Predicting reservoir volume reduction using artificial neural network

Abstract

One of the modern methods for predicting the behavior of natural phenomena is using artificial neural network (ANN) model that predicts new data through applying experimental data and finding out the rules behind the data. Area-reduction experimental model is a method for distributing sediments in the reservoirs. The main goal of the present study is to evaluate the feasibility and utilization of ANNs in predicting reservoir volume reduction under low data conditions. The present paper focuses on applying the ANN model with a multi-layer perceptron (MLP) in MATLAB software. Prediction of the volume reduction of ANN model is done by using a 10-year period of data (2003–2013) at Baroon dam reservoir in Maku, Iran as a case study. Furthermore, for modeling the distribution of sediments in dam reservoir, area-reduction experimental model is used for a 30-year period (2009–2037). The analysis of data shows that a volume of 1.67 million cubic meters per second (MCM) on annual average is reduced from the total volume of Baroon dam due to the inflow discharge of sediment loads. Also, it is observed that the reservoir of the dam will be inoperable after 30 years because sediments will reach the minimum level of operation in the dam reservoir. It would be concluded that using experimental models such as area-reduction method for modeling sediment distribution in dam reservoirs is a complicated and costly process. Also, having other datasets such as sediment particle shape and density for modeling and generalization is not considered in the area-reduction method, while application of ANN can predict sediment distribution in dam reservoirs and also reservoir volume reduction properly and precisely.