Learnings

Outcomes

The project has delivered the demonstrated modules to the ESO supported by associated documentation. These modules were tested using scenarios proposed by the ESO and the results assessed against expectations. Within the confines of current maturity (using a historic 2019 model with a limited representation of assets) the modules delivered the results expected. 
 
The ESO has started design discussions for future implementation which would link into production systems.  It is anticipated that further work will be required with IT partners to adopt the processes proposed to work with ESO systems.  
 
Initially the ESO are proposing a phased implementation of module 3 as this is the more mature module. Adopting this phased implementation would consist of getting the module working with ESO production data locally, then with centrally but with ESO users before considering how to share the application of module 3 with the networks. 
 
Implementing module 3 includes development of the interfaces that would support future implementation of modules 1 and 2.  Through future innovation projects the Network Access Planning team are also investigating additional modules that could be introduced using the same workflow. 

Lessons Learnt

Working arrangements 
In part due to working arrangements during the pandemic testing was primarily completed working remotely. Combined with the need to use a historic power system module it was therefore not possible to perform parallel running with operational planners. To test the performance of the modules historic outages from 2019 were used and demonstration sessions were run to talk through and show how the modules work with Power System Engineers in the planning teams. 
 
Power System Model 
Due to confidentiality restrictions regarding the operational GB system model maintained by the ESO there was not an available sharable model to begin development on. The project developed a model from published data sources that could then be evaluated against the ESO operational models. In future a sharable model would be beneficial for projects, for the wider research community and for start-ups. 

Beyond the efficiency gains to individual projects of having a model that can be immediately used, the project identified several other advantages of having such a generalised model available for research:  

• Opportunity for testing and refining of the shared model to improve performance over time
• The model could be supported by automated anonymisation and other techniques to control or avoid access to confidential data
• Opportunity to share development of modules that could improve solutions through optimisation problems e.g., AC and DC Power Flow and Optimal Power Flow
• Provide an opportunity for developers to test out project proposals and evaluate new services
• If the shared model is of a similar structure and a validated performance compared to the ESO operational GB power system model, then implementation of projects will be smoother 

The Virtual Energy System offers an opportunity to transform this modelling challenge by opening access to the data shared direct from the data owner. This will help to manage the confidentiality restrictions on the ESO by sharing the data direct or with permission from its owner. This will allow the data owners to decide who the appropriate audience is for their data while presenting it in an interoperable way supported by common tools to integrate and use it. 

The project concluded that to further enhance the advantages, an open-source model of the GB network could be made available, implemented in open-source software to make it widely accessible for research. This open-source consideration is something that Virtual Energy System is also investigating, linked to Data Best Practice’s Presumed Open principle. This will require triage of which data can be made open safely as well as consideration of how to share tools and techniques. 

A particular difficulty that we encountered were with ‘controller’ objects, these are short code snippets that modify settings of elements during a power flow simulation. In the physical system these are the Automatic Tap Change Controllers or Automatic Reactive Switching or Active Network Management schemes. These were necessary to achieve convergent simulation studies to achieve a credible voltage profile. These models are not easily transferable between power systems modelling packages. Nominal target values were used for this project though in the physical system variable target values can be used in optimisation. The project recommended further investigation to include these in a Virtual Energy System that does not tie it to a specific software package. 

Standardised scenario generation 
Another lesson learned from this project is that it would be beneficial to have a common methodology to generate credible system scenarios aligned to the power system model. In the project’s opinion, it presents similar advantages to having an available model of the GB network, namely: less error prone, easier comparison between similar projects, simplified transition to operational models, and improved controls of data sharing and access.  

The scenarios could be updated over time to what the ESO believes are credible scenarios, which then makes it easier to validate already completed projects in the updated scenarios (e.g., ESO winter 2022 scenario: “Reduced electricity imports from Europe combined with insufficient available gas supply in Great Britain”). This could be implemented as a module of a shared GB network model. We believe that a wide range of scenarios could be included to enhance the advantages described above.

The project has used Optimal Power Flow models to generate models for a range of potential scenarios, these are used to ensure the scenarios are credible by managing overloads using the sort of actions the Power System Engineer would take.  This sort of automated scenario generation would benefit from further study.  Including, for example, the application of voltage control circuits, running arrangements and automated switching schemes. 

Model hierarchies 
The core observation that allowed us to reduce running times was that we could use a hierarchy of different power system models to approach different problems. For many problems, we can use the coarsest level of the hierarchy, a pure graph model, instead of a full-fledged AC power flow model. And on this coarsest level, we can use classical algorithms that run much faster than a power flow simulation. This allows us to reject many obvious problematic cases in under a second and, in turn, allows us to spend more computation time on the difficult cases or refer them to an engineer. This observation was crucial to achieve the speed-ups in Module 3. We recommend to also consider providing the data for the Virtual Energy System in a hierarchical fashion to facilitate the use of simpler models if appropriate. 

Double circuits 
To check whether the system meets the requirements, it is necessary to check the situation in the system after faults. To understand which faults are needed to check, we were given a simple set of rules and an additional list of cases referred to as “double circuits”. We did not fully plan for the incorporation of these into our models, which led to more difficulties than anticipated. To fully model these, it was necessary to use graph algorithms that allow for parallel edges, which is not widely supported by standard graph libraries. It was necessary to re-implement some basic algorithms to support these exceptional cases. We suggest checking algorithm availability and support for exceptional cases early in the design process. 

The project team also observed that the list of double circuits was not published as a data set though is shown on system diagrams where lines that form double circuits are drawn closer together. The project suggests considering sharing the list or a methodology to approximate the list. It might be of further research interest to develop methodologies to keep the list updated under future changes by aligning to geospatial data and considering whether neighbouring circuits that do not share a tower could be at elevated risks for example due to wildfires or floods.