Application of GA-ANN Hybrid Algorithms in Runoff Prediction

Accurate prediction of runoffs plays an important role in managing water resources in river basins. Conducting runoff prediction is a complex and nonlinear process, therefore suitable for applying tools like Artificial Neural Networks (ANN). A typical algorithm for training the ANN is Back Propagation (BP). But there are two disadvantages for the BP algorithm: slow in convergence, and prone to fall into local extreme points. Therefore, we applied so called GA-ANN hybrid algorithms to predict runoffs. The data from Qing jiang river in between 1989 and 1991 are used as training data set, and 1992 are as testing data set. The historical 3 days daily runoff data were used to predict the upcoming (4th) day's runoff. The precision of the prediction is examined by DC and RMAE. The result showed that, the DC value of GA-ANN hybrid algorithms has increased for 0.12% compared to the traditional BP algorithm, and the RMAE value has been reduced for 1.69%. And in addition, the GA-ANN network is more stable than the traditional BP network.