基于机器学习的大气NO_(2)浓度预测模型

Machine Learning-based Prediction Model for Atmospheric NO_(2)Concentration

传统的NO_(2)监测存在响应时间滞后等问题,准确预测大气NO_(2)浓度对于环保政策制定和空气质量改善至关重要。大气NO_(2)水平与地区的气象条件、工业污染排放、社会经济发展情况等多个因素相关联,因此NO_(2)污染具有显著的区域差异。近年,机器学习被广泛应用于环境质量要素预测,其中极端梯度提升树(XGBoost)算法在分析、挖掘数据关系上具有优势。本研究搜集了2011—2022年大连市11个区县的大气NO_(2)浓度与气象、工业排放、社会经济因素的年度数据,通过时间滑动策略,结合XGBoost算法构建了空间异质的未来时间NO_(2)预测模型。模型对大连市各区域2021年与2022年NO_(2)浓度预测结果的决定系数(R^(2))达到0.611,具有良好的预测性能与泛化能力。使用沙普利加和解释(SHAP)对关注的多个因素进行分析,结果表明,污染排放氨氮、社会消费品零售额、污染排放氮氧化物与NO_(2)浓度呈现正相关。

Traditional NO_(2)monitoring technique faces challenges such as delay in response time.It is crucial to predict the atmospheric NO_(2)levels for informing environmental policy decisions and enhancing air quality.The atmospheric NO_(2)levels can be affected by various factors including regional meteorological conditions,industrial pollution emissions,and socio-economic development,leading to notable regional disparities in the NO_(2)pollution.In recent years,machine learning techniques have been generally utilized for predicting pollutant levels,with the XGBoost(eXtreme Gradient Boosting)algorithm standing out for its excellent ability to analyze data relationships.This study gathered annual data on atmospheric NO_(2)levels,meteorological conditions,industrial emissions,and socio-economic factors of 11 districts in Dalian City from 2011 to 2022.By employing a time-sliding strategy in conjunction with the XGBoost algorithm,a spatially heterogeneous model was developed to predict the NO_(2)concentrations.The coefficient of determination(R^(2))of the model for the prediction results reached 0.611,which shows that the model demonstrated has good prediction performance and generalization ability.Multiple factors of concern were analyzed by using SHAP(SHapley Additive exPlanations),and the results revealed that pollution emission of ammonia nitrogen,retail sales of social consumer goods,and pollution emission of nitrogen oxides were positively associated with the NO_(2)concentration.