基于深度学习与时空信息集成的O_(3)预测研究

Study on O_(3)Prediction Based on Deep Learning and Spatiotemporal Information Integration

高浓度臭氧会产生很多危害,精准预测臭氧浓度可为相关部门提供有效预警。基于深度学习和时空信息集成的方法,开发了一种新的集成模型,包括深度学习模块、气象与时空信息耦合预测模块与集成模块,对杭州市滨江站点的臭氧浓度进行了预测。结果显示:该模型在24 h预测中平均绝对误差(MAE)为19.35μg·m^(-3),显著优于其他模型;该模型在不同程度的臭氧污染条件下均能较好地预测臭氧变化趋势,对于高浓度臭氧的峰值捕捉能力也最为显著;模型在不同季节均表现出较好的预测性能,在秋季表现最佳;模型的臭氧空气质量分指数(IAQI)预测准确率在24 h内表现最佳,准确率为0.81,其中在前3个h可达0.9以上,可以为臭氧污染治理提供科学支撑。

High concentrations of ozone can cause many hazards,and accurate prediction of ozone concentration can provide effective early warning for relevant departments.Based on the method of deep learning and spatiotemporal information integration,this study developed a new integrated model,including a deep learning module,a meteorological and spatiotemporal information coupling prediction module and an integration module,to predict the ozone concentration at the Binjiang station in Hangzhou.The results show that the mean absolute error(MAE)of this model in 24 h prediction is 19.35μg·m^(-3),which is significantly better than other models.The model exhibits proficient capabilities in predicting ozone trends under various levels of ozone pollution,particularly excelling in capturing peak concentrations during high ozone episodes.Moreover,the model showcases consistent performance across different seasons,with its optimal performance observed in autumn.Regarding the model's accuracy in predicting the ozone individual air quality index(IAQI),it performs notably well within 24 h,reaching an accuracy of 0.81.Specifically,its accuracy exceeds 0.9 within the initial 3 h,thereby offering substantial scientific support for ozone pollution management.