Research On Short-term Power Load Forecasting Based On Hybrid Intelligent Optimization Algorithm

Accurate short-term power load prediction is helpful to ensure power system safety and reduce power generation costs.Related scholars and experts have studied power load forecasting for a long time and have achieved promising results.However,with the development of the economy,the natural climate changes has more and more non-linear factors that affect the power load,and most of the traditional prediction models rely heavily on artificial experience and lack the ability to learn independently,so it is difficult to analyze and fit highly nonlinear multi-factor power load data better;Artificial neural network has the advantages of adaptive learning ability,strong nonlinear mapping ability and high operating efficiency.It has been widely expanded in short-term power load forecasting,but for the parameters of neural network model are randomly initialized,it is still difficult to accurately predict the power load.The improved Tree Seed Optimization Algorithm combines the advantages of the Tree Seed Algorithm and the Levy Flight Algorithm,which has excellent global optimization capabilities,and is widely used in parameter optimization.In this thesis,the improved tree seed optimization algorithm is used for artificial neural network parameter optimization.Actual data has carried out related research on short-term power load forecasting.The main contents include:(1)An improved Tree Seed Optimization algorithm based on Levy flight(LTSA)is proposed.The standard Tree Seed Optimization(TSA)algorithm has weak global search ability,and the algorithm is easy to premature.This thesis proposes to combine the standard TSA optimization algorithm with the levy flight,so as to strengthen the global search ability of the algorithm,and take into account the convergence speed of the algorithm,to avoid the algorithm premature.Meanwhile,the LTSA algorithm was compared and tested with TSA,MFO,GWO and PSO intelligent optimization algorithm on 10 benchmark functions.(2)A long short-term memory network model with attention mechanism(AM-LSTM)is proposed to apply to short-term power load forecasting.Compared with the original LSTM model and other commonly used time series prediction models,the overall prediction effect of AM-LSTM is better than that of LSTM,in the early stage of prediction,the accuracy is slightly worse than models such as ELM,but in the later stage of prediction,the prediction performance is better than other models.(3)An extreme learning machine prediction model based on improved Tree Seeds Optimization algorithm(LTSA-ELM)was established,which was compared with the extreme learning machine based on other intelligent optimization algorithms,and verified on 6 UCI data sets to prove the validity and reliability of the LTSA-ELM prediction model.At the same time,it is proposed to use KPCA extraction method based on Gaussian kernel function to perform dimensionality reduction preprocessing on short-term power load data,and use the processed short-term power load as input to the LTSA-ELM model to simplify the network structure and increase the training rate of the network.Compared with other ELM application models improved by other intelligent optimization algorithms(TSA,MFO,GWO,and PSO)and other basic time-series forecasting models(BP,SVM,LSTM,and AM-LSTM),the results show that the KPCA-LTSA-ELM model is more suitable for short-term power load forecasting.