摘要
分析了遥感月蒸发蒸腾量数据的动态变化趋势,把一维遥感月蒸发蒸腾量输入空间映射到高维输入空间,将蒸发蒸腾量时间序列重构为12维相空间,建立了基于支持向量机的蒸发蒸腾量预测模型。根据预测精度,确定了损失系数ε、惩罚因子C及径向基核函数的宽度σ。通过对48个训练样本的学习,得到拟合样本平均相对误差为3.51%;将模型应用于12个样本的预测,预测平均相对误差为12.30%,预测值与实测值的确定性系数达0.97。结果表明,支持向量机(SVM)模型泛化能力强,具有较满意的预测效果。研究结论较好地解决了小样本、过学习、高维数、局部最小等问题。
Statistical learning theory is a small-sample statistics theory. Support vector machine（SVM） is a new machine learning method based on statistical learning theory. It may solve the problems of non-linear classification and regression in sample space. By analyzing the dynamic variation of remote sensing evapotranspiration（ET） series,the original one-dimension ET time series input space was first projected on a high dimension input space. ET series were reconstructed as twelve dimension phase space. ET predicting model based on support vector machine was set up. According to predicting accuracy,loss coefficient e, penalty parameter C and width of the radial basis function kernel were determined. By learning 48 training samples, the mean squared error of fitting samples was 3.51%. The model was used to predict 12 samples,the mean squared error of predicting samples was 12.30%,and the deterministic coefficient between predicting values with real values was 0. 97. The results show that SVM possesses stronger generalization ability and higher prediction accuracy. It was helpful to solve small-sample,over-fitting learning, high-dimensional and local minimum problems.
出处
《太原理工大学学报》
CAS
北大核心
2011年第2期188-192,共5页
Journal of Taiyuan University of Technology
基金
教育部国家外国专家局111创新引智计划基金(B08039)
全球环境基金(GEF)(MWR-9-2-1)
关键词
蒸发蒸腾量
统计学习理论
支持向量机
预测
evapotranspiration（ET）
statistical learning theory
support vector machineprediction
