Data-Driven Techno-Economic Analysis, Optimization, And Uncertainty Quantification of Integrated Energy Systems in Deregulated Electricity Markets

Electricity is a ubiquitous energy source in daily life, powering everything from stovetops and cellphones to vehicles and industrial processes. While wind and solar power have become increasingly common sources of electricity, the majority of electricity is still produced by burning fossil fuels, releasing greenhouse gases and propelling climate change. Wind and solar power cannot economically replace these fossil fuel energy sources on their own because they do not produce consistent power; the wind must be blowing, and the sun must be shining for them to make electricity. Nuclear power is a reliable source of energy that does not generate greenhouse gases when generating electricity but is not flexible enough to directly replace existing traditional generators. Pairing these nuclear power plants with energy storage technologies like “thermal batteries” could help them find this needed flexibility, but the economics of these plants are not well-understood. In particular, there is significant uncertainty in construction costs, operating costs, and what revenue these plants would bring in. This dissertation first puts a number to these uncertainties. While the construction costs for advanced nuclear power plants are the greatest source of uncertainty, the uncertainty in revenue is also significant for some markets, and many past studies have not considered this source of uncertainty. One way to measure the effect of variation in quantities which change over time like the electricity demand, renewable energy production, and electricity price is to use a statistical model that describes how these values change over time, then use that model to create many scenarios over which the energy system can be modeled. Previous studies have used models which are either not very well suited for modeling these quantities or use models which are not interpretable. Both having realistic scenarios and having an understanding of how these values relate to each other over time is important for understanding the electricity markets that energy systems operate in. Neural stochastic differential equations are used for the first time in energy systems studies for this purpose in this dissertation, and they are shown to perform comparably to state-of-the-art machine learning models while being more interpretable. The neural stochastic differential equation model developed here is used to optimize a nuclear power plant with thermal energy storage in the ERCOT market in Texas, which was the most sensitive to time series uncertainty of the markets considered earlier. A neural network model is used to estimate the price of electricity from the electricity demand, renewable energy generation, the amount of each type of generator in the market, and the price of natural gas, and this model is used to estimate how much the flexible nuclear plant will decrease the price of electricity and therefore the plant revenue. This analysis is performed for various cases of plant price and system sizes. The energy storage system makes the plant more profitable in almost all cases, but no benefit was seen for very expensive plants in small markets. This shows that adding energy storage to make nuclear power plants more flexible can make them more cost-competitive in electricity markets in many cases, though this should be evaluated on a market-by-market behavior.