Development of Hybrid Adaptive Neuro Fuzzy Inference System - Harris Hawks Optimizer (ANFIS-HHO) for Monthly Inlet Flow to Dam Reservoirs Prediction

Abstract

Nowadays, machine learning models are able to make good predictions based on pattern extraction between data. In this study, a neural-fuzzy network (ANFIS) was used to predict the inflow to the reservoirs of a dam namely, the Mahabad dam located in the northwestern part of Iran. A new Harris Hawk (HHO) optimization algorithm was also used to improve the ANFIS (HHO-ANFIS) structure. Monthly precipitation and temperature and inlet flow data to the reservoir one to three months ago were used as input parameters as 6 different input patterns. About 70% of the data was used for training and 30% to test the models. The results showed that the ANFIS model has good accuracy in training data although, for test data, its accuracy was greatly reduced. The development of the HHO-ANFIS model improved the accuracy of the prediction. The patterns with all input parameters had the highest prediction accuracy. In this pattern, values ​​of Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Nash Sutcliffe Efficiency coefficient (NSE) for test data were 3.9 MCM, 2.41 MCM, and 0.86, respectively. Due to the good performance of the model used, it can be recommended for time series predictions.

Keywords

• evolutionary algorithms
• rainfall-runoff
• predicting time series
• mahabad dam
• machine learning