Comparison of Back Propagation Neural Network and Genetic Algorithm Neural Network for Stream Flow Prediction

Comparison of stream flow prediction models has been presented. Stream flow prediction model was developed using typical back propagation neural network (BPNN) and genetic algorithm coupled with neural network (GANN). The study uses daily data from Nethravathi River basin (Karnataka, India). The study demonstrates the prediction ability of GANN. The statistical tests show that GANN model performs much better when compared to BPNN model. 1. Introduction Stream flow prediction for a river has been one of the most explored areas of research during recent days. Predicting the flow may facilitate its monitoring. Prediction of stream flows with good probability and reliability is of great concern. Precise prediction of stream flow gives a clear picture of the available water resources. It may also facilitate improved planning and optimum utilization of water. Many factors influence stream flow such as catchment characteristics and geographical and meteorological factors. Stream flow models may show high nonlinearity. From the second half of the last century, different methods such as physical, empirical, and numerical methods and other hybrid black box models have been practised for stream flow prediction. Main drawbacks observed in using physical model are the requirement of a more accurate and large data set which is tedious to acquire. The black box models may have an advantage at this context as they require minimum data and may provide satisfactory results. Neural network (NN), genetic algorithm, and fuzzy and hybrid algorithms are some of the methods which have received lots of attention among all modelling techniques during recent decades. The potential of NN had already been demonstrated in the context of river flow [1, 2] and dissolution kinetics [3] emphasizing the prediction ability of NN models. NN models were capable of reconstructing rainfall runoff relationships [4]. NN has proven an alternative to conventional rainfall runoff models and its strength in adaptive learning was shown for flow forecasting in the study [5]. Probabilistic forecasting accuracy was achieved using NN [6]. Modelling potential of NN was compared to a physical model and it was proven that NN has good prediction capability [7]. Good prediction accuracy and flexibility of NN were demonstrated in the studies [8, 9]. The ability of neuroemulation to imitate the behaviour of real cases and capture nonlinearity has made it a suitable method for modelling. Back propagation learning algorithm using gradient (steepest descent) based approach is widely used in the neural network