Prediction of suspended sediment load using ANN GA conjunction model with Markov chain approach at flood conditions

Development of cities, destruction of forests and pastures, population growth and other factors have increased suspended sediment load in rivers of developing countries. Measurement of suspended sediment load is a challenging issue for engineers at flood conditions. The case study of this research is Idenak hydrometric station on the Marun River in the south west of Iran. The used data consists of 42 years (1968 to 2009) flood discharge. A perceptron artificial neural network is trained and validated by observed data. For training of Artificial Neural Network (ANN), momentum and Levenberg-Marquardt training methods are applied. For decreasing of Normalized Mean Square Error (NMSE) and increasing of correlation coefficient (R), parameters of ANN are optimized by Genetic Algorithm (GA) method. GA method optimizes the number of nodes of hidden layers of ANN that is trained by Levenberg-Marquardt training method while it optimizes the number of nodes and momentum of ANN that is trained by momentum training method. GA method can reduce NMSE to 80% while GA method doesn't increase R significantly. In order to predict flood condition, the Markov chain method is applied. The results show that suspended sediment load may be increased almost from 400000 ton/day to 800000 ton/day at future.