基于复合LSTM模型的PM2.5浓度预测

在对淄博市19个空气质量监测站点监测数据进行分析后，提出了一种基于机器学习的复合模型——灰色关联度分析(GRA)-改进的完备总体经验模态分解(ICEEMD)-长短期记忆网络(LSTM)模型。通过分析淄博市2019年大气污染物和气象数据，选用LSTM模型预测PM_2.5浓度。由于传统单一模块机器学习模型具有训练时间较长和预测精度较低的问题，提出了复合LSTM模型。该模型由3部分组成：GRA,用于PM_2.5浓度影响因素变量筛选；ICEEMD,用于PM_2.5分解、分量筛选和原始大气污染物及气象数据处理；LSTM,用于PM_2.5浓度预测。预测结果表明：淄博市中部丘陵地带PM_2.5浓度高于南部山区和北部平原，东部高于西部；淄博市逐月PM_2.5浓度呈“U”形分布，1月最高，8月最低；淄博市PM_2.5浓度受PM₁₀和CO影响较大，受湿度和温度影响较小。对比单一LSTM模型和GRA-LSTM模型，GRA-ICEEMD-LSTM模型...

PM2.5 Concentration Prediction Based on the Composite LSTM Model

Based on the data analysis of 19 air quality monitoring stations in Zibo City,a composite model based on machine learning,GRA-ICEEMD-LSTM model,was proposed.By analyzing the air pollutants and meteorological data of Zibo City in 2019,LSTM model was used to predict PM_2.5 concentration.Because the traditional single module machine learning model has the problems of long training time and low prediction accuracy,the composite LSTM model was proposed.The model consists of three parts:GRA variable screening and ICEEMD component screening;ICEEMD processing raw air pollutants and meteorological data;LSTM predicting PM_2.5 concentration.The results showed that the PM_2.5 concentration in the central hilly area of Zibo is higher than that in the southern mountainous area and the northern plain,and higher in the eastern area than that in the western area.The monthly PM_2.5 concentration shows a U-shaped distribution,with the highest in January and the lowest in August.The PM_2.5 concentration is significantly affected by PM₁₀ and CO,but less affected by humidity and temperature.Compared with the single LSTM model and the GRA-LSTM model,the GRA-ICEEMD-LSTM model can effectively improve the computing efficiency,reduce the training time and improve the prediction accuracy,which shows a good performance on PM_2.5 prediction.