Learnings

Outcomes

See the “Performance Compared to the Original Project Aims, Objectives and Success Criteria” section for outcomes from each work package.

This project developed and tested a day-ahead and intra-day system inertia forecasting model with multi-time resolution. Due to the lack of GB inertia measurement, the proposed inertia forecasting model was tested with publicly available data, i.e., Nordic system. The results indicated that the model is able to provide accurate forecasting of the day-ahead frequency response procurement and that the real-time system operation can benefit in terms of optimality and efficiency. 

This project also provided a benchmark for the inertia forecasting models, the outcomes of WP4 (Forecasting Model Integration and Testing) on the Nordic system was a success, i.e., achieving on average 4% MAPE compared with 7% MAPE for the state-of-the-art time series forecasting model. When the system operator approved GB inertia data is in place, the model can be tested under GB context. If the approach is successful there is potential for a follow-on project to explore options to demonstrate the approach. This would be subject to a CBA and comparison to other approaches being developed with the system operator.

Lessons Learnt

The key lessons learnt:

• Feature selection is important in actual forecasting. Proper feature selection results in better model performance compared to more exhaustive machine learning approaches that use all available variables.
• All inertia providers need to be taken into consideration in measuring system inertia. This project captured the impact of embedded generation on inertia.
• Inertia uncertainty affects the overall frequency response procurement costs. The key impacts of different levels of inertia uncertainties on frequency response procurement costs are quantified in this project.
• Publicly available or ESO approved data is essential in tuning the model and testing the accuracy under real operation context.

Further research topics identified:
Following the conclusion of this project the recommended next steps are:

• Validate and benchmark the inertia forecasting model under GB context when inertia measurement is available. The output from this project provides a proof-of-concept of the inertia forecasting tool and verifies its performance with Nordic data, for the application in GB. Follow-up work on its validation and benchmarking based on the GB inertia measure data is still needed when system operator approved inertia data is available.
• Expand the inertia forecasting model from deterministic forecasting into probabilistic forecasting. Probabilistic forecasting will provide more complete information for decision making. Further investigation on the probabilistic inertia forecasting and its integration with operational tool is required.
• Investigate the impacts of decreasing short circuit level and system strength in high power electronics penetrated systems.
• Investigations into other challenges brought by the increasing penetration of renewable energy and power electronic based devices such as the short circuit level bad system strength.