摘要
小波神经网络作为国际上新兴的一种数学建模分析方法,充分继承了小波变换良好的时频局部化性质及神经网络的自学习功能和极强的非线性能力等优点。降水量预测模型中神经网络选择BP网络,隐含层激发函数选取Morlet小波,并利用MATLAB编写预测程序,运用吉林西部地区白城、长岭、前郭3个测站1957-2010年的降水资料对模型进行训练、检验,进而预测三站未来十年的降水量。研究结果表明,小波神经网络预测模型对降水量的变化趋势预测准确,结构简单,收敛速度快,具有较高的实际应用价值,但其对于降水量具体值的预测精度有待于进一步提高;未来十年,吉林西部地区将处于降水量变化周期的丰水阶段,各相关部门应根据实际情况做好相应的准备。
As an emerging international mathematical modeling method, wavelet neural network （WNN） tully inherits many advantages, such as excellent time-frequency localization property of wavelet transform and self-learning function and strong ability nonlinear of neural network. In this paper, BP network is selected to be the neural network, the Morlet wavelet is chosen to be the hidden excitation function of precipitation prediction model, the MATLAB is used to write WNN prediction program, and then the model is trained and tested by using the precipitation information from 1957 to 2010 of the three hydrometric stations of Western Jilin prov- ince, including Baicheng, Changling, Qianguo, to forecast their precipitation of next decade. The research results show that the WNN prediction model is of high accuracy, simple structure, fast convergence rate and high practical value, while the prediction ac- curacy for specific value of precipitation needs further improvement. In addition, Western Jilin province will be in wet stage of the precipitation variation during next decade, so related departments should make preparations according to the actual situation.
出处
《节水灌溉》
北大核心
2013年第3期31-34,共4页
Water Saving Irrigation
基金
吉林省科技发展计划项目(20080456)
作者简介
侯泽宇（1989-），男，硕士研究生，研究方向为水环境与水生态。E-mail：houzeyu890829@163．com。
通讯作者：卢文喜（1956-），男，教授，博士生导师，主要从事地下水数值模拟及水生态研究。E-mail：luwenxi@jlu．edu．cn。
