Correlated Generator Failures and Power System Reliability

This thesis contributes new knowledge toward understanding the relationship between capacity procurement and power system reliability through rigorous analysis of generator-level availability data. In Chapter 2 I analyze four years of data (2012-2015) from the Generating Availability Data System (GADS) database maintained by the North American Electric Reliability Corporation
(NERC) to evaluate key assumptions made by power system planners when determining capacity requirements. Using block subsampling and binomial modeling, I demonstrate that large unavailable capacity events have occurred with much greater frequency than should be expected if current-practice assumptions hold.
In Chapter 3 I propose a nonhomogeneous Markov model to explain the observed correlated failures. I use logistic regression to fit a simple model specification that allows generator transition probabilities to depend on ambient temperatures and system load. I fit the model using 23 years of GADS data for the PJM Interconnection (PJM), the largest system operator by generation capacity in North America. Temperature and load are each statistically significant for two-thirds of generators. Temperature dependencies are observed in all generator types, but are most pronounced for diesel and natural gas generators at low temperatures and nuclear generators at high temperatures. The nonhomogeneous Markov model predicts system-level unavailable capacity substantially better than
the homogeneous Markov model used currently by industry.
In Chapter 4, joint work with Luke Lavin, I quantify the reliability risks implied by temperature dependence in PJM’s generator fleet. We modify an open-source resource adequacy modeling tool to allow generator availability to depend on temperature. We then parameterize the tool for PJM’s system using temperature-dependent forced outage rates developed in Chapter 3. We find that temperature dependence substantially increases capacity requirements to achieve the target level of reliability, though PJM procures still more than our model finds is required. Given the seasonality in temperatures and loads, we also demonstrate that average annual capacity requirements could be significantly reduced were PJM to set separate monthly targets, rather than a single annual target. Finally, we explore the resource adequacy implications of various future generator resource and climate change scenarios for PJM.