An Experimental Design Frame for Active Dam Reserve Ratio Forecasting Using Neural Networks

Abstract

Today, one of the important and frequently spoken problems of the world is global warming and climate change. Due to these subjects, water drought and scarcity may become a trouble in the future. To prevent these problems, scientific studies are being carried out, solutions are being recommended and preventive applications are developing. In this study, to examine and foresee the decrease in water resources, active dam reserve ratio is considered and estimated using artificial neural networks. Time series analysis is performed using the active dam reserve ratio of Guzelhisar Dam, located in city of Izmir, Turkiye. Active reserve ratio data between 2012 and 2023 are considered on monthly basis. Since the data set displays high seasonality, this cyclic effect is extracted out of the data to get non-seasonal series. Then, using non-linear autoregressive artificial neural network method, both original seasonal data and non-seasonal data is forecasted. Three parameters are considered for neural network models: Input neurons, middle layer neurons and backpropagation algorithm. Results are compared according to mean absolute percent error. In the result, values of parameters to give minimum error are presented. In addition, performances of backpropagation algorithms are compared.