Methods to improve neural network performance in daily flows prediction

摘要: In this paper, three data-preprocessing techniques, moving average (MA), singular spectrum analysis (SSA), and wavelet multi-resolution (WMRA), were coupled with artificial neural network (ANN) to improve the estimate of daily flows. Six models, including original ANN model without data preprocessing, set up evaluated. Five new models ANN-MA, ANN-SSA1, ANN-SSA2, ANN-WMRA1, ANN-WMRA2. The ANN-MA was derived from raw combined MA. ANN-WMRA1 ANN-WMRA2 generated by using SSA WMRA in terms two different means. Two flow series watersheds China (Lushui Daning) used six for prediction horizons (i.e., 1-, 2-, 3-day-ahead forecast). poor performance on forecast mainly due existence lagged prediction. among performed best eradicated lag effect. performances ANN-SSA1 ANN-SSA2 similar, also similar. However, based presented better than at all horizons, which meant that is more effective improving current study. Based an overall consideration complexity modeling, optimal, then SSA, finally WMRA.