28 September 2021

Effect of the weather on nuclear production and electricity market: test our forecasts!

Heat waves and droughts are increasingly important weather hazards for the French electricity market, particularly via the availability of nuclear power plants: since 2015, these factors have led to nearly 360 shutdowns or deratings for French nuclear power plants, causing up to 6.2GW of unavailability.

As a specialist in the study of climate risks, Callendar has developed a powerful solution for predicting these unavailabilities. This solution is designed for balance managers, traders or producers who want to anticipate outages caused by weather conditions before they are announced by EDF.

Are you interested in this product? We can offer you a demo for Golfech power plant until D+4. To learn more, please contact us!

Questions and answers

What is the cause of weather unavailability for nuclear power plants?

Nuclear power plants, like most conventional power plants, need water to cool their turbines and produce electricity. in France, this use of water is subject to regulations that vary from plant to plant. For example, the temperature at the discharge point or downstream, in the case of a river, may be limited. Details of these regulations can be found here.

In periods of extreme heat or drought, these limits may become unworkable. In this case, the operator must reduce or stop reactors’ production.

What is the impact of weather-related outages on the electricity market?

Between 2015 and 2020, we identified 357 total or partial unavailabilities caused by the weather. The most severe weather related disruption occurred on July 25, 2019: it affected 9 nuclear reactors simultaneously, for a total unavailable power of 6.2GW. This episode led to a significant increase in the price of electricity: over 70€/MWh on the spot market in France.

How does our prediction of weather related unavailability for nuclear power plants work?

At the heart of our predictions is a machine learning algorithm trained to predict the temperature of rivers. The training use both public data and on-site measurements mined from EDF’s communication. This innovative approach is described in more detail here.

Our system evaluate the temperature of the cooling water used by the power plants in function of the meteorological conditions: recent temperatures, weather forecasts, flow rate, length of the days…. This temperature is then used to calculate the maximum cooling capacity, taking into account the authorized thermal discharges.

The shutdown program communicated by EDF is used to calculate the cooling requirement. If this need exceeds the cooling capacity, the plant will be forced to reduce its production.

How do we know if our predictions are performing well?

In line with good practices for this type of project, the performance of the system is evaluated on a validation sample: real data that have not been used during design and training. Under these conditions, the average absolute error of the temperature predictions is generally less than 0.5°C.

The re-training with data from EDF also allows to eliminate the biases that may exist between the public measurements and those made by the operator. After this re-training, the frequency of errors decreases significantly as illustrated below:

Availability forecast for a nuclear power plant depending on weather

Example of availability forecasts: contrary to a conventional model, our system correctly anticipates the interruption of electricity production at the Golfech power plant on July 23, 2019.

What are the known limitations of these forecasts?

Even if it already offers good performances, our forecast of the meteorological unavailability of the French nuclear power plants remains in development. It therefore has some limitations:

  • The water flow is taken into account but not modeled: it is considered constant and equal to its average of the last 24 hours between D and D+4,
  • The effect of temperature on the efficiency of the turbines is not taken into account, this decrease in efficiency is about 1 to 2 points per 10°C.

We are continuing our work in order to offer even more accurate and complete forecasts.

You want to know more? Contact us!