基于深度学习的沈阳市春节期间PM_(2.5)浓度预测研究  

Prediction of PM_(2.5) Concentration During Spring Festival in Shenyang Based on Deep Learning

作　　者：刘思洋 曹馨元 刘照[2] 李晓妍 Liu Siyang;Cao Xinyuan;Liu Zhao;Li Xiaoyan(School of Energy and Environment,Shenyang Aerospace University,Liaoning,110136;Northeast Institute of Geography and Agroecology,Chinese Academy of Sciences,Jilin,130102)

机构地区：[1]沈阳航空航天大学能源与环境学院,辽宁110136 [2]中国科学院东北地理与农业生态研究所,吉林130102

出　　处：《当代化工研究》2023年第4期86-88,共3页Modern Chemical Research

摘　　要：本文利用2016年—2022年沈阳市春节期间逐时空气质量监测数据和气象因子观测资料,基于贝叶斯参数优化,建立了GRU、LSTM深度学习模型以及LGBM、XGBOOST、RF、GBDT树集成学习模型,并且通过4种错误度量标准,将模型进行对比进而得出预测PM_(2.5)浓度的最优模型。结果表明:GRU模型的PM_(2.5)浓度预测准确度最高、训练速度最快、模型最简单,其MSE为32.160,R2为0.973,其次为LSTM模型,GBDT模型的预测效果最差。同时整体来看,深度学习模型要优于常见的树集成学习模型。In this paper, based on the hourly air quality monitoring data and meteorological factors observation data during the Spring Festival in Shenyang from 2016 to 2022, GRU, LSTM deep learning model and LGBM, XGBOOST, RF and GBDT tree integrated learning model are established based on Bayesian parameter optimization, and the optimal model for predicting PM_(2.5)concentration is obtained by comparing the models with four error metrics. The results show that GRU model has the highest accuracy of PM_(2.5)concentration prediction, the fastest training speed and the simplest model, with MSE of 32.160 and R2 of 0.973, followed by LSTM model and GBDT model with the worst prediction effect. On the whole, the deep learning model is better than the common tree ensemble learning model.

关 键 词：春节期间 PM_(2.5)浓度预测 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]