基于多步时间序列的空气质量预测研究  

Research on Air Quality Prediction Based onMultistep Time Series

作　　者：胡予昕 陆文浩 徐子俊 李媛媛[1] Hu Yu-xin;Lu Wen-hao;Xu Zi-jun;Li Yuan-yuan(School of Artificial Intelligence,JianghanUniversity,Wuhan 430056,Hubei Province,China)

机构地区：[1]江汉大学人工智能学院,湖北武汉430056

出　　处：《科学与信息化》2024年第15期48-50,共3页Technology and Information

基　　金：江汉大学在线课程建设项目,项目名称:数学建模与仿真,项目编号:2022-8。

摘　　要：研究空气污染水平,更精准的预测PM2.5浓度和AQI指数对于解析污染影响因素和有效制订控制策略具有重要意义。本文基于2015-2023年同一地区的污染物浓度和气象数据,根据PM2.5浓度和AQI非线性、时序性的特征,构建了ARIMA和LSTM多步预测模型,对PM2.5浓度和AQI等级进行预测。结果显示,对于PM2.5浓度的真实数据,基于3步预测的ARIMA模型RMSE值最小,更适合PM2.5浓度的预测;而在AQI的真实数据集上,LTSM模型较ARIMA模型准确性更高。Studyingthe level of air pollution and more accurate prediction of PM2.5 concentration and AQI index are of great significance for analyzing the influencing factors of pollution and formulating control strategies effectively.Based on pollutant concentrations and meteorological data inthe same region from 2015 to 2023,and according to the nonlinear and sequential characteristics of PM2.5 concentration and AQI,the multi-step prediction models of ARIMA and LSTM are constructed to predict the PM2.5 concentration and AQI grade.The results show that for the real data of PM2.5 concentration,the ARIMA model based on three-step prediction has the smallest RMSE value and is more suitable for the prediction of PM2.5 concentration,while in the real data set of AQI,LTSM model is more accurate than ARIMA model.

关 键 词：ARIMA模型 LSTM模型 空气质量预测 时间序列 

分 类 号：X51[环境科学与工程—环境工程]