Electromagnetic Transient (EMT) studies are necessary to accurately assess system stability at times when renewables, batteries or HVDC links are the main sources of energy. EMT stability studies currently rely on manufacturer-provided “black box” models (compiled code). The Neural BB project (NIA2_NGESO082) which was completed in September 2025 showed that a neural network “surrogate model” of a windfarm could be used in place of its black-box model. This gives the potential for faster execution, recompilation without equipment-manufacturer support (which is often unavailable) and creating simple models to represent multi-converter subsystems. Neural BB II aims to extend this work from just a “proof of concept” windfarm to a diverse range of six models by April 2028. It will also investigate the potential for accelerating studies by implementing the surrogate models on a Graphical Processing Unit (GPU).

Benefits

The project should allow EMT studies to run more quickly (by eliminating the factors in 2.1 above whereby black box models slow studies down), while eliminating accuracy problems that could be caused by the often-poor long-term support provided for black box models. These improvements make it feasible to use EMT studies to accurately verify the grid stability of more scenarios than would otherwise be the case. This in turn would lead to at least one of the following benefits:

i)                Higher grid security due to the ability to check potentially unstable scenarios using accurate EMT studies rather than less accurate approaches, and/or

ii)              Lower system constraints, and hence lower constraint costs, as it will not be necessary to reduce system boundary capacities by a margin of safety in order to allow for potential inaccuracies in non-EMT methods of stability assessment.

 Knowledge benefits

Impact is High and is expected during project and after rollout:

The use of machine learning to train surrogates of EMT models of full AC/DC converters (as found in HVDC, wind, solar, battery, etc) has not previously been demonstrated in the GB or elsewhere with this level of detail.

Neural BB Phase 1 succeeded in proving the concept is possible and developed the technology into TRL 4-6. This project would move the concept into TRL 7-8 and into testing in business cases. Additionally, this project would investigate leveraging GPU acceleration methodologies for machine learning which is currently an area of interest internationally.

Confidence of Impact is high:

The project aims to assess accuracy using more metrics, test and improve versatility with different converter types and introduce best practice methodologies that can be rolled out across NESO. This means that regardless of the results of these assessments increase of progress in the field of neural network surrogatisation of EMT networks is guaranteed.

Environmental benefits

Impact is Low and is dependent on rollout after project:

EMT studies have a track record in more accurate constraint assessment resulting in improved boundary flow limits. This facilitates renewable and interconnector resource utilisation, reducing the need for using dispatchable fossil fuels to offset constraint.

Confidence of impact is medium:

EMT analysis creates the most impact when applied to constraints affected by converter stability and fast sub-millisecond transients. These generally correspond to inverter-based resources which cover the majority of renewable generation and HVDC interconnectors. The limiting factor is often speed and scaling of computational power, this project offers a technology that directly increases speed while halfling computational load, However, this depends on rollout scale and constraints change with energy mix and network topology introducing uncertainty that makes exact prediction difficult.

Financial / Operational benefits

Impact is High and is expected after project and is dependent on rollout:

EMT studies have a track record of saving millions in constraint costs,  however the studies may require 4-6 week on reducing a large EMT model and dispatching it to reduce run times. This would amount to an estimate between £20k and £100k per study depending on the complexity of the original model.  Additionally, engineers would need to integrate EMT models with often different timesteps into the network model with simulation time increasing exponentially with the number of models and their complexity and the model timestep being fixed to the smallest timestep between these models. Neural BB has already proven a doubling of speed due to increase of timestep from 10-20us to 40-50us, effectively halving simulation time and due to the neural network nature does not introduce the same  
complexity as detailed models. This means with Neural BB network model reduction may not be necessary, model numbers would have a small effect on computational load, and the model would be able to run at twice the speed.

Confidence of impact is Medium:

The financial impact will depend on the results of the limitations assessment and updated accuracy tests which will affect the uses cases and hence size of rollout within the organisation. In WP5 in the consultation stage we will consulting internal NESO teams to assess the possibility of NESO wider rollout. The actual rollout will not occur after project completion but will happen slowly and gradually building confidence on a small scale before moving on to justify the business case costs for wider rollout stages.