Data-driven approach for emission monitoring and management

• Main Author: Si, Minxing
• Other Authors: Du, Ke, Li, Simon, Mohamad, Abdulmajeed
• Format: Doctoral or Postdoctoral Thesis
• Language: English
• Published: Schulich School of Engineering 2023
• Subjects: Predictive emission moniotirng; GHG monitoring; machine learning; artificial intelligence; Canada Carbon Policy; Alberta GHG Policy; Engineering > Environmental; Canada; Peace River
• Online Access: http://hdl.handle.net/1880/115864; https://doi.org/10.11575/PRISM/40755

Data-driven approach for emission monitoring and management is the process of making decisions that are informed by collecting, processing, and analyzing data. In this research, we developed and assessed predictive models using six algorithms, including linear regression, lasso regression, ridge regression, adaptive boosting, gradient boosting, and artificial neural networks (ANN), to monitor NOx emissions from point sources. The long-term evaluation showed that the moderate complexity algorithm, adaptive boosting, had the best long-term monitoring performance with a root mean square error (RMSE) of 0.48 kg/hr. The two algorithms with the high-complexity developed by gradient boosting and ANN algorithms had the worst RMSE score, 0.51 kg/hr and 0.57 kg/hr, during the long-term monitoring period. Additionally, we used machine learning methods to calibrate a low-cost fine particulate matter (PM) sensor. After calibration by gradient boosting and ANN, the variances of the PM2.5 values were not statistically significantly different from the variance of the PM2.5 values measured by the reference method. The ANN method generated the lowest RMSE of 3.91 in the test dataset with 610 samples. Moreover, we applied data-driven discovery to an oil and gas database. The use of clustering and association rules implied that: (1) the cyclic steam stimulation (CSS) recovery method was less efficient than Steam-Assisted Gravity Drainage (SAGD) recovery as schemes proceed toward maturity; (2) gas co-injection resulted in low Steam Oil Ratio (SOR)s; and (3) the Cold Lake region had higher solution gas oil ratio compared to the two other regions, including Athabasca and Peace River. Finally, data-driven approaches were used to analyze GHG emissions from in-situ oil sands operations. The weighted averages of the fuel use for the schemes using SAGD and CSS were 0.20 103m3 fuel to produce 1 m3 bitumen (0.24 103m3/m3) and 0.34 103m3 fuel to produce 1 m3 bitumen (0.34 103m3/m3), respectively. The average emission intensities (EIs) for ...