报告主题： 机器学习和数据同化在空气质量预测中的应用
报告人：林海翔 教授 （荷兰代尔夫特理工大学）
报告时间：2018年11月22日（周四）15:30
报告地点：校本部G508
邀请人：白延琴教授
主办部门：理学院数学系
报告摘要：The advance of sensor technology and reduced cost of monitoring systems lead to a vast increase of meteorological and air quality observation data. Numerical models of chemical transport have been developed and in operational use to simulate the complex processes involved in the formation and transport of air pollutants. Although these models can predict the spatiotemporal variability of the concentrations of various chemical species, the accuracy of these models is often limited due to the coarse resolution of the used numerical grids (constraint by the computer capacity). Therefore, data assimilation methods have been applied to use the continuously measured data to improve the forecast. This practice of using (big) data to improve the model parameters and initial values had already been done in the past two decades. Now, machine learning techniques provide new possibilities to use observation data for improving the air quality forecast. We will report our experience with the recurrent neural network, LSTM, for visibility prediction at an airport. A second example considers the prediction of PM2.5 and PM10 concentrations using a numerical chemical transport model. It is known that the concentration of pollutants depends on local emissions (by human activities), however, estimation of local emissions are usually only available after three to five years. The lack of current emission data is one of the main reasons that current numerical chemical transport models cannot produce accurate forecasts. Using machine learning to generate local emissions based on real-time observations is a promising approach which can be combined with data assimilation to greatly improve the accuracy of air quality forecast
欢迎教师、学生参加 ！
