| 引用本文： |
-
张洪波，石建军，冯民权，陈晓楠，王义民.基于EMD和IDAR的灌区年降水量预测研究[J].大英文模板1,2014,(3):196-201. [点击复制]
- ZHANG Hong-bo, SHI Jian-jun, FENG Min-quan, CHEN Xiao-nan, WANG Yi-min.Research on the annual rainfall prediction for irrigation area based on the EMD and IDAR[J].大英文模板1,2014,(3):196-201. [点击复制]
|
|
| 摘要: |
| 针对降水量这样一种非线性、非平稳序列，研究经验模态分解方法(EMD)和信息扩散近似推理方法（IDAR）在年降水量预测中的组合应用，解决资料序列不充分情况下的区域降水量预测问题。首先，通过EMD方法对具有典型非线性与非平稳特征的年降水量时间序列进行处理，分解出包含原信号不同特征尺度的分量数据系列；然后应用信息扩散近似推理技术对各降水量分量间的复杂非线性关系进行描述，建立当前趋势以及相邻年份之间的预测规则，并进行预测。以文献案例灌区长系列降水资料为样本进行实例计算，并与其它预测方法进行了对比。结果表明：基于EMD和信息扩散近似推理的预测方法效果较好，误差绝对值和为1.30，优于人工神经网络、线性自回归方法以及单纯信息扩散近似推理的统计结果。同时，为了验证该方法的适用性，将该方法应用于文峪河灌区的降雨量预测，取得了满意的效果。研究中发现：信息扩散近似推理可将样本点转换成模糊集，部分弥补了由于数据的不完备性所造成的信息空白，并可将矛盾模式转换成兼容模式。而EMD方法可有效分解具有非线性、非平稳特征的降水序列，保留其原序列在空间（或时间）各种尺度上的分布规律。两者结合对解决样本不完备的非平稳序列的预测问题是非常有价值的。通过与其它预测方法比较，发现该模型能够很好地光滑样本数据以及能够较好地发掘知识，有较高的预测精度和推广应用价值。 |
| 关键词: 经验模态分解方法(EMD) 信息扩散近似推理(IDAR) 年降水量 预测模型 灌区 |
| DOI： |
|
| 基金项目:国家自然科学基金（51009009，51379014）；流域水循环模拟与调控国家重点实验室开放研究基金（IWHR-SKL-201109）；国土资源部干旱半干旱地区水资源与国土环境开放研究实验室开放基金（2013G1502044） |
|
| Research on the annual rainfall prediction for irrigation area based on the EMD and IDAR |
| ZHANG Hong-bo1, SHI Jian-jun2, FENG Min-quan2, CHEN Xiao-nan3, WANG Yi-min2 |
| () |
| Abstract: |
| According to rainfall such a nonlinear, non-stationary sequence, this paper carried out a research combining empirical mode decomposition(EMD) and information diffusion approximate reasoning method (IDAR) method to solve the problem on annual precipitation forecast under the condition of insufficient data material sequence. First, the annual precipitation time series with typical nonlinear and non-stationary feature was made data processing by the EMD method. Some sub-data series including different scale features were decomposed out from the original material. Then we applied the information diffusion approximate reasoning technology to describe the nonlinear relation within precipitation component, and established a series of prediction rules between adjacent years based on the current trend for forecasting. In this paper a long irrigation precipitation data was as the samples for calculation, and its computing result was compared with other prediction method. The
contrast results showed that: The comprehensive forecast method based on IDAR and EMD was highly effective when precipitation forecasting. The sum of absolute values of the result error was only 1.34, which was better than other statistical results by artificial neural network, linear regression method and original information diffusion approximate reasoning. At the same time, this method has b
een used in the rainfall forecast in Wenyu River irrigation district, has obtained satisfactory results. In the research process, we also have found: The information diffusion approximate reasoning technology can convert a sample points to a fuzzy set, which can partly offset the information shortage due to sample data insufficient to a certain extent. Furthermore, the fuzzy treatment in IDAR also can convert the problem from contradiction mode into compatibility mode. The EMD method can effectively decompose precipitation sequence with nonlinear and non-stationary characteristics, and retain various distribution rule in space (or time) scales within their original sequence. The combination of IDAR and EMD methods is very valuable for forecast problems of the incomplete and non-stationary sequence. Through the comparison with other forecast methods, it find that the combination model has high forecasting precision and application value, due to it well smoothing the sample data and exploring information knowledge. |
| Key words: empirical mode decomposition(EMD) information diffusion approximate reasoning(IDAR) annual precipitation prediction model irrigation area |
