The Forecasting system has two steps which are outliers detection using neural network and the deteced outliers replace using interpolation. Therefore the effects of outliers are removed, the one-step ahead prediction neural network is constructed by using the filtered series. The two step prediction neural network is also trained by using the previous predicted values. Experimental results indicated that the algorithm show much better than the neural network method. And in conclusion, we can get more accurated forecasting data by using the proposed method.