Genetic programming and Neural Network for Model air pollution monitoring application

This study focuses on real-time monitoring of pollutant concentrations in large urban areas for effective air quality control. Using estimation models to enhance the precision of low-cost multisensor data through soft-calibration, the research introduces an integrated genetic programming dynamic neural network model. Specifically designed for more accurate estimation of carbon monoxide and nitrogen dioxide pollutant concentrations from multisensor data, the model combines a genetic programming-based polynomial former estimator with a neural estimator. The former estimator utilizes a short-term memory to feed the neural model, improving pollutant concentration estimations. The integration approach benefits from a correlation enrichment strategy, and a two-stage training procedure is proposed for model training. Experimental results show that the integrated model reduces mean relative error by approximately 10% compared to the standalone artificial neural network and about 28% compared to standalone genetic programming models. The authors suggest the integrated model’s potential for accurate soft-calibration of multisensor electronic noses in broad-scale air-quality monitoring applications.

Paper Url: https://link.springer.com/article/10.1007/s00521-022-07129-0