Abstract

Computer
models are widely used as decision-support tools for planning energy systems
in both industry and government. These computer models are often
computationally intensive and have high-dimensional input spaces, making it
difficult to quantify the impact that different sources of uncertainty have
on model output. Without a complete picture of the effect of these
uncertainties it is difficult to take planning decisions that are robust in
the real-world. This presentation will discuss methodology for accounting for
uncertainties in computationally intensive energy planning models. Both input
uncertainty and uncertainty in the structure of the model itself will be
considered. An emulator, or statistical model of the underlying computer
model, will be used to quantify uncertainty in areas of the input space where
it has not been possible to make model runs. This emulator will be combined
with a description of the uncertainty over the input space and a description
of the structural error to quantify uncertainty in model outputs. Several
real-world examples in energy planning will be discussed, including the
modelling of wholesale electricity prices and making decisions about
renewable support schemes.