Prediction and optimization of runoff via ANFIS and GA

In planning of water resource projects, the estimation of the availability of water plays an important role. The first step in the water availability estimation is the computation of runoff resulting from the precipitation on river catchments. The length of the runoff measured in a stream may be of short period or long period depending upon the catchment characteristics. Keeping this in mind the present work is focused on two different model generation. In the first phase of this study, runoff rating curves are developed considering present day water level (H(t)) as input and present day runoff (Q(t)) as the model output. In the second phase of the study runoff prediction models are developed considering 1 day lag water level (H(t − 1)), 2 day lag water level (H(t − 2)) and 1 day lag runoff (Q(t − 1)) as inputs and 1 day ahead runoff (Q(t + 1)) as the output of the model. Models developed and used for prediction of runoff are Non-Linear Multiple Regression (NLMR) and Adaptive Neuro-Fuzzy Inference System (ANFIS). Both the models were trained and tested to predict the performance of models. Genetic Algorithm (GA) is then coupled with NLMR model to obtain the condition of hydrological parameter for which the runoff is maximum.