摘要
城市计算的数据具有鲜明的时空特性,即时间维度相似性与空间维度相近性的耦合关系,因此,对时空数据的分析和处理,已成为城市计算中亟需解决的热点问题。面向城市污染中的空气质量问题,提出时空卷积残差网络(spatio-temporal convolution residual network,ST-ResNet),通过分析空气质量指数(air quality index,AQI),实现预测和预警。时空卷积残差网络子组件是由以卷积层为基础的单元通过全等映射残差连接构成,将AQI数据通过空间转换组件转换成AQI像素图,利用卷积运算捕获其空间特性;而时间趋势性、周期性和时间接近度等属性分别被三个子组件捕获,将三者的输出加权连接得到时空卷积残差网络并输出AQI的预测结果。最后,将ST-ResNet网络与经典的长短时记忆网络(long short-term memory,LSTM)进行对比,结果表明ST-ResNet网络在准确率上比LSTM网络提高了7%,有望对城市环境监测预测和精细化管理提供理论依据和技术支撑。
The data of urban computing have distinct spatio-temporal characteristics,that is,the coupling relationship between the similarity of time dimension and the similarity of space dimension. Therefore,the analysis and processing of spatio-temporal data has become a hot topic that needs to be solved in urban computing. Aiming at the problem of air quality in urban pollution,the spatio-temporal convolution residual network(ST-ResNet) is proposed to realize prediction and early warning by analyzing air quality index(AQI). Sub-components in ST-ResNet are composed of unitCCFs based on convolution layer connected by congruent mapping residuals. AQI data are converted into AQI pixel maps by spatial conversion components,and their spatial characteristics are captured by convolution operation. Meanwhile,the time tendency,periodicity and time proximity are captured by three sub-components respectively,and their outputs are weighted connected to obtain the ST-ResNet and output the prediction results of AQI.Finally,by comparing the ST-ResNet with the classical long short-term memory network(LSTM),the results show that the accuracy of ST-ResNet is 7% higher than that of LSTM,which is expected to provide theoretical basis and technical support for urban environmental monitoring and prediction and fine management.
作者
李栋
张蕾
郭茂祖
刘银龙
LI Dong;ZHANG Lei;GUO Mao-zu;LIU Yin-long(School of Electrical and Information Engineering,Beijing University of Civil Engineering and Architecture,Beijing 100044,China;Institute of Information Engineering,Chinese Academy of Sciences,Beijing 100093,China)
出处
《计算机技术与发展》
2020年第6期124-129,共6页
Computer Technology and Development
基金
住房城乡建设部科学技术计划项目(2017-R2-018)
国家自然科学基金面上项目(61871020)
北京市教委科技计划重点项目(KZ201810016019)
北京市属高校高水平创新团队建设计划项目(IDHT20190506)
北京建筑大学市属高校基本科研业务费专项资金资助(X18013,X18197,X18203)。
关键词
城市计算
时空数据
空气质量指数
卷积神经网络
残差网络
urban computing
spatio-temporal data
air quality index
convolution neural network
residual network
