代理模型对预测荷兰港口区泥沙浓度的应用

he use of both process-based model and data driven model (DDM) in simulating sediment processes have shown to be useful by previous research. However, both approaches have disadvantages. This paper explored several data driven methods to build simple models in predicting SPM based on output from the process-based models. Artificial neural network (ANN) is adopted as surrogate model to predict suspended particulate matter (SPM) concentration in the Southern North Sea. Surrogate model is essentially a simple and fast ‘model of the model’. The simulation by surrogate models is acceptable and simulation time reduces dramatically. Surrogate models are also built with linear regression method which refers to ‘parsimonious model’. Parsimonious model is the simplest feasible model with the fewest possible number of variables It requires less computation time, the simulation is transparent and results are easy to interpret.