Application of Genetic Algorithms and K-Nearest Neighbor Method in Developing Reservoir Operation Policies During Floods

In Iran, the reservoir operation during floods is mostly carried out based on engineering judgments. There is no operation policy for the majority of the reservoirs especially those with no flood prediction and warning system. In this research, operation policies during flood are developed for Shahid Abbaspour Reservoir on Karun river southwestern Iran. A Genetic Algorithm (GA) model is developed to determine the reservoir releases. The parametric objective function of the GA model is to minimize the flood losses associated with releases exceeding the safe carrying capacity of the downstream river. The GA model performance is evaluated using historical flood data as well as simulated floods with 50-year return period. The GA model output has been used along with a K-Nearest Neighbor (K-NN) model to develop the reservoir operation policies for controlling floods. Comparison between the results of K-NN and linear regression model shows the accuracy of the GA and K-NN models to develop the operation policies for reservoir operation during floods.