Living in the age of data and the new era of digitalization of cities have created a large volume of datasets and data flows associated with the urban environments. It is significantly vital to capture and analyze the data from various resources in smart cities. For instance, the real-time air pollution data are remarkably important in controlling air pollution for urban sustainability and protecting humans against the air pollution damages. However, in reality, the average construction investment and maintenance costs in the air pollution stations are too high. This paper intends to investigate whether and how we can measure air pollution using cost effective means and without using the expensive pollution sensors and facilities. In order to realize such a goal, a predictive model for particulate matter prediction was developed. The proposed model consists of multiple components to integrate heterogeneous multiple sources of urban data and predict the particulate matter based on transfer learning perspective in which neural network and regression was leveraged as the core of the prediction. The results of the particulate matter prediction exposed that while these data sources are capable of proper prediction of the particulate matter, they can also yield better results over the models, which were based only on the features of the air pollution sensors. This work provides an opportunity for evaluation of the model with the urban data from the city of Aarhus, in Denmark, and comparison of the model performance against various specified baselines. The superiority of the model over the baselines shows the practicality of the model.

处于数据时代和城市数字化新时代，已经创建了大量与城市环境相关的数据集和数据流。从智能城市的各种资源中捕获和分析数据至关重要。例如，实时空气污染数据对于控制空气污染以实现城市可持续发展以及保护人类免受空气污染损害非常重要。但是，实际上，空气污染站的平均建设投资和维护成本太高。本文旨在研究是否以及如何使用成本有效的方法而不使用昂贵的污染传感器和设施来测量空气污染。为了实现这一目标，开发了用于颗粒物预测的预测模型。所提出的模型由多个组件组成，用于集成异质的多个城市数据源，并基于传递学习的观点来预测颗粒物，其中以神经网络和回归作为预测的核心。颗粒物预测的结果表明，尽管这些数据源能够正确预测颗粒物，但是它们也可以比仅基于空气污染传感器特征的模型产生更好的结果。这项工作为利用丹麦奥尔胡斯市的城市数据对模型进行评估以及将模型性能与各种指定基准进行比较提供了机会。该模型相对于基线的优越性表明了该模型的实用性。