Author(s): A. S. Farias Camilo; Suzuki Koichi; Kadota Akihiro; B. Celeste Alcigeimes
Linked Author(s):
Keywords: Sustainability; Artificial intelligence; Monte Carlo optimization
Abstract: A risk assessment of reservoir operating rules defined by Monte Carlo Optimization (MCO) and Artificial Neural Networks (ANNs) is examined. The MCO technique generates synthetic inflow scenarios which are used by an optimization model to find optimal reservoir releases. The ensemble of optimal release data is conditioned on storage and inflow in order to define operating policies. Unlike the common use of regression equations relating releases to the other variables, this study makes use of ANNs to find the releases to be implemented at each period. The procedure was applied to operate the Ishitegawa Dam Reservoir that supplies water to the city of Matsuyama, Japan. The operating rules set up by the ANNs were used to simulate a 10-year monthly operation. The results are shown to be equivalent to those obtained by optimization under perfect forecast and superior to the ones found by the socalled Standard Linear Operation Policy (SLOP). The outcomes from the operation are used for determining sustainability criteria and the drought risk index for different levels of demand. Although SLOP provides higher reliability and resiliency for greater levels of demand, its vulnerability is much more elevated than those produced by optimization under perfect forecast and by the MCO-ANN-generated rules, making the system less sustainable. In general, the sustainability and drought risk indexes obtained by the latter approaches are better than those of the SLOP. However, the optimization under perfect forecast needs accurate inflow forecasts for the whole horizon, which is not always the case. In this way, MCO-ANN-generated policies, which require information only on the initial reservoir storage and current inflow, may be useful in the decision-making process of reservoir operation.
Year: 2007