基于SVR多维时间序列分析及其在农业科学中应用.pdfVIP

中国农业科学 2008,41(8):2485-2492 Scientia Agricultura Sinica doi: 10.3864/j.issn.0578-1752.2008.08.039 基于SVR的多维时间序列分析及其在农业科学中的应用 袁哲明，张永生，熊洁仪 （湖南农业大学生物安全科学技术学院，长沙 410128 ） 摘要：【目的】建立一种基于结构风险最小、既反映样本集动态特征又体现环境因子影响的高精度非线性多 维时间序列预测方法。【方法】耦合支持向量机回归（SVR）和带受控项的自回归模型（CAR），以留一法基于 MSE 最小原则实施模型定阶和变量筛选，以一步预测法检验新模型 SVR-CAR 的有效性，并通过强制汰选给出各保留变 量对预测的相对重要性次序。【结果】3 个农业科学实例验证表明，SVR-CAR 在 7 种参比模型中预测精度最高，且 可更精细地反映样本集的非线性动态特征，依各保留变量对预测的相对重要性次序及其动态变化可赋予保留变量 部分解释能力。【结论】SVR-CAR是一种基于SVR并融合时间序列分析和回归分析的非线性多维时间序列分析方法， 具结构风险最小、非线性、适于小样本，能有效克服过拟合、维数灾和局极小，非线性定阶和非线性筛选变量， 自动选择核函数及其相应参数，泛化推广能力优异、预测精度高等诸多优点，在农业科学、生态学、经济学等领 域有广泛应用前景。 关键词：多维时间序列；支持向量机回归；预测；均方误差 Multidimensional Time Series Analysis Based on Support Vector Machine Regression and Its Application in Agriculture YUAN Zhe-ming, ZHANG Yong-sheng, XIONG Jie-yi (Bio-safety Science and Technology College, Hunan Agricultural University, Changsha 410128) Abstract: 【Objective 】To construct a novel nonlinear multidimensional time series approach based on structural risk minimization, which shows the dynamic characteristics of sample set as well as the effect of environmental factors. 【Method 】 Integrated controlled autoregressive (CAR) into support vector machine regression (SVR), a novel nonlinear multidimensional time series model named as SVR-CAR was proposed. After estimated the order and screened the variable by leave-one-out method based on the minimum mean square error, the reliability of SVR-CAR was validated by one-step prediction method. 【Result 】 The prediction results of the agricultural sample set showed that SVR-CAR had the highest prediction precision in all reference models, characterized t