Performance of Enhanced CNN-LSTM Prediction Model for Vehicle-Mounted Air Quality Monitoring

CNN-LSTM has been established for predictive modeling in urban air pollution. However, while the method is accurate for time series forecasting, it struggles with accurately capturing spatial variability in air quality data. To address this limitation, we propose an Enhanced CNN-LSTM Prediction Model for vehicle-mounted air quality monitoring systems. This model integrates Convolutional Neural Networks (CNNs), spatial attention mechanisms, and Long Short-Term Memory (LSTM) networks to enhance the capture of both spatial and temporal dependencies in the data. Performance evaluation shows a Mean Absolute Error (MAE) of 5.2 µg/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>for PM2.5, 1.3 µg/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>for PM10, 0.1 ppm for CO, 0.1 ppm for VOCs, and 0.68°C for temperature, with Root Mean Squared Error (RMSE) values of 7.8µg/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>for PM2.5, 1.5µg/m<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup>for PM10, 0.10 ppm for CO, 0.10 ppm for VOCs, and 0.75°C for temperature. Compared to traditional models like Linear Regression and standalone LSTM networks, the Enhanced CNN-LSTM model demonstrates substantial improvements in prediction accuracy. These findings highlight the model's superior performance in delivering accurate, real-time air quality predictions, offering a robust framework for pollution management and public health interventions.