Modelling Of Fusion Artificial Neural Networks For Assessment Of Air Pollution In Smart City Environment

Escalating impact of human action, industrial development and urban expansion has given rise to critical global issue: air pollution. This pressing concern has significant menace in the welfare of individuals in numerous countries. Among the various contaminants of air, Particulate Matter having a width less than 2.5µm (PM2.5) stands out as particularly menacing hazard. PM2.5 is connected with severe health issues, including lung and cardiovascular diseases. Consequently, it has become imperative to develop precise predictive models for PM2.5 concentrations, allowing for early intervention to safeguard the population from the perilous consequences of contaminated air. The fluctuation of PM2.5 levels are affected by a multitude of factors, including meteorological conditions as well as concentrations of additional contaminants within the city region. Current research work has undertaken the task of forecasting hourly PM2.5 concentrations in New Delhi, India, utilizing a deep learning (DL) approach. Specifically, we have employed a CNN-LSTM model, which incorporates space-time features through combining past information about contaminants, climatic records as well as PM2.5 concentrations from neighboring monitoring stations. Our study also entails a comparative analysis of various DL algorithms, encompassing, a hybrid CNN-LSTM, Bi-GRU, Bi-LSTM, GRU, CNN, and LSTM model. The outcomes of these experiments unequivocally demonstrate the proposed technique, the "hybrid CNN-LSTM multivariate" model, consistently outperforms all the traditional models listed, delivering superior predictive accuracy and efficiency.