干旱气象

• 业务技术应用 • 上一篇    下一篇

张掖湿地公园水域结冰厚度预报的BP 神经网络与统计回归方法对比

刘洪兰张 强赵小强张浩文
  

  1. 1. 甘肃省张掖市气象局,甘肃 张掖 734000;2. 中国气象局兰州干旱气象研究所,甘肃省干旱气候变化与减灾重点实验室,中国气象局干旱气候变化
    与减灾重点开放实验室,甘肃 兰州 730020
  • 收稿日期:2012-11-28 出版日期:2013-07-10 发布日期:2013-03-21
  • 作者简介:刘洪兰(1968 - ),女,山东招远人,高级工程师,主要从事天气、气候变化和预测的业务和科研工作. E - mail:gszylhl@ 126. com
  • 基金资助:

    甘肃省气象局气象科研项目“张掖国家湿地公园水域结冰厚度预报服务系统研究”(2012 -08)及甘肃省气象局第六批“十人计划”共同资助

Comparison Analysis of BP Neural Network and Statistical Models for Forecasting Icing Thickness of the Zhangye National Wetland Park

LIU Honglan1,ZHANG Qiang2,ZHAO Xiaoqiang1,ZHANG Haowen1   

  1. 1. Zhangye Meteorological Bureau of Gansu Province,Zhangye 734000,China;
    2. Key Laboratory of Arid Climatic Change and Reducing Disaster of Gansu Province,Key Open Laboratory of Arid Climatic Change and Disaster Reduction of China Meteorological Administration,Lanzhou Institute of Arid Meteorology,CMA,Lanzhou 730020,China
  • Received:2012-11-28 Online:2013-07-10 Published:2013-03-21

摘要:

利用张掖国家湿地公园冬季水域结冰厚度观测资料和张掖观象台的气温、地温气象资料,运用统计学方法和BP神经网络方法建立了张掖国家湿地公园水域结冰厚度预报方程。通过对不同的预报方法进行预报效果验证,该结冰厚度的预报模型能够对结冰厚度有比较理想的预报效果,流动水域结冰厚度预报历史拟合率分别为:80.6% (多元回归)、74.6(逐步回归)、100% (BP  神经网络);模型试报准确率分别为:72.7%(多元回归)、72.7% (逐步回归)、81.8%(BP  神经网络)。静止水域结冰厚度预测历史拟合率分别为:76.9%(多元回归)、71.8%(逐步回归)、93.5%(BP 神经网络);模型试报准确率分别为:76.0%(多元回归)、72.0% (逐步回归)、84.0%(BP  神经网络)。结果表明:多元回归方法优于逐步回归方法,而BP  神经网络又明显优于传统的统计学方法,数据显示该结冰厚度的预报模型能够对结冰厚度有较好的预报效果,预报模型能够对水域结冰厚度进行有效的短期预报,其性能指标符合实际要求,具有很好的实际应用价值。

关键词: 水域, BP 神经网络, 统计预报, 模型, 结冰厚

Abstract:

Based on the observations of the ice thickness of the water area of the Zhangye National Wetland Park,and air temperature and ground temperature observed by the Zhangye Meteorological Observatory,the icing thickness prediction equation was established by using statistical and BP neural network methods. Through verifying the forecast effects of different forecasting methods,the prediction model could forecast the icing thickness better. The historical fitting rate of icing thickness forecast of flowing waters is 80. 6% ( multivariate regression) ,74. 6% ( stepwise regression) ,100% ( BP neural network) ,respectively,and the model forecast accuracy is 72.7% ( multivariate regression) ,72. 7% ( stepwise regression) ,81. 8% ( BP neural network) ,respectively. The historical fitting rate of icing thickness forecast of static water is 76. 9% ( multivariate regression) ,71. 8% ( stepwise regression) ,93. 5% ( BP neural network)
,respectively,and the model forecast accuracy is 76. 0% ( multivariate regression) ,72. 0% ( stepwise regression) ,84. 0%( BP neural network) ,respectively. Results show that the multiple regression method is better than that of stepwise regression method,and the BP neural network is superior to traditional statistical methods,and the prediction model of icing thickness had better forecast effect. It is also showed that these prediction models have significant short - term forecast skills and can be used practically.

Key words: water area, BP neural network, statistical forecast, forecast model

中图分类号: