Artificial Neural Networks Model for Multireservoir Water Allocation

Abstract

This research aimed to develop the multireservoir water allocation model for the Upper Mun basin in Northeast of Thailand where there were 4 large scale multipurpose reservoirs namely Lam Chae, Mun Bon, Lam Phra Ploeng and Lam Takong located. The study area covered 7,190 sq.km. The water in the basin was used for 4 purposes namely agriculture, municipal and industrial water supply, preserving the ecological system equilibrium and the water requirements at the river basin outlet for downstream users. There were 4 big irrigation systems covering the area of 353,650 rai and 8 municipal and industrial water supply systems. 25 years of monthly data were used in the analysis of the water shortage in the basin. The optimum water allocation among the 4 water user groups was derived by the multiobjective optimization techniques called Epsilon-constraint linear programming. The best water allocation policy that satisfied the profitability, equity and reliability criteria was derived by the multicriteria decision making process called AHP. The priority weights for profitability, reliability and equity were 41, 32.3 and 26.7% respectively. The best water allocation policy was then simulated by HEC-3 using the cases of dry and normal year. The simulation result was used to develop the artificial neural networks model for multireservoir water allocation. The 12-16-4-4 MLFF-BP neural networks were the best fitted model. The model showed very high R² of 0.9578-0.9900 for training and R² of 0.8041-0.9658 for the testing. The 12-16-4-4 MLFF-BP neural networks model was tested by comparison with the actual data in 2001 and 2002. This research illustrated that the artificial neural networks was simple and could practically
be applied to the multireservoir water allocation.