关键词: 小波变换, GRNN神经网络, 出山径流, 逼近系数, 细节系数

Abstract: The arid area of Northwest China is situated in the inland area of Asia far from the oceans. However, many enormous mountains receive much more precipitation compared with the low land area in front of the mountains. Therefore, glaciers and snow storage develop very well in the mountains. In an inland basin, the mountainous watersheds are runoff drainage basin, while the low land plains and basins in front of the mountains are the areas of water resources consumption and runoff scatterings. Therefore, the runoff amount generated from the mountains, while runs out off the outlets of the mountainous watersheds, basically represents the amount of water resources of the inland arid area. To some degree, runoff from mountainous watersheds is the radical factor of controlling the development of this region. Therefore, we should firstly find the changing regularities of runoff from mountainous watersheds. In this paper, the author simulated and predicted the runoff from mountainous watersheds of Heihe River that is the biggest river of Hexi Corridor. Since wavelet conversion and neural network are used extensively recently, we use Haar wavelet and GRNN neural network to simulated and predicted the runoff from mountainous watersheds of Heihe River. We have monthly runoff be converted, acquire the reconstructed approximation and detail coefficients, and use them as input samples of the GRNN neural network. The results are perfect. However, this method can t be used to predict the runoff in the future. For this reason, we take the second path. Let the precipitation and air temperatures converted, and then use them as the input samples of GRNN neural network, and the runoff as output samples. The simulated and predicted results are all perfect. Now, we use the global changing results to predict the changes of yearly runoff. Now are the results. If precipitation is unchanged, the air temperature rises 0.5℃ and 1℃ respectively; the yearly runoff from mountainous watersheds of Heihe River may arise 2.15% and 1.82%. If air temperature does not change and precipitation rises 10% and 20%, the yearly runoff will rise 11.35% and 26.77% respectively. If air temperature rises 0.5℃, and precipitation rises 10% at the same time, the yearly runoff should rise 11.21%. From these results, we can conclude that the yearly runoff from mountainous watersheds of Heihe River should arise in the near future, and decrease in the last. This result agrees with that of other models such as HBV model. We find this model is not sensitive to air temperature, and too sensitive to precipitation. That is not suitable for the Northwest China. One of the reasons is that the reconstructed detail coefficients of air temperature are obtuse to changing of air temperature. Therefore, we eliminate these detail coefficients. In this time, the results show much perfect. However, we find that yearly runoff arises with the air temperature. This is not suitable because of evapotranspiration. At last, we conclude that this model may suitable for the region where the runoff is obtuse to the temperature.

Key words: Wavelet Conversion, Generalized Regression Neural Network, runoff from mountainous watersheds, approximation coefficient, detail coefficient