Modelling uncertainty when planning for energy systems
Amy Wilson (University of Edinburgh)
Friday 24th March, 2017 15:00-16:00 Maths 203
The changing energy landscape has led to an increased need for mathematical and statistical modelling of the future energy system to help make evidence-based policy decisions. The management of uncertainty in these models is critical so that modelling results can be related back to the real-world under study.
In this talk, two aspects of statistical modelling for energy system planning will be considered. The first concerns the assessment of the risk of shortfalls due to insufficient generating capacity. Statistical methodology for assessing this risk and for deciding how much future capacity to procure based on the resulting risk level will be discussed. The second aspect focuses on the quantification of uncertainty when using time-consuming computer simulators to make policy decisions. This uncertainty arises because no simulator can perfectly replicate the complex real-world interactions making up an energy system. In addition, outputs from computer models are often heavily dependent on uncertain inputs (e.g. electricity demand in future years). Using traditional Monte Carlo simulation to assess these uncertainties is infeasible because of the long run-time of the computer simulator. To resolve this issue, regression models with correlated errors will be used to emulate the computer simulator. A varying coefficient model will be presented for circumstances when few simulator runs can be performed. Results will be given for two case studies. In the first, a sensitivity analysis is performed on future projections of the wholesale electricity price. In the second, the level of government support required for investment in renewable technologies so that EU targets on renewable generation and emissions are met is investigated.