Based on Singular Spectrum Analysis(SSA)method the standardized sample series have been reconstructed using the quasi-period signal weight.A BP neural network multi-step prediction model with the Mean Generating Function(MGF)prolongation matrix of the reconstruction series as the input factor,and the original sample series as the output factor has been formulated.Results show that the model is superior in predictions over the other three models,indicating that denoising by SSA and constructing BP neural network prediction model are relatively effective for improving forecast accuracy.Therefore,the model is an useful one for the actual operational forecasting.