基于变权重组合的沙尘污染物浓度预测  

Prediction of dust pollutant concentration based on variable weight combination

作　　者：王永生 谭诗怡 刘文静 刘广文 张德龙 WANG Yongsheng;TAN Shiyi;LIU Wenjing;LIU Guangwen;ZHANG Delong(School of Data Science and Application,Inner Mongolia University of Technology,Hohhot O10080;Meteorological Data Center of Inner Mongolia Autonomous Region,Hohhot 010051,China)

机构地区：[1]内蒙古工业大学数据科学与应用学院,呼和浩特010080 [2]内蒙古自治区气象数据中心,呼和浩特010051

出　　处：《干旱区资源与环境》2025年第3期152-162,共11页Journal of Arid Land Resources and Environment

基　　金：国家自然科学基金(62366039);内蒙古自治区自然科学基金(2021LHMS06001;2023LHMS06023);内蒙古自治区科技计划项目(2023YFSH0066);内蒙古自治区高等学校科学研究项目(JMZD202301);基本科研业务费(JY20220273;JY20230040;JY20240002;JY20240061);内蒙古自治区高等学校青年科技英才支持计划(NJYT23104)资助。

摘　　要：沙尘天气会使空气浑浊、空气污染物含量剧增,给人类生产生活带来恶劣影响。研究可吸入颗粒物(PM_(10))浓度变化趋势并提前预测结果,有利于提前预防沙尘空气污染物带来的危害。为提高沙尘污染物浓度预测精度,解决传统单一模型预测结果滞后及拟合性较差等问题。为此,提出一种基于变权重组合的沙尘PM_(10)浓度混合预测方法。1)利用极端梯度提升回归树算法筛选重要气象特征,并将PM_(10)和筛选出来的气象特征通过长短期记忆提取输入变量中隐藏的时序特征关系。2)使用能量熵优化变分模态分解算法对PM_(10)进行分解,利用门控循环单元对分量进行预测。3)采用变权值动态分配法组合单一预测模型的PM_(10)浓度预测值,构建基于变权重组合的沙尘PM_(10)浓度混合预测模型。在内蒙古二连浩特数据集上进行实验,实验结果表明文中提出的混合预测模型的预测精度优于单一预测模型,混合预测模型的3个评价指标都优于其他单一预测模型,拟合优度提高2%~20.3%,平均绝对误差降低45%~79.5%,均方根误差降低45.5%~76%。证明本模型能够对PM_(10)浓度进行有效的预测。Sand-dust weather renders the air turbid and significantly increases the content of air pollutants,thereby exerting a detrimental impact on human production and life.Studying the trends in the concentration of inhalable particulate matter(PM_(10))and predicting the outcomes in advance are beneficial for preventing the harm caused by dust air pollutants beforehand.To enhance the precision of predicting sand-dust pollutant concentrations and address issues such as delayed prediction outcomes and inadequate fitting in traditional single models,a hybrid prediction approach for sand and dust PM_(10)concentrations,utilizing variable weight combinations,is proposed.1)The extreme gradient boosting regression tree algorithm is used to select important meteorological features,and PM_(10)and the selected meteorological features are extracted through the long short-term memory to reveal the hidden temporal feature relationships in the input variables.2)The energy entropy-optimized variational mode decomposition algorithm is used to decompose PM_(10),and the gate control recurrent unit is used to predict the components.3)By employing a variable weight dynamic allocation method,the PM_(10)concentration prediction values from individual prediction models were combined to construct a hybrid prediction model for dust PM_(10)concentration based on variable weight combination.An experiment is conducted on the data set of Erlianhot,Inner Mongolia,and the experimental results show that the prediction accuracy of the hybrid prediction model proposed in this paper is better than that of the single prediction models,and the three evaluation indicators of the hybrid prediction model are all better than those of single prediction models.The coefficient of determination is improved by 2%to 20.3%,the average absolute error is reduced by 45%to 79.5%,and the root mean square error is reduced by 45.5%to 76%.It is proved that the model can effectively predict the PM_(10)concentration.

关 键 词：PM_(10)浓度预测 沙尘污染物 梯度提升回归树算法 变分模态分解算法 长短期记忆网络 门控循环单元 

分 类 号：X16[环境科学与工程—环境科学]