Data-driven stochastic reliability assessment of the US electricity grid under large penetration of variable renewable energy resources

The impacts of climate change will exacerbate humanitarian crises at a global level, and there is an urgent need to eliminate greenhouse gas (GHG) emissions from the power system infrastructure. While deploying clean energy resources in the current grid will help in decarbonization, variable and uncertain availability of solar and wind resources will introduce additional challenges in the operation of a grid. Thus to achieve a reliable clean electricity transition it is important to think about how renewable energy resources can be increased in the grid while minimizing potential challenges. This dissertation begins by examining the reliability contribution of incorporation of large solar photovoltaic (PV), onshore wind, and offshore wind generators for the case of New York

Independent System Operator (NYISO) using a method called Effective Load Carrying Capability (ELCC). ELCC quantities the reliability benefit of adding generators with certain nameplate capacity on top of the existing base fleet of generators. We define five different future scenarios to account for potential pathways in energy transition of the base fleet through 2030. In these future scenarios, we then add the generator of interest to assess its contribution to the grid and repeat the process for multiple nameplate capacities for offshore wind, onshore wind, and solar power across the entire footprint of New York. We conclude that, from a reliability perspective choosing offshore wind generators irrespective of size (capacity) is more worthwhile as compared to onshore wind farms for serving high demand periods as median capacity contributions (ELCC values) of offshore is 20 times greater than solar generators. Furthermore, analyses using our scenarios indicate that addition of solar generators in a base fleet with abundance of onshore and offshore wind generators contributes towards increasing system reliability and vice-versa. Thus, the diversification of future base fleet is necessary to meet demand shortfalls effectively.

Next, we reconstruct proxies of temperature-driven hourly electricity demand data to investigate the response of electricity demand to the variability of temperature over multiple decades to understand the change in peak load. Investments in power system capacity expansion projects are based on understanding of accurate, region-specific peak demand requirements. Balancing Authorities of the US report hourly demand records only from 2015. This constrains the scope of analysis to understand change in demand only to five or six years. Electricity use is strongly influenced by temperature and as the grid is designed to handle maximum load days, which tend to be the hottest days in many areas, the increasing intensity of extreme heat days will require additional investments in peak generation capacity, transmission, or storage. Along with changing demand, the scope to analyze the generating capacity gap after adding variable solar and wind resources is also limited, as demand data corresponding to hourly solar radiation and wind speed records are unavailable. Thus, there is a need to reconstruct demand data based on observed historical temperature, rather than forecast demand based on simulated future temperature records from climate change models.

We attempt to fit various advanced machine learning models to understand the best choice for reconstruction based on performance on the validation set. We conclude that a "one-model fits all" approach as suggested in existing literature performs poorly. We also find that within the largest balancing authorities, ranked in the order of size and maximum demand consumption, Tennessee Valley Authority, Midcontinent Independent System Operator, and Electricity Reliability Council of Texas are most sensitive to temperature changes with the coefficient of variation of 20 largest demand hours (representative of peak demand) ranging between 15 - 19%.

In Chapter 4, we underscore the need for assessing grid reliability while accounting for long-term inter-annual variability in supply as well as demand side. We enlist the limitations of chapter 2 result

because of unavailability of long-term consistent demand records. Thus, we propose to conduct reliability analysis of three different Balancing Authorities, that is ISO-NE, CA-ISO, and ERCOT to account for spatial heterogeneity which influences temperature, and also use hourly reconstructed demand proxies from Chapter 3 in conjunction with synchronous solar, onshore wind, and offshore wind capacity factors over four decades. We find that ELCC values for offshore wind generators vary significantly across years and has a coefficient of variation value that is 3 times larger than the coefficient of variation value of ELCCs from solar generators over 40 years between 1980 - 2019. We conclude that because offshore wind ELCC contributions are significantly large in east and west coast, ISO-NE and CA-ISO should include more offshore wind irrespective of large variability. Whereas ERCOT witnessed the largest capacity contributions from solar generators, which is also

less sensitive to interannual variability effects of weather variables. Thus, it is worthwhile to provide capital subsidies for solar generators in Texas.