Performance Evaluation of Multilayer Perceptron Neural Network and Adaptive Neuro-Fuzzy Inference Systems for Reservoir Operation Optimization: a Case Study of Cheffia Reservoir, Algeria

Abstract

Artificial Intelligence based prediction has wide applications, including hydrology, water resources management and particularly reservoir operation. Thus, two black box models based on Artificial Neural Networks and Fuzzy Logic methods are implemented and tested in forecasting reservoir operation; the first is a Multilayer Perceptron Neural Network and the second is an Adaptive Neuro-Fuzzy Inference System combining the two methods. The developed models consist of predicting evaporation, inflows and reservoir storage from their historical records and that, with aim of providing the best fit between predicted and observed values and of improving operating rules on storage and releases. The performance of achieved results demonstrated the pertinence of Artificial Neural Networks and fuzzy Logic methods in predicting cyclical state variables, such as evaporation and storage, while the prediction of inflows to reservoir generally gave better results compared to other research works available in open literature on one hand. On other hand, it is deduced from testing different prediction models that these methods are unable to predict random variables.