Air quality forecasting using artificial neural networks with real time dynamic error correction

Proposed by a group of researchers from Belgium and India in 2020, a novel method of air quality forecasting using Artificial Neural Networks (ANN) was developed. Given the grave effects of air pollution on human health, providing highly accurate pollution forecasts is necessary for heavily polluted areas. Powered by machine learning, Artificial Neural Networks can forecast pollution for up to five days. The ANN focuses on forecasting the pollution concentrations of PM10, PM2.5, NO2, and O3. Given the demographics of the research team that trained the ANN, the model was used to forecast pollution in highly polluted locations, with 32 different areas in Delhi. The ANN was trained using hourly pollution concentration data from 2018, as well as corresponding meteorological parameters. Air quality forecasts can be rendered in real-time, and adjust dynamically based on the latest accuracy of its predictions using Real Time Correction. The model was validated for predictions in both 2018 and 2019, but given the ever-changing dynamics of air pollution within some of these densely populated cities, the model may need additional data inputs surrounding incidents such as wildfires or days which call for fireworks celebrations.


Artificial Intelligence, Climate Change, Data, Internet of Things, Pollution, Regulation, Visual Technologies