A New Time Series Forecasting Model And Its Application In Short Term Electric Load Forecasting

Nonlinear time series forecasting model has been a hot topic in the field of science and engineering applications.Aiming at the low prediction accuracy and the poor generalization ability of single time series model,a new time series forecasting model combined complete ensemble empirical mode decomposition(CEEMD)and neural network(including extreme learning machine and kernel extreme learning machine)was proposed.Firstly,the historical data series is decomposed into a series of relatively stable subsequences self-adaptively by CEEMD.Then,by calculating the sample entropy(SE)values of each component to measure its complexity,merger and reorganize components whose entropy is similar so as to reduce the computational scale.Finally,a neural network model is established to predict the values of each component,and then the final prediction results are obtained.The CEEMD algorithm,as the improvement of EMD and EEMD,not only overcomes the mode mixing problem of EMD,but also inherits the advantage of minimal reconstruction error.The components of historical sequences after Multi-scale decomposing and reorganizing by CEEMD,are predictive modeling respectively by the BP neural network,the Elman neural network,the extreme learning machine(ELM),the Gauss kernel extreme learning machine(KELM)and the wavelet kernel extreme learning machine(WKELM).ELM as a new type of single hidden layer feedforward neural network has the advantages of fast training speed and strong generalization ability.Two kinds of nuclear extreme learning machine is the improvement of ELM,and the kernel function concept is introduced,avoiding setting the number of hidden layer node number,and avoiding the system instability caused by the random setting of the weights at the same time,this paper applies it to the field of time series forecasting.Through the combination of CEEMD and several neural networks,it can effectively improve the accuracy of the time series forecasting model.The power system management and scheduling has a very high requirement for the accuracy of the load forecasting model.This paper forecasts the daily load curve by the proposed combined forecasting model and the five single forecasting models.The CEEMD-SE-ELM model which is more efficient was chosen to predict the weekly load curve of different regions.The numerical example shows that the new model has certain advantages in the forecast accuracy and efficiency,and has practical application potential.