Locally weighted projection regression for predicting hydraulic parameters

摘要

With the advent of electronics and control technology many new computing approaches are being made available for their use to the general scientific community including hydraulic engineers. One such latest technique coming under the category of data driven modelling is locally weighted projection regression (LWPR). In the present work, the suitability of this technique to solve two problems of uncertain nature, namely prediction of littoral drift and that of scour downstream of a flip bucket spillway, is assessed. Although traditional techniques to predict these hydraulic phenomena exist, they may not always yield satisfactory results owing to the complexity of underlying physical processes. Alternative approaches in this regard should therefore be welcome. For the prediction of littoral drift field measurements of waves and sediments collected near the beach of Karwar along the west coast of India are used while the scour prediction is based on laboratory model observations. It is found that the estimation of the rate of littoral drift as well as that of the spillway scour made by this new computing approach is better than traditional empirical formulae and further it rivals the results of the other and previously attempted data driven methods like artificial neural networks and genetic programming. There is thus a case to apply the LWPR technique in future to a variety of problems in hydraulic engineering.