基于时空规律的PCA-LSTM-Attention空气质量预测研究  

Air Quality Prediction by PCA-LSTM-AttentionBased on Spatio-temporal Regularity

作　　者：栗治杰 贾东水 Li Zhijie;Jia Dongshui(School of Management Engineering and Business,Heibei University of Engineering,Handan 056004,China)

机构地区：[1]河北工程大学管理工程与商学院,河北邯郸056004

出　　处：《环境科学与管理》2024年第11期172-177,共6页Environmental Science and Management

摘　　要：空气质量指数(AQI)是考量空气质量好坏的综合指标,由于各地区空气受风向影响不断流动,使传统预测模型难以从时间单一角度进行建模。针对此问题提出一种利用主成分分析(PCA)降维思想,考虑不同地区时空规律的模型。通过收集目标城市和周边几个城市的所需数据,使用PCA求解所有城市的综合空气得分作为空间信息,再输入LSTM提取时间规律,最后通过注意力模块输出AQI预测。通过对沧州、唐山、廊坊、保定和天津的大气污染物和气象数据的分析,证明该算法比只考虑时间因素的LSTM模型、RNN模型和ARIMA(1,1,1)模型精度更高,可以有助于提高AQI预测精度。The air quality index(AQI)is a comprehensive indicator to consider the good or bad air quality.The constant flow of air in each region due to the wind direction makes it difficult for the traditional prediction model to model from a single perspective of time.To address this problem,a model is proposed to consider the spatial and temporal patterns of different regions using the idea of dimensionality reduction by principal component analysis(PCA).By collecting the required data from the target city and several neighboring cities,the composite air scores of all cities are solved as spatial information using PCA,and then inputted into LSTM to extract the temporal pattern,and finally the AQI prediction is outputted through the attention module.By analyzing the air pollutants and meteorological data of Cangzhou,Tangshan,Langfang,Baoding and Tianjin,it is proved that the algorithm is more accurate than the LSTM model,RNN model and ARIMA(1,1,1)model which only consider the time factor,and can help to improve the AQI prediction accuracy.

关 键 词：空气质量指数 长短期记忆网络 注意力机制 主成分分析法 

分 类 号：X823[环境科学与工程—环境工程]