Research On Short-term Power Load Forecasting Method Based On Sequence Decomposition And Optimized Kernel Extreme Learning Machine

As one of the key energy sources for the development of national economy,electric power system undertakes the important task of producing,transmitting,distributing and using electric power,which constitutes a very complicated system.High precision short-term load forecasting is very important to ensure the reliability of power system.However,the development of various industries,especially the growth of energy-intensive industries,further promoted the rapid expansion of electricity demand.This not only brings great pressure to power system,but also puts forward higher requirement for short term power load forecasting.It is difficult for traditional forecasting methods and a single intelligent forecasting model to achieve accurate load prediction.Therefore,this thesis builds a shortterm power load combination forecasting model based on swarm intelligent optimization algorithm,machine learning and data preprocessing.The main research contents of this thesis are as follows:(1)In view of the low accuracy of a single prediction model and the complexity of various influencing factors,this thesis proposes a short-term load prediction model based on the improved sparrow search algorithm(ISSA)to optimize the kernel Extreme Learning Machine(KELM).Firstly,to solve the problem that the original sparrow search algorithm(SSA)has insufficient optimization effect,nonlinear decreasing weight and adaptive firefly disturbance strategy are added to control the balance between global search and local search,and the ISSA performance is tested by reference function.Then,ISSA was used to optimize KELM’s kernel parameters and regularization coefficient to improve its model stability.Pearson correlation coefficient(PCC)was used to screen the best feature set to reduce the influence of low correlation redundancy features on model accuracy,so as to improve the overall prediction performance of the model.(2)Aiming at the fluctuation and nonlinearity of power load data,a combined prediction model based on sequence decomposition and optimized kernel extreme learning machine is proposed in this thesis.The variational mode decomposition(VMD)is used to carry out adaptive decomposition of power load data,and the optimized KELM model is selected to predict each sequence after decomposition one by one,and then the predicted value is added to get the final result.In view of the fact that VMD is affected by its parameters,this thesis proposes three decomposition methods of NGO-VMD,ISA-VMD and IVMD to carry out adaptive decomposition of power load data,effectively removing the impact of noise in the data,improving the fitting ability of the forecasting model to the power load data,and further improving the overall forecasting performance of the model.(3)In this study,we compared the combined prediction model proposed in this thesis with other classical machine learning and deep learning prediction models through simulation experiments,and verified that the prediction method proposed in this thesis has better prediction performance.It provides a new and effective solution for short-term power load forecasting and helps to improve the operation efficiency and reliability of power system.