Are America's Nuclear Plants Doomed?

The Project

This project aims to forecast changes in the wholesale price of electricity based on the factors we have identified as influencing the price, giving policy makers a tool to guide their decisions.

Recently, owners of nuclear power plants in the midwest have seen their profitability decline. In particular, Exelon corporation, the owner of several nuclear plants in northern Illinois, has lobbied for policy changes to secure their revenues and ensure their ongoing financial viability. These arguments are cast against the backdrop of an electricity sector being pushed to integrate more low-carbon sources of electricity. As nuclear currently constitutes the largest source of low-carbon power in the US, shuttering of these plants would make it more difficult to comply with the EPA's Clean Power Plan regulations, and, more generally, would make it more difficult to achieve the long-term emissions reductions neccessary to avoid the worst effects of climate change.

---

Location of Exelon's Nuclear Plants: At the Intersection of Two Power Systems

Exelon's nuclear plants in northern Illinois are within an area served by the PJM power system that is surrounded by the MISO power system.

---

Policymakers are now faced with the dilemma of whether or not and by how much to support these plants. As nuclear plant revenues are a direct function of wholesale power prices, the risks and tradeoffs of the debate could be clarified with an accurate forecast of future electricity prices—this is the challenge our project attempts to address.

The Price of Electricity

What determines the cost of your electric bill? At first glance, it may seem obvious: how much electricity you use over the course of a billing cycle. But the cost of turning on that light, or running that air conditioner, may be different from day to day, even if you power them on for the same amount of time. The reason for this is that the wholesale price your utility pays for electricity varies from day to day—in fact, it varies from hour to hour (your utility averages these hourly variations and charges you a daily price). The wholesale price your utility pays for electricity depends, in turn, on how much power your neighbors use, and your neighbors' neighbors, and so on. Thus, your electric bill is determined not only by how much electricity you use over the course of a billing cycle, but also by how much electricity was used by the people in your area. (If everyone turned on their air conditioners, the demand for electricity would increase, and its price would rise accordingly.)

Tour the Project

CS109 - Wholesale Electricity Price Prediction from Max DeCurtins on Vimeo.

---

Our Process

Process notebook on GitHub

---

Source: Manitoba Public Utilities Board

---

Predictors of Price

by size of effect

1. Total Demand
2. Natural Gas Prices
3. Hour of Day
4. Wind Generation Output
5. Coal Prices
6. Month
7. Day of the Week

Gathering Data

According to the US Energy Information Administration, the key factors that influence wholesale power prices are demand for electricity and the cost of fuel for generators. We gathered historical electricity demand data from wholesale market operator websites, and since natural gas and coal are the largest sources of electricity production in the region. Though wind generation still only makes up a small portion of total electricity generation, it has been fingered as a culprit for depressed wholesale prices, and therefore we also include it as a feature in our model.

Unlike the rate for electricity you pay at home, wholesale power prices vary by time and location. For this reason it was important for us to train our model on the price at the specific locations of Exelon's nuclear plants in question. The regional breakdown of power generation for this area is shown in the chart to the left.

Initial Observations

Observations and predictions were made based on a dataset of 32,017 records.

The figure below displays the results of our data collection efforts: the historical values of wholesale, electricity, gas and coal prices, electricity demand, and wind generation in the Midwest region. It is notable that wholesale electricity and gas prices both fell significantly in 2008–2009. Coal prices fluctuate but show no clear trend, while wind generation capacity has steadily increased since 2008 and shows increasingly sharp seasonal variation (while wind and demand are both displayed in units of MWh and scaled in a way to reveal trends, as wind is actually a tiny fraction of total electricity demand). Megawatt-hours (MWh) are a common unit of measure for both electricity generation capacity and demand, and the price data gathered by the project team for wind generation and demand is in dollars per Megawatt-hour ($/MWh).

Predicting Electricity Prices

The final prediction model captured 93% of the variance in wholesale electricity prices.

The project team evaluated a number of analytical techniques to predict wholesale electricity prices using the available data for the years 2008–2015 for overall demand, natural gas prices, coal prices, and wind generation capacity. Linear regression techniques generally produced poor results, achieving a maximum accuracy of only 44% (as measured by r-squared). This may partly be explained by the numerous non-linearities we observed in the relationships between features in our model. With this in mind, we decided to try a Random Forest Regressor, which is an ensemble of different regression trees and is considered appropriate for non-linear multiple regression. This approach resulted in enormous gains in accuracy: the final predictive model was able to capture 93% of the variance in wholesale electricity prices using the selected predictors. The project team suspects that the Random Forest approach was better able to capture the cyclical (i.e., non-linear) nature of our time series data, which fluctuates depending on the hour of the day, day of the week, and month of the year.

Forecasting Future Scenarios

Wholesale electricity prices rise dramatically as demand increases.

To test the model, we created several hypothetical scenarios including: 1) a doubling of electricity demand over the next 5 years, 2) a doubling of the natural gas price, which is roughly equivalent to a $50/tonne carbon tax, and 3) a tenfold increase in wind generation. The results are displayed in the graph below - in each case we see that future prices vary significantly from the baseline.

What happens?