World Journal of Engineering Research and Technology

A.O. Ibeje*

ABSTRACT

In the study, simulation model is developed for the prediction of daily inflow into Dadin-Kowa Reservoir (River Gongola) in Northern Nigeria. In the study, the 1991-2001 records of observed and forecasted daily rainfall amounts are used as predictors and the reservoir daily inflow as predicted targets for Multilayer Perceptron Artificial Neural Networks (MLP-ANNs). With a learning rate of 0.01 and momentum coefficient of 0.85, the MLP-ANN model is developed using 1 input node, 7 hidden nodes, 1000 training epoches and 24 adjustable parameters. Error measures such as the Mean Absolute Error (MAE), the Mean Squared Relative Error (MSRE) and the Coefficient of Determination (R2) are employed to evaluate the performance of the developed model for data calibration (1991-1998), verification (1991-2001) and validation (2010-2011). The result revealed: MAE={0.7156, 0.6717, 1.046} x 10-5; MSRE = {1.4984, 1.5087, 1.1478 }x 10-7; and R = {0.9957,0.9958,0.9688}. Furthermore, dynamic model is developed based on observed and simulated daily reservoir inflow to obtain optimal allocation policy to irrigation, industrial and domestic user sectors for each month of the year. The research reveals that only the months with prolonged dry spells have optimal returns to the user sectors while the months with records of rainfall could not produce optimized returns in the model. Therefore, the application of the results will lead to saving N175, 298,126 annually in the dam provision of water to the region.

[Full Text Article] [Download Certificate]