Comparison of Neural Networks to Ormsbee's Method for Rain Generation - applied to Toronto, Ontario
Rainfall rate is a key input function for the analysis and design of hydrologic and hydraulic systems. One common problem with existing records of rain is that the time increments are not fine enough for use in urban storm water models. To solve this problem, observed rainfall data can be disaggregated into shorter time steps.
In this chapter two artificial neural networks are used to disaggregate hourly rainfall data into 5 min time steps. One model is a multi-layer perceptron (MLP) with a fast back propagation learning algorithm, while the other is a radial basis function (RBF) network with an orthogonal least-squared error-learning algorithm. Both models are described and evaluated.
It is shown that the RBF model performed poorly and its use is not recommended for rainfall disaggregation. However the MLP model achieved generally comparable results to Ormsbee’s continuous deterministic model, and did better in the prediction of maximum incremental rainfall depth, but at significantly higher computational effort.
This paper is only available in PDF Format:
View in PDF Format