On-line nonlinear modelling and forecasting of streamflow using neural network

• Autores: Mohd Yusoff Mashor
• Localización: International journal of the computer, the internet and management, ISSN 0858-7027, Vol. 17, Nº. 1 (ENE-ABR), 2009, págs. 78-86
• Idioma: inglés
• Texto completo no disponible (Saber más ...)
• Resumen

  • This study explores the capability of neural networks for on-line nonlinear modelling of streamflow. Three neural networks were considered, namely multilayered perceptron network (MLP), radial basis function network (RBF) and hybrid multilayered perceptron network (HMLP). The comparative results showed that, HMLP and RBF networks have much better performance than MLP network. RBF network produced better forecasting results than HMLP network for training data set, however, HMLP network produced slightly better results than RBF network for testing data set. Due to simpler structure and better generalisation performance, HMLP network was selected for on-line nonlinear modelling of streamflow. The on-line forecasting results indicating that the HMLP network starts to give good forecasting performance of up to 3-hours lead-time after it has been trained only with 150 data samples. After 300 training samples, the HMLP network is ready to provide good forecasting up to 24- hours lead-time. These results suggest that the HMLP network is suitable for on-line nonlinear modelling and forecasting of streamflow.