Comparison of different deep learning optimizations in river flow prediction

Water is the most important source of life in human life for centuries. Due to the growing population, the need for water on earth is increasing day by day. In contrast to this increase, water resources; global warming, drought, climate changes, unplanned consumption is desiphering and losing sustainability. The forward forecast of river flows is of great importance in order to ensure sustainability. If the predictions are made correctly; major improvements can be made in the future management, operation, storage and correct use of water. The input-output account of the waters that have recently rened in rivers is predicted by forward-looking artificial intelligence techniques. The long-term estimates provide suitable planning for both the producer and the user for the production of water resources for living beings, irrigation, hydroelectric power generation and the transfer of water to future generations. In this study, flow measurements made obtained 2002-2011 at a selected FMS in the Euphrates River Basin, the largest basin in our country, are analyzed. In this analysis, Artificial Neural Networks and Deep Learning enhancers are used. Also in this analysis, 4 different scenario models were used and R²= 0.9923 in Adam optimizer and Logcosh loss function for the most statistically significant scenario. The statistical success of the model will provide much more accurate and easy estimation of the new input parameters that will occur on this basin, and will also shed light on other studies in this area. Keywords: ANN, Deep learning, River Stream, Flow Measurement Station