Principal Component Analysis(PCA)algorithm and DSAE algorithm and classifying the processed data,it is proved that the proposed method can reduce the data dimension and simplify the data structure more effectively.With the development of intelligent and complicated power grid,the traditional power load forecasting model fails to meet the target of forecasting accuracy.Therefore,a deep learning model--Gated Recyclable Unit Networks(GRU)model was introduced in this paper and improved.Firstly,the Attention Mechanism and Convoutional Neural Network(CNN)were introduced into the improved GRU model(IGRU).By using the added attention mechanism,more attention can be paid to the key information through probability distribution of the input data features,so as to select specific inputs and reduce or even ignore some unimportant features.By introducing CNN model,the correlation between data can be fully and effectively mined without destroying load sequence,and important features can be extracted.Then Adaptive Chaos(AC)algorithm was used to optimize the Fruit Fly Optimization Algorithm(FOA)to overcome the disadvantage that the FOA model was prone to fall into extreme values in the optimization process.The improved FOA algorithm optimized the super parameters of the Bidirectional Gated Cyclological Unit Networks(BIGRU)model,and improved the prediction accuracy of the model.Finally,the effectiveness of the proposed method is verified by the analysis of simulation experiments.