基于数据分解和权重优化的VOCs浓度区间预测  

VOCs concentration interval prediction based on data decomposition and weight optimization

作　　者：杨帆[1] 江松[1,2] 黄光球[1] 蒋国炜 YANG Fan;JIANG Song;HUANG Guangqiu;JIANG Guowei(School of Management,Xi'an University of Architecture and Technology,Xi'an 710055;School of Resources Engineering,Xi'an University of Architecture and Technology,Xi'an 710055;Inner Mongolia Xianhong Technology Co.,LTD,Hothot 011517)

机构地区：[1]西安建筑科技大学管理学院,西安710055 [2]西安建筑科技大学资源工程学院,西安710055 [3]内蒙古显鸿科技股份有限公司,呼和浩特011517

出　　处：《环境科学学报》2024年第11期50-61,共12页Acta Scientiae Circumstantiae

基　　金：国家自然科学基金(No.52374136,71874134);内蒙古呼和浩特市科技局项目(No.2023-高-12)。

摘　　要：准确的VOCs浓度预测对预防和控制空气污染具有重要意义.针对VOCs浓度波动的随机性和不确定性特征,本文提出了分解集成模型用于VOCs浓度的点和区间预测.首先,采用变分模态分解将VOCs浓度序列分解为多个平稳的模态分量以降低数据的复杂度;然后,基于深度自编码器模型实现各模态分量的独立预测,并通过鲸鱼优化算法计算集成权重得到VOCs浓度预测结果;最后,基于核密度估计得到预测区间,以量化点预测的不确定性.采用实际监测数据对模型的有效性进行验证.结果表明,数据分解和集成权重优化有效提高了VOCs浓度预测精度,且基于核密度估计得到的预测区间在保证区间覆盖率的同时能够缩小区间宽度,增强了预测区间可靠性.Accurate VOCs concentration prediction is significant for air pollution prevention and control.To address the stochasticity and uncertainty of VOCs concentration fluctuations,we propose a decomposition integration model for VOCs concentration point and interval prediction.First,variational modal decomposition is used to decompose the VOCs concentration series into multiple stable modal components to reduce the data complexity.Then,each modal component is independently predicted based on deep autoencoder model,and the integrated weight is calculated by whale optimization algorithm to get the VOCs concentration prediction results.Finally,prediction intervals are obtained based on kernel density estimation to quantify the uncertainty of point predictions.The actual monitoring data were used to validate the validity of the model.The results show that data decomposition and integrated weight optimization can effectively improve the prediction accuracy of VOCs concentration,and the prediction intervals obtained based on kernel density estimation can narrow the width of the intervals while guaranteeing the coverage of the intervals,which enhances the reliability of the prediction intervals.

关 键 词：VOCs浓度 区间预测 变分模态分解 深度自编码器 鲸鱼优化算法 核密度估计 

分 类 号：X511[环境科学与工程—环境工程]