Accurate prediction of urban air quality is vital for protecting public health and the environment. The current challenges in air quality prediction models include noisy data, missing data, interactions of pollutants, temporal and spatial variations, external variables, lack of generalization, and real-time prediction. To overcome these challenges, a hybrid bio-inspired model for predicting urban air pollution using deep learning (AQP-SAPINN-HMRFO) is proposed. The input data is obtained from the Global Urban Air Quality Index dataset. The data is preprocessed with Implicit Bulk Surface Filtering (IBSF) to normalize the data and treat missing values to provide high-quality input data. The Quadratic-Phase Wave Packet Transform (QPWPT) is used to extract relevant features such as concentrations of pollutants, interactions between pollutants, and past concentrations of pollutants. Air quality forecasting is carried out by using a Self-Adaptive Physics-Informed Neural Network (SAPINN) that predicts concentrations of five major air pollutants like Particulate Matter PM₂. ₅, Carbon Monoxide (CO), Nitrogen Dioxide (NO₂), Ozone (O₃), and Sulfur Dioxide (SO₂), along with meteorological factors like temperature, wind speed, and humidity at various locations. The Hierarchical Manta Ray Foraging Optimization (HMRFO) technique is used to optimize the weight parameters of SAPINN to improve the accuracy of the model. The AQP-SAPINN-HMRFO model is a combination of SAPINN and HMRFO techniques. This model is capable of handling the interactions and gaps of the pollutants effectively. The proposed method is implemented in Python and the experiments demonstrate that AQP-SAPINN-HMRFO achieves 99% accuracy. This is a significant improvement over existing approaches and indicates its potential for real-time applications in urban air quality monitoring, environmental management, and strategic urban planning.
Chaudhary et al. (Thu,) studied this question.