基于深度学习的大数据空气污染预报

Deep Learning based Air Pollutant Forecasting with Big Data

中文关键词:  空气污染预报  深度学习  深度信念网络  大数据

英文关键词:air pollutant forecasting  deep learning  deep brief network  big data

中文摘要:

      为了更好地反映环境污染变化趋势,为环境管理决策提供及时、全面的环境质量信息,预防严重污染事件发生,开展城市空气质量预报研究是十分必要的.本文针对环境大数据时代下的城市空气质量预报,提出了一种基于深度学习的新方法.该方法通过模拟人类大脑的神经连接结构,将数据在原空间的特征表示转换到具有语义特征的新特征空间,自动地学习得到层次化的特征表示,从而提高预报性能.得益于这种方式,新方法与传统方法相比,不仅可以利用空气质量监测、气象监测及预报等环境大数据,充分考虑污染物的时空变化、空间分布,得到语义性的污染物变化规律,还可以基于其他空气污染预测方法的结果(如数值预报模式),自动分析其适用范围、优势劣势.因此,新方法通过模拟人脑思考过程实现更充分的大数据集成,一定程度上克服了现有方法的缺陷,应用上更加具有灵活性和可操作性.最后,通过实验证明新方法可以提高空气污染预报性能.

英文摘要:

      It is necessary to do research about urban air quality forecasting to better refl ect the changing trend of the air pollution and provide prompt and complete environment quality information for environment management decision, as well as to avoid serious air pollution accident. For the urban air quality forecasting in the era of environmental big data, this paper proposes a novel method based on deep learning. Via simulating neural connecting structure of human brain, the deep learning method transforms the feature representation of data in the original space to a new feature space with semantic feature, and obtains hierarchical feature representation automatically to improve the performance of forecasting. Due to the merits of the deep learning, compared with traditional methods, the deep learning based model can not only utilize the environmental big data, including the air quality monitoring, weather monitoring and forecasting, and consider the spatiotemporal change and spatial distribution of air pollutant sufficiently to get the semantic change regulation of air pollutant, but also analyze the scope of its application, advantages and disadvantages based on results of other air quality forecasting methods (such as, numerical forecasting model). Therefore, the deep learning based method realizes the comprehensive integration of big data via simulating the thinking progress of human brain. The novel method is of fl exibility and feasibility for application, and overcomes the weak of the existing forecasting methods. Finally, the numerical test demonstrates that the novel method can improve the performance of air pollutant forecasting.