A Machine learning approach to surrogate development for Canadian power system toward decarbonization
| dc.contributor.author | Jahangiri, Zahra | |
| dc.contributor.supervisor | McPherson, Madeleine | |
| dc.date.accessioned | 2025-09-04T20:35:53Z | |
| dc.date.available | 2025-09-04T20:35:53Z | |
| dc.date.issued | 2025 | |
| dc.degree.department | Department of Civil Engineering | |
| dc.degree.level | Doctor of Philosophy PhD | |
| dc.description.abstract | As Canada works toward a net-zero emissions economy by 2050, understanding optimal strategies for power sector expansion and decarbonization is crucial. To address this challenge, this thesis uses machine learning, specifically neural networks, to conduct a detailed sensitivity analysis, uncertainty analysis and provincial analysis. We developed a supervised learning surrogate model for a capacity expansion model, reducing computation costs by five orders of magnitude. Using this model, we perform sensitivity analysis to evaluate how changes in input variables, such as generation technology capital costs, electricity demand, and carbon taxation, impact model outputs. Additionally, we perform an uncertainty analysis to explore the behavior of the model’s outputs in response to variability, uncertainty, and potential fluctuations in these inputs. This approach allows for a more advanced exploration of the design options for Canadian national and provincial power systems. This model reduces computational time from 11–72 hours to milliseconds with minimal resource requirements. The computational efficiency enables integration into various platforms and tools for decision-making. It’s essential because it makes the model accessible to users who may not have technical expertise, such as stakeholders and decision-makers. By reducing the need for extensive technical resources, these users can leverage the model's outputs to inform real-time decisions without relying on advanced computing power. The study in chapter 4, uses unsupervised machine learning and statistical techniques to identify key factors influencing system outcomes. These include the increasing importance of gas combined cycles in a low-carbon system and the strong potential of wind energy in Canada's decarbonization. Our methodology identifies key patterns in power system outcomes. For example, it uncovers critical correlations like that between variable renewable energy capacity factors and transmission expansion. The results in chapter 5, underscore the importance of flexible grid systems and offering a province-specific roadmap. This thesis introduces the use of machine learning for large-scale energy system planning. It contributes by developing analytical frameworks for model usage and offering a detailed discussion of the results. These insights provide a foundation for strategic planning and policy formulation, particularly in supporting Canada’s transition to a sustainable energy future. | |
| dc.description.scholarlevel | Graduate | |
| dc.identifier.uri | https://hdl.handle.net/1828/22720 | |
| dc.language | English | eng |
| dc.language.iso | en | |
| dc.rights | Available to the World Wide Web | |
| dc.subject | Machine Learning | |
| dc.subject | Deep learning | |
| dc.subject | Energy decarbonization | |
| dc.subject | Energy planning | |
| dc.subject | Uncertainty Analysis | |
| dc.subject | Sensitivity Analysis | |
| dc.subject | Provincial Analysis | |
| dc.subject | K-Means clustering | |
| dc.subject | Power systems | |
| dc.subject | Residual neural networks | |
| dc.title | A Machine learning approach to surrogate development for Canadian power system toward decarbonization | |
| dc.type | Thesis |