Multi-frequency Combination Short-term Load Forecasting Based On EEMD

Short-term load forecasting is an integral part of energy management system and an important part of power system operation and dispatch.It has a very important significance for the operation,control and planning of power system.The accuracy of short-term load forecasting directly affects the security and economy of power system.In order to make power system dispatch reasonably,formulate reasonable generation plan,reduce generation cost,save energy and reduce emission,reduce rotating reserve,avoid major accidents,reduce operational risks,determine the installed capacity of generating units,etc.,the technical level of power load forecasting should be improved.It can be seen that the accuracy of load forecasting is one of the yardsticks to measure the modernization of power enterprise management.After many years of development,power system load forecasting has formed many forecasting methods,many of which have been very mature and began to be used in social production.Among the existing forecasting methods,different forecasting methods have their own advantages and disadvantages.Power system load is susceptible to the influence of politics,historical load value,weather and time,so it has non-linearity and multi-frequency.The existing load forecasting methods alone can not capture the multi-frequency and non-linearity of power system load.Based on this,according to the advantages and disadvantages of existing forecasting methods,this paper proposes a "decomposition-multi-model combination forecasting-integration "forecasting method,aiming at making full use of the multifrequency and non-linearity of power system load.In the existing research,empirical mode decomposition(EMD)has been used to decompose loads,but there will be modal aliasing and false components in the process of EMD decomposition.Therefore,the EMD-based ensemble false components in the process of EMD decomposition.Therefore,the EMD-based ensemble empirical mode decomposition(EEMD)is firstly used to improve the modal aliasing and false components in the process of signal decomposition,and the decomposition method is used to decompose the power load data to obtain the modal components with diffe rent cha racteristics,which are divided into high-frequency components and medium-frequency components according to their fluctuation characteristics.And low frequency components.Then,according to the fluctuation characteristics of each frequency component,the wavelet neural network method,the non-linear autoregressive model method and the support vector machine method are used to predict the high-frequency,medium-frequency and low-frequency components respectively,in order to improve the limitations of single-model forecasting method and make full use of the multi-frequency of power system load.In addition,particle swarm optimization(PSO)algorithm is used to optimize the parameters of each prediction model in order to improve the prediction accuracy of each model.After obtaining the predicted values of each frequency component,the support vector machine(SVM)method is used to integrate the training model,which takes the sample values of each component as the training input,and the actual load values as the training output.After the model training is completed,the predicted values of each component are taken as input,and the predicted values of each frequency component are integrated through the trained model to obtain the final load fo recasting value of power system.Finally,the data of Xi'an and Yulin land dispatch in one year are used as research samples,that is,after establishing the forecasting model,the number of the two areas are input into the model for power load forecasting.At the end of the prediction,the final prediction results are compared with single prediction model and EMD decomposition model through mean absolute error(MAE),root mean square error(RMSE)and mean absolute relative error(MAPE).Through comparison,it is found that the method used in this paper is more accurate than other methods.