A Neural Attention Model for Urban Air Quality Inference: Learning the Weights of Monitoring Stations

Abstract

Urban air pollution has attracted much attention these years for its adverse impacts on human health. While monitoring stations have been established to collect pollutant statistics, the number of stations is very limited due to the high cost. Thus, inferring fine-grained urban air quality information is becoming an essential issue for both government and people. In this paper, we propose a generic neural approach, named ADAIN, for urban air quality inference. We leverage both the information from monitoring stations and urban data that are closely related to air quality, including POIs, road networks and meteorology. ADAIN combines feedforward and recurrent neural networks for modeling static and sequential features as well as capturing deep feature interactions effectively. A novel attempt of ADAIN is an attention-based pooling layer that automatically learns the weights of features from different monitoring stations, to boost the performance. We conduct experiments on a real-world air quality dataset and our approach achieves the highest performance compared with various state-of-the-art solutions.

Accurate and Scalable Gaussian Processes for Fine-Grained Air Quality Inference

“…We use fewer features due to public data availability and focus on the regression task instead of classification. Recent state-of-theart work (Cheng et al 2018) proposes a neural attentionbased approach to incorporate time-invariant and time-series features together. They also learn the effect of individual train stations on a test location via an attention net.…”

“…We now describe the further details and preprocessing for each of the datasets. Beijing dataset We use the hourly PM 2.5 data from 36 stations in Beijing and meteorological data (temperature, humidity, pressure, wind speed, wind direction and weather) from the stations in the same district (Cheng et al 2018;Zheng et al 2015). Among these features, wind direction and weather are categorical and others are continuous features.…”

“…We observe a large amount of missing data in different stations at different time intervals. To enable the comparison with state-of-the-art neural baseline (Cheng et al 2018), we chose a particular month (March 2015) having the minimum amount of missing data. We carry out a dataset preprocessing step to handle further issues with the dataset, such as missing values and anomalies.…”

“…Existing techniques to generate AQ maps rely on interpolation approaches like Kriging, Spline, and Trend. Some recent approaches have taken a data-driven approach using deep learning to generate AQ maps (Cheng et al 2018;Xu and Zhu 2016;Zheng, Liu, and Hsieh 2013). These methods: i) do not quantify uncertainty which may help policymakers make informed decisions; and ii) do not incorporate domain knowledge in modeling.…”

“…We use the hourly AQ and meteorological data from Beijing (Cheng et al 2018) and London over a month to evaluate our approaches. We compare our approach against: i) conventional spatial interpolation baselines such as Inverse Distance Weighting, K-Nearest Neighbors; ii) standard machine learning models such as XGBoost and Random For-est; and iii) a state-of-the-art neural attention based model ADAIN (Cheng et al 2018). Our approach outperforms the baselines for both datasets in a cross-validation setting.…”