哈尔滨市大气污染物变化规律分析与空气质量预测

【作者】 陈明；

【导师】 苗守雷；

【作者基本信息】 黑龙江大学 ， 资源与环境， 2024， 硕士

【摘要】 本文利用2020-2023年哈尔滨市大气污染物监测数据,深入分析了六种主要空气污染物(SO₂、NO₂、O₃、CO、PM₁₀和PM_2.5)的时空变化特征;并利用空气质量指数（AQI）评估系统对大气环境的现状进行了直观的说明。使用SPSS统计分析工具,对不同气象要素与大气污染物之间的相关性进行了分析。采用聚类分析和潜在源贡献法两种研究方法探索大气污染的传播路径以及污染潜在源区。最后建立粒子群优化算法（PSO）和长短期记忆神经网络（LSTM）相结合的神经网络预测模型对哈尔滨市首要污染物的浓度进行预测。从时间变化的污染特征分析来看,各种污染物均展现出鲜明的季节变化特征。在空间分布上,PM₁₀、PM_2.5、SO₂和NO₂的浓度在中间区域相对较高,而O₃则呈现出北高南低的态势,CO的浓度则是南部高于中间区域。在AQI评价体系中,四年中重污染及严重污染天数占比分别为5.7%、1.4%、2.5%、2.7%;哈尔滨市的首要污染物是PM_2.5、PM₁₀和O₃。在空气污染路径研究中,对500米处的气流进行了聚类分析,发现气流来源在不同季节存在显著差异。根据潜在源贡献因子和浓度权重法分析,哈尔滨市O₃污染的潜在源区为山东渤海、辽宁、吉林省。PM₁₀及PM_2.5的污染除本地污染外,主要来源于西南方向的吉林和辽宁省。应用建立的粒子群优化算法和长短期记忆网络相结合（PSO-LSTM）的深度学习预测模型,分别对哈尔滨市未来7小时PM_2.5、PM₁₀和O₃浓度进行了预测。结果表明,PSO-LSTM模型预测效果优于RF神经网络模型和单个的LSTM神经网络模型,预测浓度与实际浓度之间的平均相对误差为6.86%、12.62%、11.92%。为空气质量预测提供了一种更准确的方法。更多还原

【Abstract】 In this paper,the spatial and temporal variation characteristics of six major air pollutants(SO₂,NO₂,O₃,CO,PM₁₀,and PM_2.5)were analyzed in depth using the monitoring data of atmospheric pollutants in Harbin City from 2020-2023;and the current status of the atmospheric environment was visualized by using the Air Quality Index（AQI）assessment system.The correlations between different meteorological elements and atmospheric pollutants were analyzed using the SPSS statistical analysis tool.Two research methods,cluster analysis and potential source contribution method,were used to explore the transmission pathways of air pollution and the potential source areas of pollution.Finally,a neural network prediction model combining Particle Swarm Optimization（PSO）algorithm and Long Short-Term Memory（LSTM）neural network is established to predict the concentration of the top pollutants in Harbin City.The time-varying pollution characterization shows that all pollutants exhibit distinct seasonal variations.In terms of spatial distribution,the concentrations of PM_10,PM_2.5,SO₂,and NO₂ were relatively higher in the middle region,while O₃ showed a trend of higher in the north and lower in the south,and the concentration of CO was higher in the south than in the middle region.In the AQI evaluation system,The percentage of heavily and severely polluted days in the four years was 5.7%,1.4%,2.5%and 2.7%respectively.and the top pollutants in Harbin were PM_2.5,PM₁₀,and O₃.In the air pollution pathway study,the airflow at 500 meters was clustered and analyzed,and significant differences in airflow sources were found in different seasons.According to the analysis of potential source contribution factor（PSCF）and concentration weighting method（CWM）,the potential source areas of O₃ pollution in Harbin are Bohai Sea in Shandong Province,Liaoning Province,and Jilin Province.PM₁₀and PM_2.5pollution mainly originates from Jilin and Liaoning provinces in the southwest direction,in addition to local pollution.The established particle swarm optimization algorithm and deep learning prediction model combining long and short-term memory networks（PSO-LSTM）were applied to predict the PM_2.5,PM_10,and O₃concentrations in Harbin for the next 7 hours,respectively.The results showed that its prediction is better than that of the RF neural network model and the individual LSTM neural network model.the average relative errors between the predicted and actual concentrations were 6.86%,12.62%,and 11.92%,which provides a more accurate method for air quality prediction.更多还原