Air pollution is a global problem with severe impacts on health and well-being. Air pollution is responsible for seven million deaths per year according to the World Health Organization (WHO). To address this situation, there is a need for an Air Quality Management System (AQMS) that can provide air quality information at high temporal (within a few minutes) and spatial resolution (section of a street) to the public and relevant agencies. It should be situation-aware and help autonomous decision making for air quality mitigations. There is currently no AQMS satisfying all these needs. This project is aimed to address these needs through a software solution that allows air quality (AQ) data processing from the Cloud to multiple levels of network devices towards the edge of the network (e.g., switches, routers, and embedded servers) in an autonomous and intelligent manner. To this end, AirQMan proposes a novel data processing model that autonomously determines the optimal AQ data processing flow and the right trained machine learning (ML) models to maximize the accuracy of a prediction related to an AQ request. AirQMan also features a data processing framework that determines the optimal deployment plan over the aforementioned computing platforms for efficient computation of ML models while satisfying requirements on accurate and low-latency AQ prediction and information provisioning.
In the first period of the project, we started doing some processing and prediction experiments on the air quality data provided by our partner NILU. This will serve as a basis for exploring distributed learning challenges.
Existing air quality (AQ) monitoring and management (AQMS) methods and evolving modelling practices across Norwegian and European cities have achieved significant improvements of AQ but further progress is needed due to some quality-driven requirements, such as low-latency AQ prediction. This can only be achieved by intelligent data processing at multiple levels of granularity. To this end, affordable, effective and intelligent tools are needed that utilize the current advances in digitization of all spheres of society, providing radical innovation of air quality management. The AirQMan project promises autonomous computational methods and techniques that can be used to develop such solutions, and has the potential for opening up a new era in air quality management. Our strong belief is that such a system can be realized across the Edge-Fog-Cloud continuum, extending data processing and computational intelligence from the Cloud to multiple levels of Fog nodes towards the edge of the network.
The project will develop AirQDM – a novel data processing design model that will autonomously determine the optimal data fusion processing flow, the right data sources, and the right trained deep learning (DL) model for maximizing the accuracy of a prediction related to an AQ request. A second innovation of the project, AirQWare will determine (predict) the optimal distributed deployment for an efficient computation of the DL model while satisfying requirements on accuracy and latency, and adapt the deployment of the DL model during runtime as necessary to maintain accuracy and latency requirements.
By applying the AirQMan approach, the new generation of AQMS will provide: i) low-latency data validation and fusion to increase the accuracy of air quality evaluation, and to support intelligent services, respectively, and ii) cognitive decision making with various degrees of autonomy enabling low-latency actuations of AQ mitigations.