基于数据分解的大气污染物短期预测组合方法综述

A REVIEW OF HYBRID FORECASTING METHODS FOR ATMOSPHERIC POLLUTANTS IN SHORT-TERM BASED ON DATA DECOMPOSITION

大气污染物的短期预测,对制定有效的大气环境治理措施和降低居民健康风险具有重要的实际参考价值。组合模型通过数据分解挖掘时间序列中蕴含的时频信息,可以进行精准且可靠的预测,已成为大气污染物短期预测的发展趋势。从时间尺度,将现有的大气污染物短期预测方法进行梳理,重点综述了基于小波分解、经验模态分解和变分模态分解的组合模型。随后,依据处理目的,将现有模型组合结构的优化方向归纳为数据降噪、二次分解、分量处理和误差修正,并对各结构的优缺点与适用范围进行总结。结果发现:4种组合结构并非普遍适用于所有预测情况,应根据数据特征等条件有选择性地使用。最后,总结了现存组合预测模型存在的问题,指出未来应从自适应组合结构、数据特征对性能影响和模型多性能平衡的角度开展相关研究。

Atmospheric pollutants’short-term prediction is of great significance to formulate effective control measures for the atmospheric environment and reduce the health risk on residents.Hybrid model can make accurate and reliable prediction by mining time-frequency information contained in time series via data decomposition,becoming the development trend of atmospheric pollutants’short-term prediction.The existing short-term prediction models of atmospheric pollutants were sorted out from the time scale.Meanwhile,the hybrid models based on wavelet decomposition,empirical mode decomposition and variational mode decomposition were reviewed.Subsequently,according to the aims,the prediction structure of the hybrid model was summarized into data de-noising,secondary decomposition,component processing and error modification.The advantages,disadvantages,and application range of each structure were summarized.The results showed that the four hybrid structures were not universally applicable to all situations and should be used selectively according to data characteristics and other conditions.Finally,the issues of existing hybrid prediction models were summarized.It was pointed out that the future research should be carried out from the perspectives of the adaptive hybrid structure,the impact of data characteristics on performance and the balance of multi-performance of the model.