Is eLMP Harder to Predict than LMP?

Abstract

The integration of large-scale renewable energy poses significant challenge to the real-time supply-demand balancing in the power grid, which reshapes the landscape of electricity pricing. To handle the uncertainty in the renewable generation outputs, system operators need to start up and shut down conventional generators more frequently, whereas the widely adopted locational marginal price (LMP) scheme fails to recover these frequent start-up costs, which causes inadequate incentive issues in the markets. To this end, extended LMP (eLMP) was proposed, which employs the uplift payment to compensate for the start-up costs. As eLMP is more complicated than LMP, it is commonly believed that the eLMP prediction will be much harder than the LMP prediction. However, in this paper, we submit that this common belief is unfounded through comparative study. We compare the prediction performances for the two pricing schemes measured by various evaluation metrics, including MAE, RMSE, and MAPE. The results highlight that eLMP scheme is in fact easier to predict than the LMP scheme in terms of prediction accuracy, and the prediction models trained by LMP can be directly used to predict eLMP with remarkable performance. However, through the robustness test, we find that the robustness of eLMP prediction is not as good as that of LMP prediction, which implies the complexity of eLMP scheme.