Single-reservoir operating rules for a year using multiobjective genetic algorithm

A monthly operating rule for single-reservoir operation is developed in this study. Synthetic inflow data over 100 years are generated by using a time series model, AR(1), and piecewise-linear operating rules consisting of 4 and 5 linear lines are found using the implicit stochastic optimization method. In order to consider multiobjective functions in reservoir system operation, a multiobjective genetic algorithm (NSGA-II) is adopted to obtain the optimization results. The search space of NSGA-II is carefully refined using frequency analysis of historical data, and the relationship between inflow and constraints is also investigated. It is determined that 4 and 5 segments are the optimal number of segments for the piecewise-linear operating rule, and the effect of random number seeding on NSGA-II is evaluated. Six years of historical inflow data are used for the simulation model and the results show that the developed operating rule would handle various inflow series. As a result, probabilistic reservoir storage forecasts can be provided to a system operator so as to enable the operator to evaluate the current status of a reservoir quantitatively.