On the importance of training methods and ensemble aggregation for runoff prediction by means of artificial neural networks

摘要: ABSTRACTArtificial neural networks (ANNs) become widely used for runoff forecasting in numerous studies. Usually classical gradient-based methods are applied ANN training and a single model is used. To improve the modelling performance, some papers ensemble aggregation approaches whilst others, novel proposed. In this study, usefulness of both concepts analysed. First, applicability large number population-based metaheuristics to tested on data collected from four catchments, namely upper Annapolis (Nova Scotia, Canada), Biala Tarnowska (Poland), Allier (France) Axe Creek (Victoria, Australia). Then, importance search compared with use very simple approach. It shown that although may slightly outperform Levenberg-Marquardt algorithm specific catchm...