Re-dispatch simplification analysis: Confirmation holism and assessing the impact of simplifications on energy system model performance

Confirmation holism presents a serious but little-discussed challenge for testing energy system models using historical datasets. Confirmation holism is the thesis that no individual theory can be tested in isolation, but rather testing a theory relies on supporting axillary theories, and the primary and axillary theories can only be tested as a whole. The problem arises when using fit to historical data to assess if modifications have improved or degraded a model's representation of a real-world energy system. If the base model is highly simplified, then it is very challenging to tell if an improved fit to data is caused by a corresponding improvement in the representation of an energy system component or if, instead, the additional error introduced by the modification simply better balances out other errors in the model. We define a method, which we term re-dispatch simplification analysis, that addresses the issue of confirmation holism by using a highly detailed and accurate model of a centralised dispatch procedure to test simplified representations of the electricity system, re-dispatching the model with limited simplification to reduce sources of confounding error. We demonstrate the application of re-dispatch simplification analysis using a model of the Australian National Electricity Market dispatch procedure to test a range of commonly used energy system modelling simplifications and assess their relative impact on unit dispatch and market pricing outcomes. The study finds that neglecting the representation of Frequency Control Ancillary Service provision and unit ramp rates has a relatively small impact on model outcomes. However, the constraints associated with these model components can impact the testing of other simplifications. Simplifications to unit bidding behaviour are shown to have a relatively large impact on dispatch and pricing outcomes. While neglecting network and security constraints is shown to primarily impact peak pricing outcomes. Lastly, neglecting unit commitment and outages is shown to have a larger impact on dispatch outcomes than neglecting capacity withholding and partial outages. In this work, we use re-dispatch simplification analysis to compare the importance of simplifications across different components of electricity system models. However, it can also be used for testing alternative simplifications for a single component, such as different methods for formulating simplified network constraints. Further, re-dispatch simplification analysis should be applicable to any energy system that uses centralised dispatch, although the ability to construct the necessary dispatch model from public documentation and data, or to access the actual dispatch software used by the system operator, will likely vary between jurisdictions.