GA-BP神经网络模型在流域面雨量预报的应用研究
GA-BP ANN MODEL FOR RIVER CATCHMENT PRECIPITATION FORECAST
-
摘要: 通过采用遗传算法优化网络初始权重的方法,将遗传算法(GA)和前馈误差反传播(BP)算法有机地结合,优势互补,并应用于流域面雨量预报。以广东省东北部的滨江流域为试验区域,以1995~2001年气象探空资料为基础,利用最优子集回归技术进行预报因子筛选,建立了流域面雨量预报的GA-BP神经网络模型,取得了满意的结果。试验表明:(1)6小时流域面雨量预报神经网络的优化结构是7-7-1,转移函数的组合方式为tansig-线性函数。(2)训练算法为Levenberg-Marquardt算法(LM)。(3)遗传算法具有快速学习网络权重的能力,对BP网络易陷于局部极小点。(4)利用GA-BP神经网络模型对未来6小时流域面雨量的预报精度明显高于其它统计方法,证明了这种方法的有效性和可靠性。Abstract: The method is taken to join the genetic algorithm(GA) and BP algorithm together and supplement mutually by optimizing the initial weights of ANN with GA,and some applications have been made in the Binjiang River catchment for precipitation forecast.The ANN model by GA has been established in which forecasting variables are selected by optimizing the subclass regression technique on the base of radiosonde data(from 1995 to 2001) and the optimized ANN model for 6 hours precipitation of Binjiang River catchment has been obtained.The optimized ANN structure is 7-7-1 and its transfer function is tansig-pureline and training functions is Levenberg-Marquardt(LM).The genetic algorithm can speed up the learning process of network weights and solve the local searching problem of BP network.The experiment result shows that this method can enhance the forecast precision of 6-hours precipitation compared with other statistical methods,and its effectiveness and the reliability have been proved.
-
[1] NIELSEN R Hecht.Theory of the back propagation neural network[J].Proc of IJCNN,1989,l1:593-603. [2] 王成刚,吴宝俊.BP网络在鲁西南地区西南涡降水量级预报中的应用试验[J].气象科学,1999,19(2):158-165. [3] ROBERT J K,BARROS A P.Experiments in short-term precipitation forecasting using artificial neural networks[J].Mon Wea Rev,1998,126(3):470-482. [4] BAIK J J,WANG H S.Tropical cyclone intensity prediction using regress method and neural network[J].J Meteor Soc Japan,1998,76(5):711-717. [5] 严绍瑾,彭永清,郭光.神经网络BP模型用于月降水预报的研究[J].热带气象学报,1995,11(3):252-257. [6] 胡江林,涂松柏,冯光柳.基于人工神经网络的暴雨预报方法探讨[J].热带气象学报,2003,19(4):422-428. [7] 王繁强,徐文金,陈杰伦.BP算法在青海省降水雨区分区分级预报中的应用[J].高原气象,1997,16(2):105-112. [8] 王向东,葛文忠.基于神经网络方法的华东地区强对流天气短时预报[J].模式识别与人工智能,1998,11(3):323-327. [9] 王巧霞,王洪俊.人工神经元网络中期分级降水预报系统[J].山西师大学报(自然科学版),1999,13(1):64-68. [10] 金龙,陈宁.基于人工神经网络的集成预报方法研究和比较[J].气象学报,1999,57(2):198-207. [11] 刘还珠,汤桂生.暴雨落区预报实用方法[M].北京:气象出版社,2000:30-37. [12] 陈云浩,史培军.不同热力背景对城市降雨(暴雨)的影响Ⅲ--基于人工神经网络的集成预报模型[J].自然灾害学报,2001,10(3):26-31. [13] 刘国东,丁晶.神经网络用于水文预测的几个问题探讨[J].水利学报,1999,1(1):65-70. [14] 齐卫忠,郑雄波.气象卫星云图降雨预报系统的应用[J].华中电力,2003,14(3):57-59. [15] 孙永刚,李彰俊.天气动力学组合因子在MOS降雨预报中的应用[J].气象,1998,2(1):27-30. [16] 俞善贤,汪锋.试用最优子集与岭迹分析相结合的方法确定回归方程[J].大气科学,1998,12(4):382-388. [17] 施能,曹鸿兴.基于所有可能回归的最优气候预测模型[J].南京气象学院学报,1992,15(4):459-466. [18] 魏凤英,曹鸿兴.长期预测的数学模型及其应用[M].北京:气象出版社,1990:9-90. [19] 张敏,赵金城.全局优化神经网络拓扑结构及权值的遗传算法[J].大连大学学报,1999,20(6):9-13. [20] HSU K,CUPTA H V,SORROSHIANS.Artificial neural net workmodling of the rainfall runoff process[J].Water Resources Research,1995,31(10):2517-2530. [21] HOWARD Demuth,MARK Beale.Neural network toolbox user's guide for use with matlab[M].The Mathworks Inc,1998:195-239. [22] 翟宜峰,李鸿雁.用遗传算法优化神经网络初始权重的方法[J].吉林大学学报(工学版),2003,33(2):45-50.
点击查看大图
计量
- 文章访问数: 1030
- HTML全文浏览量: 0
- PDF下载量: 1145
- 被引次数: 0