River Flow Prediction Using Genetic Programming (Case Study: Lighvan River Watershed)

The genetic programming and artificial neural networks as well as time series and (neuron) fuzzy logic, are used in predicting the river flow. In the present study, the genetic programming was applied to predict daily river flow of Ligvan river in Urmia lake watershed for the period of 1997 to 2001 and the memory rule was investigated in decreasing or increasing of forecasting accuracy. In order to model the river flow by genetic programming, the river flow discharge of over 5 days with daily time steps were used. The resulted values of river flow were evaluated by statistical measures, includes root mean square error and correlation coefficient. The results showed the feasibility of employed genetic programming for over 4 days predictions intervals in term of correlation coefficient (0.959) and root mean square error (0.029). Application of artificial neural networks in prediction of river flow had the same trend as for the genetic programming, but with a relatively low accuracy. The best structure of the neural network was three layered network with 4, 6, and 1 neuron in input, hidden and output layers, respectively, with a correlation coefficient of 0.948 and root mean square of 0.215. Therefore, the proposed genetic programming model could be successfully used in modeling the daily river flow.