A Climate Resilient Electric Power Grid is both Imperative and Possible: Preliminary evidence from the British Onshore Wind Energy System

Oral Presentation

This paper begins by reporting a statistical relationship between hourly CO2 concentrations and the hourly net radiation flux at the Earth’s surface. This relationship is of interest because the net radiation flux at the Earth’s surface, which represents the balance between incoming and outgoing energy, is recognized as an important driver of the weather and climate system. Based on the statistical relationship between CO2 levels and the net radiation flux, there is a possibility that the shift to wind energy generation to ameliorate the effects of climate change could give rise to a less resilient power grid because of climate change. While a loss in resiliency pales relative to the damages associated with climate change, a decline in the electricity system’s resiliency could have high economic costs, given that electricity is essential to all modern economies. 

The analysis proceeds by presenting a method to improve the short-term predictability of wind energy generation in the British onshore electricity system to enhance the power grid’s resiliency. The method is largely statistical but also explicitly recognizes the importance of expected meteorological conditions. The analysis also recognizes that the wind energy generation data at the half-hour level of granularity are significantly autoregressive but also highly volatile at times. Based on these properties, a statistical machine learning approach known as ARCH/ARMAX ( Autoregressive Conditional Heteroskedasticity/ Autoregressive Moving Average with Exogenous Inputs) model is formulated. The model’s exogenous inputs include simulated meteorological variables such as temperature, air pressure, wind speeds, and air density. Another key exogenous input is the final physical notification (FPN) of the expected generation that wind farm operators provide to the system operator one hour before real-time. The model is estimated using 30-minute data from Jan 1, 2018, through Dec 31, 2021. The preliminary out-of-sample predictions have a weighted mean absolute percentage error (WMAPE) of about three percent, which is more accurate than the FPNs that the operators report to the system operator. These results indicate that this research approach may be useful in reducing climate change’s challenge to the power grid’s resiliency.