Combination of Genetic Algorithm and Support Vector Machine for Daily Flow Forecasting

This paper applied a genetic algorithm (GA) to optimize the parameters of support vector machine (SVM) for daily flow forecasting of Chickasaw creek located in Mobile County. To investigate the impact of variable enabling/disabling of flow, rainfall and evaporation on model prediction accuracy, four model structures with different input vectors were developed and the performance of them was evaluated in terms of the mean square error and the coefficient of determination. The results show that the third model structure consisting of the past 3 days´ flow, the past rainfall and evaporation as the inputs is superior to other model structures in performance. Compared with the back-propagation network (BPN), experimental results show that the prediction accuracy of the proposed SVM model is better than the former and can be used for forecasting the daily flow in engineering management.