太原市COVID-19防控前后空气质量分析及预测

Air quality analysis and prediction before and after the prevention and control of COVID-19 in Taiyuan

为了探索特殊情况下太原市空气质量预测和评价方法,采用基于灰色关联度法的模糊综合评价方法对太原市疫情防控前后的空气质量进行评价,对相关联的污染物浓度变化进行分析,并以太原市AQI监测数据为基础,结合长短期记忆循环神经网络(LSTM)以及随机梯度下降算法(Adam)建立了太原市空气质量预测模型(即Adam-LSTM模型),并与LSTM模型的预测结果进行了比较。结果显示,在启动一级应急响应加强防控后,太原市的整体空气质量得到改善,个别污染物由于气象及春节等因素未降低,LSTM模型和Adam-LSTM模型预测结果的均方根误差和训练速度分别为0.203 s和12.15 s,0.183 s和10.35 s。提出的Adam优化算法能够有效提高LSTM神经网络的训练精度和收敛速度,同时具有较小的预测误差,可为环保部门制定提升空气质量相关决策提供数据支持和方法借鉴。

In order to explore the prediction and evaluation methods of air quality under special circumstances in Taiyuan,the air quality trend before and after the prevention and control of COVID-19 in Taiyuan was evaluated and the changes of related pollutant concentrations were analyzed by using the fuzzy comprehensive evaluation method based on grey correlation method.On the basis of the AQI monitoring data of Taiyuan,Long Short-Term Memory(LSTM)loop neural network and the stochastic gradient descent algorithm(Adam),the Taiyuan air quality prediction model(Adam-LSTM model)was established,and the prediction results were compared with that of LSTM model.The results show that the air quality of Taiyuan is improved after the start of the first level emergency response,and some pollutants are not reduced due to meteorological and Spring Festival factors.The root mean square error and training speed of LSTM and Adam-LSTM model are 0.203 s and 12.15 s,0.183 s and 10.35 s,respectively.It shows that the proposed Adam optimization algorithm can effectively improve the training accuracy and convergence speed of LSTM model.With relatively small prediction errors,it can provide data support and prediction methods for environmental protection departments to make related air quality decisions.