Application of Shannon Entropy for Selecting the Optimum input Variables in River Flow Simulation using Intelligent Models (Case Study: SofyChay)

Accurate prediction of the river flow is an important element in the management of surface water resources, dam reservoir operation, flood control and drought. Selecting appropriate inputs for intelligent models is vital to increase the accuracy and efficiency of the models. Since river flow prediction is of great importance in water resources, researchers have been exploring different approaches over the past several decades. Various methods have been devised to predict the flow of the river over the past years. In general, we can classify conceptual models and data-driven methods. Over the past four decades, time series models have been widely used in river flow prediction (Dawson et al., 2008). Intelligent systems are used to predict nonlinear phenomena. The Bayesian Network and the Artificial Neural Network are among these methods. Ahmadi et al. (2014) studied the comparison of performance of support vector machine and network methods in forecasting daily flow of the Barandozachay River. The results showed that, both methods are close to each other and are suitable for river flow simulation. But in mid-range forecasting and the minimum backup car model, it's much better than the business network model. Shannon entropy theory was first developed by Shannon and then widely used in various scientific issues.
The purpose of this study is to use the Shannon Entropy Theory to find the best combination of input variables for artificial neural network and Bayesian network models to predict the flow. Therefore, for this purpose, the Sufi River of the studied area was selected.