|
|
Publicações
Campos, R. e Guedes Soares, C. (2016), “An hybrid model to forecast significant wave heights”, Maritime Technology and Engineering 3, Guedes Soares, C. & Santos T. A., (Eds.), 2016, Taylor & Francis Group, London, UK, pp. 1027-1035
This paper presents and evaluates hybrid models using statistical tools to reduce the bias of forecasts or hindcasts of significant wave heights. The “hybrid model” consists of two models working together: (1) the traditional numerical wave model (in this case, WAM) and (2) the statistical model. The numerical model predicts the wave heights while the target of the statistical model is to predict bias (difference of measurement minus model); both predictions are combined to provide a final accurate estimation at the position 24.39S / 44.04W in southern Brazilian coast. Two multivariate statistical models are applied and compared, one of linear regression (degrees 1 to 4) and another of neural network (2 to 64 neurons at the intermediate layer) using backpropagation batch learning. The model using linear regression proved to be efficient in reducing bias, but not the scatter index (SI) or the root mean square error (RMSE). The neural network model presented the best results, especially with neurons from 16 to 32. The final bias was reduced from 0.13 to 0.06 meters and SI from 0.12 to 0.03.
Se nao conseguiu obter uma copia desta publicação: Peça uma cópia desta publicação
Para informação sobre todas as publicações do CENTEC pode descarregar: Lista Completa de Publicações do CENTEC
|