基于EMD的相空间重构极限学习机预测模型  

Prediction model for phase space reconstruction of extreme learning machine based on EMD

作　　者：余宇婷 金文[1] 张珊珊[1] 邱桃荣[1] 白小明[1] 

机构地区：[1]南昌大学信息工程学院,江西南昌330031

出　　处：《计算机工程与设计》2017年第9期2515-2524,共10页Computer Engineering and Design

基　　金：国家自然科学基金项目(61070139;81460769);江西省科技计划基金项目(20112BBG70087)

摘　　要：针对气象领域中现有气象预测模型存在的准确率不高、建模时间过久、受噪音影响太大等问题,通过分析气象数据具有的时序性、混沌性和存在噪音等不确定特征,拟采用一种通过模态分解并构建于相空间基础之上的极限学习机技术。通过利用经验模态分解减少数据噪音,利用相空间重构技术提升学习规则提取和模型构建的有效性,并根据极限学习机所具备的快速构建的特性来构建预测模型。针对经典经验模态分解技术在IMF的筛选的阈值计算方法存在过筛选或者噪音数据去除不完整、去噪效果不理想等问题,提出动态阈值设定方法。在真实的数据集上对该预测模型进行对比测试,测试结果表明,该模型具有更好的预测精度,有效降低了数据噪音的影响。Aiming at solving the problems of the existing meteorological elements model, such as low accuracy, long modeling time and serious noise interference, by analyzing the features of meteorological data such as sequential order, chaos and the exis-tence of uncertainty like noise, the sequential order predicting model of phase space reconstruction extreme learning machine based on the empirical mode decomposition was studied and constructed. Empirical mode decomposition was used to effectively reduce noise data, phase space reconstruction technology was used to improve the effectiveness of the learning rules extraction and model construction, and extreme learning machine was used to quickly build effective prediction model. As for the problems of excessive filtering, removing noise data incompletely, ineffectively denoising, etc. in the IMF screening threshold calculation methods of classical empirical mode decomposition technique, a dynamic threshold setting method was proposed. Results of test and analysis on real data sets show that the forecast model proposed not only has better prediction precision, but also effectively reduces the effects of noise data, which illustrates the effectiveness of the model.

关 键 词：经验模态分解 相空间重构 极限学习机 气象预测 时序预测模型 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]