Short-term Power Load Forecasting Based On Intelligent Optimization LSSVM

Accurate short-term power load forecasting can effectively ensure the safe and stable operation of power systems and improve power quality.Traditional forecasting methods can hardly reflect the influence of nonlinear regression factors on the forecasting results,and it has become a trend to apply intelligent optimization algorithms to the field of short-term power load forecasting.To address the problems of non-smooth sequences in power load data and the selection of prediction model parameters,this paper proposes an intelligent optimization-based least squares support vector machine(LSSVM)prediction model for short-term power load forecasting research.The main research contents of this paper are as follows:(1)Firstly,the load data are pre-processed to correct the abnormal and missing values in the data set,and then the degree of influence of relevant meteorological factors on the load change is analyzed by Pearson analysis,and the meteorological data(maximum temperature,minimum temperature and average temperature)with greater correlation with the load data are selected as the influencing factors for short-term power load forecasting under the condition of ensuring the accuracy of load forecasting.(2)LSSVM model based on statistical theory is proposed to input load data and influence factor data into the model for short-term electricity load forecasting.Different support vector machine(SVM)and LSSVM models are built by selecting different kernel functions,and they are trained,validated and tested by the same data set to finally compare and obtain a better performance forecasting model.(3)Considering the non-stationary and non-periodic components in the power load data,in order to reduce the influence of these factors on the short-term power load forecasting results,this paper proposes to decompose the load data using ensemble empirical modal decomposition(EEMD),which effectively improves the modal confounding phenomenon that can occur in the empirical modal decomposition(EMD)and decomposes the original load data into multiple IMF components,which improves the accuracy of load prediction to a certain extent.Later,the different IMF components are classified and synthesized by gray relation analysis(GRA),which effectively reduces the amount of model training,improves the efficiency of load prediction,and ensures the prediction accuracy.(4)Considering that the parameter selection of the LSSVM model is subjective and empirical,and the different choices of parameters will affect the prediction effect of the LSSVM model,the first proposal is to use the traditional particle swarm optimization algorithm(PSO)to select its parameters for the optimization search,and although the PSO-LSSVM model improves the load prediction accuracy,its effect has not yet reached the expectation,so it is proposed to use the sparrow search algorithm(SSA)is proposed to further improve the optimization-seeking effect.Finally,the comparative analysis shows that the model optimized by SSA has higher prediction accuracy.