Research On Short-term Load Forecasting Method Based On IMA-GRU

With the expansion of China’s power grid,the increasing demand for industrial electricity,and the integration of a large amount of new energy generation into the grid,there are higher requirements for the operation of the power system.The factors that need to be considered in load forecasting are increasing day by day.Load forecasting is an important work in the operation process of the power industry,and accurate forecasting is particularly important for the stability and economic operation of the power system.Reasonable forecasting can provide the correct decision-making basis for the power grid dispatch and operation of the power system.In order to improve the accuracy of load forecasting under multiple factors,this thesis constructs a combination model based on IMA-GRU.The main research content is as follows:(1)In the actual process of using electricity,various influencing factors cause fluctuations and instability in the load data sequence.To address this issue,firstly,this thesis decomposes the load data into multiple components with certain frequency domain characteristics through data decomposition,in order to weaken the influencing factors;Secondly,by comparing wavelet decomposition,ensemble empirical mode decomposition(EEMD),and variational mode decomposition(VMD),it was found that VMD decomposition has the best performance,but the selection of two parameters has a significant impact on the decomposition results.This thesis uses the flower pollination algorithm(FPA)to optimize it and improve the VMD decomposition effect;Finally,FPA-VMD decomposed the data,and the decomposition results showed that FPA-VMD had the best decomposition effect compared to the other three methods.(2)During the model prediction process,excessive input or insignificant correlation of feature variables results in a decrease in the accuracy and efficiency of the model prediction.In response to this issue,this thesis adopts the maximum correlation minimum redundancy(m RMR)method for feature selection,selects features for each component after FPA-VMD decomposition,analyzes the correlation between each component and feature information(daily type,temperature and humidity,real-time electricity price,etc.),and finds a set of feature sets that have the highest correlation with the output results of the components and the lowest correlation between features,laying the foundation for future model prediction work.(3)In the process of GRU prediction,improper parameter selection leads to a problem of reduced model prediction accuracy.To address this issue,firstly,this thesis uses Monkey Algorithm(MA)to optimize the parameters in GRU;Secondly,in the actual optimization process,MA has the problem of falling into local optima.Therefore,the Monkey Algorithm(MA)is improved from four aspects: initialization,climbing process,viewing process,and jumping process;Finally,a test function was used to verify the optimization performance of the Improved Monkey Algorithm(IMA).Simulation results showed that the convergence speed and solution accuracy of the improved IMA were superior to Particle Swarm Optimization,Artificial Bee Colony,and standard Monkey Algorithm.(4)The combination model constructed in this thesis is used to analyze the actual load data of a certain region in Australia.The prediction results show that m RMR improves the model’s prediction efficiency to a certain extent.FPA-VMD data decomposition,m RMR feature selection,and IMA optimization of GRU parameters can all improve the model’s prediction accuracy,verifying the effectiveness of the model in this thesis.