摘要
积温插值是分布式积温获取的重要途径。为了提高积温插值的精度,应用BP(Back Propagation Learning Algorithm)神经网络模型和支持向量机(Support Vector Machine,SVM)模型建立甘肃省及周边地区的积温插值模型。结果显示:1)从积温插值的空间分布来看,SVM模型比BP神经网络模型能够体现出更多的细节;2)SVM模型的插值精度总体上显著高于BP神经网络模型;3)在平均相对误差(MRE)最大的西区,相比BP神经网络模型的7.19%,SVM模型将误差降低到了5.47%;4)东区两种模型的MRE最小,BP神经网络模型为2.97%,SVM模型为2.03%;5)与分区建模前相比,分区后的插值精度有所提高,BP神经网络模型将MRE降低了0.04%,SVM模型降低了0.11%。
In order to improve the rainfall interpolation accuracy in Gansu province,the precipitation interpolation models were built using two artificial intelligence techniques： Artificial neural network about Back Propagation Learning Algorithm and Support Vector Machine（ SVM） model. The preliminary results were as follows：（ 1） From the spatial distribution of accumulated temperature perspective,SVM model can reflect more details.（ 2） SVM model has higher interpolation accuracy than BP neural network model.（ 3） In the west district which has the largest MRE,compared to 7. 19% of the BP neural network model,the SVM model is reduced to 5.47%.（ 4） The east district has the smallest MRE,with 2. 97% of the BP neural network model and 2. 03% of the SVM model.（ 5） Modeling separated could improve the interpolation accuracy. But the reduction of two models was different,the BP neural network model reduced MRE by 5. 08%,while the SVM mode reduced by0. 66%. It exhibits that the SVM model is more stable.
出处
《干旱区资源与环境》
CSSCI
CSCD
北大核心
2014年第5期158-165,共8页
Journal of Arid Land Resources and Environment
基金
国家自然科学基金"中国草地分类系统的比较分析与整合研究(31160475)"
"基于近60年来高密度气象资料的前期降水和泥石流发生概率关系分析(40971016)"资助
关键词
积温
插值
BP神经网络
支持向量机
accumulated temperature
interpolation
BP neural network
support vector machine
