Study on the Precipitation Forecast Based on BP-Neural Network and Wavelet Analysis

作者: 熊海晶 * , 王式成 :安徽淮河水资源科技有限公司; 王栋 :南京大学地球科学与工程学院水科学系;

关键词: BP神经网络小波神经网络年降水量降水预报Back Propagation Neural Network Wavelet Neural Network Annual Precipitation Rainfall Forecast

摘要: 江苏省地处江淮流域,是受旱涝灾害影响最为严重的地区之一。在该地区开展降水预报的研究,对防汛抗旱具有重要意义。本文采用江苏省徐州站、赣榆站、东台站和南京站的降水数据,建立了BP神经网络和小波神经网络降水预报模型。通过实例分析得出:1) BP网络模型预报的最小相对误差为1.16%,最大相对误差为16.35%,最优确定性系数0.87,均方误差4.27%;2) WNN网络模型预报的最小相对误差为0.7%,最大相对误差为88.65%,最优确定性系数0.94,均方误差4.2%。结果表明:1) BP神经网络模型预报降水具有可行性,该模型能在一定程度上反映降水变化的趋势;2) WNN模型在某些年份预报误差较大,可在实践中将多种预报方法相互验证,相互校核,提高预报精度。

Abstract: Jiangsu province is located in the Yangtze and Huaihe River basins, which is one of the areas most severely affected by droughts and floods. Therefore, it is important to study on the precipitation forecast in this area for the flood control with drought relief. In this paper, the precipitation forecast model based on BP Neural Network and Wavelet Neural Network is established with the precipitation data from four rainfall sta-tions in Jiangsu Province, which include the Xuzhou, Ganyu, Dongtai and Nanjing Stations. The following conclusions can be obtained by the example analysis: 1) The minimum relative error of BP neural network is 1.16%, the maximum relative error is 16.35%, the determine coefficient is 0.87, the mean square error is 4.27%; 2) The minimum relative error of WNN neural network is 0.7%, the maximum relative error is 88.65%, the determine coefficient is 0.94, the mean square error is 4.2%. The results show that: 1) It is feasi-ble to apply the back propagation neural network precipitation forecast model. To a certain extent, this model can reflect the trends of the precipitation; 2) There is more error of WNN in some years, so a variety of forecasting methods in practice are used to mutual authentication and mutual checking to improve forecast accuracy.

文章引用: 熊海晶 , 王式成 , 王栋 (2012) BP神经网络和小波分析在年降水预报中的应用研究。 水资源研究, 1, 340-346. doi: 10.12677/JWRR.2012.15052


[1] 崔锦泰. 小波分析导论[M]. 西安: 西安交通大学出版社, 1995. CUI Jingtai. Introduction of wavelet analysis. Xi’an: Xi’an Communication University Press, 1995. (in Chinese)

[2] ZHANG, Q. H., BENVENISE, A. Wavelet network. IEEE Transon Neural Network, 1992, 3(6): 889-897.

[3] CHUI, C. K. An introduction to wavelet. Boston: Academic Press, 1992.

[4] AUSSEN, A., MURTAGH, F. Combining neural network forecasts on wavelet-transfomed time series. 1997, 113-121.

[5] KITAHARA, M., ACHENBACH, J. D. and GUO, Q. C. Neural network for crack-dapth determination from ultrasonic back- scattering data. Review of Progress in Quantitative Nondestructive Evaluation, 1992, 11: 701-708.

[6] DAUBECHIES, I. Orthonormal bases of compactly supported wavelets. Communications on Pure and Applied Mathematics, 1988, 41(7): 909-996.