Abstract:
Air quality control requires real-time monitoring of pollutant concentration distributions in large urban areas. Estimation models are used for the soft-calibration of low-cost multisensor data to improve precision of pollutant concentration measurements. This study introduces an integrated genetic programming dynamic neural network model for more accurate estimation of carbon monoxide and nitrogen dioxide pollutant concentrations from the multisensor measurement data. This model combines a genetic programming-based estimation model with a neural estimator model and improves estimation performances. In this structure, a genetic programming-based polynomial model works as a former estimator and it feeds the neural estimator model via a short-term former estimation memory. Then, the neural model utilizes this former estimation memory in order to enhance pollutant concentration estimations. This integration approach benefits from the correlation enrichment strategy that is performed by the former model. The proposed two-stage training procedure facilitates the training of the integrated models. In experimental study, the standalone genetic programming model, artificial neural network model, and the proposed integrated model are implemented to estimate carbon monoxide and nitrogen dioxide pollutant concentrations from the experimental multisensor air quality data. Results demonstrate that the proposed integrated model can decrease mean relative error about 10% compared to the standalone artificial neural network and about 28% compared to the standalone genetic programming estimation models. Authors suggested that the integrated estimation model can be used for more accurate soft-calibration of multisensor electronic noses in a wide-area air-quality monitoring application.