基于EEMD-SVM-ELM模型的月降水量预测研究  

Monthly Precipitation Prediction Based on EEMD-SVM-ELM Model

作　　者：李明[1,2] 刘东岳 赵良伟 蒋一波[3] LI Ming;LIU Dong-yue;ZHAO Liang-wei;JIANG Yi-bo(School of Business,Hohai University,Nanjing 211100,China;Institute of Project Management Informatization,Hohai University,Nanjing 211100,China;Jiangsu Huaiyin Water Conservancy Construction Co.,Ltd.,Huaian 223005,China)

机构地区：[1]河海大学商学院,江苏南京211100 [2]河海大学项目管理信息化研究所,江苏南京211100 [3]江苏淮阴水利建设有限公司,江苏淮安223005

出　　处：《水电能源科学》2024年第5期19-23,共5页Water Resources and Power

基　　金：国家社会科学规划基金资助项目(17BGL156);河海大学中央高校基本科研业务费项目(B220207039)。

摘　　要：针对地表降水量数据的非线性、非平稳特征,首先利用EEMD对月降水量初始数据进行分解,再利用Lempel-Ziv复杂度算法将分量划分为高频及低频分量,使用粒子群算法(PSO)优化基学习器参数,最终构建EEMD-SVR-ELM月降水量预测模型,并采用该模型对长江下游部分城市的月降水量实际数据进行预测。结果表明,该模型的综合性能最优,具有更高的精确度。相较于单一模型,在M_(MAE)、R_(RMSE)、M_(MAPE)指标上分别降低了37.4%、41.4%、42.5%,DM检验表明该模型显著优于其他模型,说明该模型可作为月降水量预测的一种有效新方法。Aiming at the nonlinearity and non-stationary characteristics of surface precipitation data,a support vector regression(SVR)and extreme learning machine(ELM)are constructed as base learners.Firstly,the initial monthly precipitation data is decomposed based on Empirical Mode Decomposition(EEMD).Then the Lempel-Ziv complexity algorithm is used to divide the components into high-frequency and low-frequency components.The parameters of the base learner are optimized by particle swarm optimization(PSO).Finally,the EEMD-SVR-ELM monthly precipitation prediction model was constructed.Compared with other models,the model has the best comprehensive performance,higher accuracy and generalization.Especially compared with the single model,the M MAE,RRMSE,and M MAPE indicators were reduced by 37.4%,41.4%and 42.5%.The DM test showed that this model was significantly better than other models.This model can be used as an effective new method for monthly precipitation prediction.

关 键 词：月降水量预测 经验模态分解 极限学习机 支持向量回归 

分 类 号：TV125[水利工程—水文学及水资源]