The need to meet climate targets, resulting in a rise in renewable generation in GB, has led to an increased need for long-term mathematical and statistical models of the future electricity system. The management and understanding of uncertainty in these models is critical if robust policy decisions are to be made based on model output.
In this talk, two aspects of statistical modelling for energy system planning will be presented. The first part of the talk focuses on statistical models for assessing the risk of insufficient electricity generating capacity to meet demand. As data on electricity shortfalls in GB are limited, methods from extreme value theory are used to assess this risk. Results from a model of the marginal distribution of electricity shortfall over one year will be given. Methods for extending this model to a full time-series model will be discussed and the initial results presented.
The second part of the talk focuses on the quantification of uncertainty when using computer simulators to make energy policy decisions. This uncertainty arises because these computer simulators have uncertain inputs and are simplifications of the complex interactions making up a real-world energy system. Using traditional Monte Carlo simulation to obtain probability intervals for simulator outputs is usually not possible because the simulators are computationally demanding and so can only be run a limited number of times. To resolve this issue, regression models with correlated errors can be used to emulate the computer simulator. The emulator can then be used to approximate the simulator output for a particular set of inputs and provide uncertainty bounds on this approximation. A computer simulator used to determine the level of government support required to encourage investment in renewable technologies will be used to illustrate these ideas. Emulation of this simulator was challenging because the number of simulator runs available was extremely limited. The methods used to emulate the simulator alongside results from the uncertainty quantification will be described.