Benchmarking Explanatory Models for Inertia Forecasting using Public Data of the Nordic Area

TL;DR

This study benchmarks a day-ahead explanatory model for inertia forecasting in the Nordic power system, demonstrating its accuracy, generalizability, and the impact of feature enhancements, with publicly available data for further research.

Contribution

It introduces a novel explanatory model for inertia forecasting that outperforms traditional time-series models and provides a publicly available dataset for benchmarking.

Findings

The explanatory model reduces MAPE by 43% compared to time-series models.

The model performs consistently across Nordic and Great Britain datasets.

Feature enhancements further improve accuracy, with a monthly interaction variable reducing MAPE by up to 18%.

Abstract

This paper investigates the performance of a day-ahead explanatory model for inertia forecasting based on field data in the Nordic system, which achieves a 43% reduction in mean absolute percentage error (MAPE) against a state-of-the-art time-series forecast model. The generalizability of the explanatory model is verified by its consistent performance on Nordic and Great Britain datasets. Also, it appears that a long duration of training data is not required to obtain accurate results with this model, but taking a more spatially granular approach reduces the MAPE by 3.6%. Finally, two further model enhancements are studied considering the specific features in Nordic system: (i) a monthly interaction variable applied to the day-ahead national demand forecast feature, reducing the MAPE by up to 18%; and (ii) a feature based on the inertia from hydropower, although this has a negligible impact. The field dataset used for benchmarking is also made publicly available.