A Study On Runoff Prediction In Qira River Basin Based On Artificial Neural Network

Water resources are the most important factor in sustaining the biological survival and socio-economic development of an area.At the same time,the changes in runoff flowing through this area are dominated by water resources of the entire region.The changes in runoff are affected by the combined effects of various factors such as the geographical characteristics of the basin,climate changes,biological activities,and social development.At the same time,the variation of the inner-diameter flow in the basin has random features with multiple time scales and non-linearity,and has strong uncertainties and randomness and other uncertainties.Therefore,the establishment of appropriate runoff forecasting models in specific basins is a necessary condition for ensuring economic growth,social harmony,ecological restoration,and environmental friendliness in the basin.In this paper,the cumulative runoff data for the 28 years from 1981 to 2008 monitored by the Qira River Hydrology Station in the lower reaches of the Qira River basin was used as the research object.In addition,the meteorological data obtained from the weather station in the middle reaches of the Qira River is set as input variables for the model.The monthly cumulative runoff data for 20 years from 1981 to 2000 was used as the training sample.Monthly cumulative runoff data from 2001 to 2008 for a total of 8years was used as a test sample,and a neural network prediction model was established.Many existing research results show that runoff simulation of rainfall and runoff models with feed-forward artificial neural networks has extremely high prediction accuracy and applicability in rainfall-rich areas.However,in runoff basins with less rainfall or more drought,the research on the rainfall runoff model is very scarce.Especially in the arid oasis plains,the existence and development was depend mainly on runoff originated from high altitude mountains.The quantification of runoff inflows with significant climate change can provide a good basis for management decision makers in arid oasis areas in water resources management and oasis development planning.Therefore,this paper selects three types of three-layer back-propagation feed-forward artificial neural networks with similar structure as the model,and uses the Qira River flowing through the Qira Oasis Plain as a typical study object,to train and predict Qira River's monthly cumulative runoff over the next month.The results show that the training accuracy of the compact wavelet neural network is low,but from the analysis of the prediction results,which can better reflect the overall law of the sample in the learning process.Both the general artificial neural network model and the radial basis neural network model have higher precision in the training and prediction stages.However,the training and prediction accuracy of traditional artificial neural network models are more sensitive to the selection of input variables,and a large number of numerical simulations are required to determine the appropriate input variables and hidden layer neurons.Therefore,the radial basis neural network model is more suitable for the study of such problems and can be extended to other similar research areas on the southern edge of the Kunlun Mountains.