基于改进支持向量机的气象干旱预测方法  

Meteorological Drought Prediction Method Based on Improved Support Vector Machine

作　　者：王雪妮 郝舒哲[2] 常建波 邓显羽 WANG Xue-ni;HAO Shu-zhe;CHANG Jian-bo;DENG Xian-yu(Shanxi Geological Survey Institute,Taiyuan 030006,China;College of Water Resource Science and Engineering,Taiyuan University of Technology,Taiyuan 030024,China;Water Northeastern Investigation and Design Research Co.Ltd.,Changchun 130021,China)

机构地区：[1]山西省地质调查院,太原030006 [2]太原理工大学水利科学与工程学院,太原030024 [3]中水东北勘测设计研究有限责任公司,长春130021

出　　处：《沈阳农业大学学报》2023年第5期599-606,共8页Journal of Shenyang Agricultural University

基　　金：山西省科技厅基础研究计划项目(202203021222112)。

摘　　要：气象干旱制约着区域农业生产活动,通过监测预报、提前部署可有效降低其带来的危害。基于“先分解,后重构”思想,将自适应噪声完备经验模态分解法(complete ensemble empirical mode decomposition with adaptive noise,CEEMDAN)、灰狼优化算法(grey wolf optimization,GWO)和支持向量机(support vector machine,SVM)相结合,构建CEEMDAN-GWO-SVM组合机器学习模型。以山西省1956−2007年逐月自适应帕默尔干旱指数(self-calibrated palmer drought severity index,scPDSI)为训练集,以2008−2020年逐月scPDSI值为测试集。采用scPDSI预测值与实际值之间的均方误差、平均绝对误差及决定系数作为评价指标,对该模型的预测效果进行了评价,并比较了该模型与常用的GWO-SVM和随机森林模型预测结果。结果表明:相较于两种常用机器学习模型,基于CEEMDAN-GWO-SVM模型预测得到的均方根误差可分别降低7.6%和19.5%,平均绝对误差分别降低19.9%和27.4%,决定系数则分别提高5.6%和20.3%,预测的干旱等级和实际情况接近。CEEMDAN-GWO-SVM模型预测精度优于GWO-SVM模型和随机森林模型,在气象干旱预测中具有较好的适用性。Regional agricultural production activities are restricted by meteorological drought,whose harm can be effectively reduced by monitoring forecasting and early deployment.With the idea of“first decomposition,then reconstruction”,the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN),the Grey Wolf Optimization(GWO)and the Support Vector Machine(SVM)are combined to construct a combine machine learning model,CEEMDAN-GWO-SVM.Taking the monthly Self-Calibrated Palmer Drought Severity Index(scPDSI)from 1956 to 2007 in Shanxi Province as the training set,the monthly scPDSI values from 2008 to 2020 are used as the test set.The Root Mean Squared Error,Mean Absolute Rrror and Coefficient of Determination between the predicted value of scPDSI series and the actual value are used as evaluation indexes,and the prediction effect of the model is evaluated.The prediction results of the model are compared with the GWO-SVM and Random Forest model.The results show that compared with the two machine learning models,the root mean square error predicted by the CEEMDAN-GWOSVM model can be reduced by 7.6% and 19.5%,the average absolute error can be reduced by 19.9%and 27.4%,the coefficient of determination can be increased by 5.6% and 20.3%,respectively.The predicted drought level is close to the actual situation.It indicates that the prediction accuracy of CEEMDAN-GWO-SVM model is better than that of GWO-SVM model and Random Forest model,the proposed CEEMDAN-GWO-SVM model has good applicability in meteorological drought prediction.

关 键 词：自适应帕默尔干旱指数 自适应噪声 完备经验模态分解法 灰狼优化算法 支持向量机 干旱预测 山西省 

分 类 号：S27[农业科学—农业水土工程] P338[农业科学—农业工程]