Implicit stochastic optimization for deriving operating rules for a multi-purpose multi-reservoir system

In this study, optimal operating policies for a multi-purpose multi-reservoir system have been derived using Implicit Stochastic Optimization (ISO) method. In order to do so, four models: (1) all possible regression, (2) stepwise regression, (3) decomposition and (4) simulation, have been proposed that deals with the statistical significance as well as the physical significance of the presence of independent variables in a particular regression model. The first two models deal with the variable selection based on statistical significance. In the decomposition models, operation problem is divided into sub-problems to generate smaller subsets for regression retaining variables of physical significance. The simulation model examines the performance and the predictive efficiency of different regression models. Fifty six years historical inflow data (1962–2017) have been used for this analyses. These four models along with 1000 years monthly synthetic inflow sequences, generated by the multivariate gamma AR(1) model, are applied to the Damodar Valley (DV), a multipurpose multi-reservoir system in India. The effect of variability in those inflow values on the release decisions are taken into account implicitly and the results are presented. Analysis of result indicates that decomposition model with less number of predictor variables is the most preferred model for DV system.