Determining Effective Features for Estimating the Volume of Water Delivered to the Irrigation and Drainage Network using Artificial Intelligence Methods (Case study: Irrigation and Drainage Network of Sefidrood Dam)

In this study, the ability of water cycle algorithm in combination with ANN to model the volume of water delivered to the Irrigation and drainage network and determine the effective characteristics, was investigated. For the input of artificial intelligence models, five Features include the volume of water delivered to Sefidrood Irrigation and drainage network for one lags, flow and inlet volume for seven lags, level and reservoir volume for ten lags, were chosen. Using a hybrid algorithm, feature selection method and sensitivity analysis, a triple combination of features (with MSE of 0.00045) is the best input combination. The volume of water delivered to Sefidrood Irrigation and drainage network for one lags according to sensitivity analysis has been the most effective feature in its modeling. Then the artificial neural network weights were optimized to increase efficiency and with the help of two meta-innovative algorithms: water cycle without evaporation (WCA) and water cycle with evaporation (ER.WCA). The ANN-ER.WCA hybrid model with the highest accuracy (with values of R, NRMSE, MAE and NS equal to 0.9915, 0.0975, 0.0090 and 0.9829 in the test period, respectively) is in the first priority and the ANN-WCA and ANN models are next in line, respectively.