INFILLING MAXIMA ANNUAL MONTHLY FLOWS USING FEEDFORWARD BACKPROPAGATION (BP) ARTIFICIAL NEURAL NETWORKS (ANNs)

摘要

The standard backpropagation (BP) artificial neural network (ANNs) and the pseudo Mac Laurin power series order 1 (McLlBP) and order 2 (McL2BP) derivatives techniques are used to in-fill maxima annual monthly flows. The data infilling techniques (ANNs) are firstly compared using the Root Mean Square Error of Predictions (RMSEp) as criterion. Then ANNs are briefly compared to selected regression methods (RMs) using the same criterion. South African flow gauges (i.e. the Diepkloof (control) gauge and the Molteno (target) gauge of the Orange drainage river systems are used as a case study. Generally, the study demonstrated that the ANNs techniques performed almost at the same level when maximum annual monthly flows are used. The three techniques showed a relatively substantial impact on the accuracy of the estimated missing values at the Molteno gauge for missing data proportions beyond 10 %. ANNs were shown to perform slightly better than their RMs.