摘要
文章分析了BP神经网络结构对预报精度的影响,并提出了相应的改善方法:(1)对于简单的单输出网络预报模型,输入层神经元设置为3个时,取得较好的预报精度;(2)隐含层单元数的确定中,试算法优势明显;(3)采用COS型变换对实测数据进行处理,可以提高数据平滑度,有利于提高预报精度;(4)在网络训练中采用增设监控样本的方法可以防止过适应现象的产生。文章最后将BP网络应用于白杨河水库进行入库流量中长期水文预报,取得了较好的预报结果,合格率达84.2%,接近国家标准规定的甲级预报水平。
This paper analyses influences of BP network model structure on the precision of hydrological forecast, and proposes improvement methods： （1）For the simple single output network forecasting model, if input layer of neural network were set to be 3 neurons, there will be better forecast precision. （2）In the determination of the number of hidden layer unit, trail method has obvious advantages. （3）Using COS transform to process measured data can improve the data smoothness, and be beneficial to improve forecast precision.（4）Adding monitory sample in network training could avoid over-adaption phenomenon. In the last part of this paper BP network model was used to medium and long-term prediction of reservoir inflow in Bai Yang river reservoir, and got the better forecasting result. The qualified rate was 84.2%, close to the first grade of national standard.
作者
王鹏
李立
谢鹏
Wang Peng Li Li Xie Peng(Inner Mongolia Resources and Hydropower Survey and Design Institute, Hohhot 010020, China)
出处
《江苏科技信息》
2017年第3期27-31,共5页
Jiangsu Science and Technology Information
关键词
水文预报
BP神经网络
网络结构
预报精度
hydrologic forecast
BP network model
network structure
forecast precision
