This research has examined the application of machine learning and statistical methods for developing roadside particle (number/mass concentrations) prediction models that can be used for air quality management. Data collected from continuous monitoring stations including pollutants, traffic and meteorological variables were used for training the models. A hybrid feature selection method involving Genetic Algorithms and Random Forests was successfully used in selecting the most relevant predictor variables for the models from the variables selected based on their correlation with the PM\(_+\), PM\(_{2.5}\) and PNC concentrations. The study found that the hybrid feature selection can be used with both statistical and machine learning methods to produce less expensive and more efficient air quality prediction models. Among the machine learning models studied the Boosted Regression Trees (BRT), Random Forests (RF), Extreme Learning Machines (ELM) and Deep Learning Algorithms were found to be the most suitable for the predictions of roadside PM\(_+\), PM\(_{2.5}\), and PNC concentrations. The machine learning models performed better than the ADMS-road model in spatiotemporal predictions involving monitoring sites locations. Moreover, they performed much better in predicting the concentrations in street Canyons. The ANN and BRT were found to be suitable for air quality management applications involving traffic management scenarios.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:694739 |
| Date | January 2016 |
| Creators | Suleiman, Aminu |
| Publisher | University of Birmingham |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://etheses.bham.ac.uk//id/eprint/6945/ |
Page generated in 0.0109 seconds