Storage-Yield Evaluation and Operation of Mula Reservoir, India

Abstract

A set of nested models was applied to provide useful strategy to evaluate the storage, water yield and the operational performance of the multipurpose Mula reservoir in India. Insufficient yield from the reservoir for the purpose of water supply and irrigation has led to the need for reevaluation. These nested models were applied in tandem using linear programming (LP), dynamic programming (DP), artificial neural networks (ANN), hedging rules (HRs), and simulation. An LP-based yield model (YM) has been used to reevaluate the annual yields available from the reservoir for water supply and irrigation. The yields obtained from the YM have been refined by two DP models, viz; the controlled output DP (CODP) and the controlled inventory DP (CIDP). The prespecified annual release reliabilities and the yield deficits were similar to that used in the YM. In this approach, the ANN models use a hybrid model in the stochastic generation of monthly inflows to the reservoir for studying its operational performance. With the ANN, both the sigmoidal (DPN) and the radial basis function network (RBN) evaluated reservoir performance using results of the DP model. Continuous HR (CHR) and discrete HR (DHR) were also used for modeling the reservoir performance. The results show that the YM estimates the annual multiyields accurately. The data generated by the hybrid models preserve the characteristics of the historical data. The RBN is better than the DPN models. In the CHR, the sets of hedging trigger obtained from the YM gave an acceptable performance. The DHR reduces the number of reservoir empty conditions and hazards of droughts. The study reveals that the project may fail to fulfill the irrigation demands.