Learnings

Outcomes

 
The project outcomes so far include major enhancements to the Co-optimisation tool this includes: 
 
Over 500 generators with comprehensive Unit commitment parameters (e.g., MNZT (Minimum Non-Zero Time), MZT (Minimum Zero Time), etc.) have been incorporated into the co-optimisation model.  
All frequency-related constraints have been included in the tool for co-optimisation of Energy and Frequency (COEF) - Rate-of-Change-of-Frequency (RoCoF), Frequency Nadir, and quasi-steady-state requirements.  
Frequency services related to inertia, primary frequency response (PFR), dynamic containment (DC), the influence of dynamic regulation (DR), dynamic moderation (DM), and the optimised largest power infeed/outfeed are involved in the co-optimisation model.  
The interconnectors power flows (power import and export) are optimised in the model to minimize the whole-system costs are included to model the largest power infeed or outfeed, depending on its status of power import and export.  
The model is now capable of capturing the influence of demand-side inertia.  
Requirements for both Low and High Frequency Security have been included in the COEF model.  
Operating reserve requirements have been included in the COEF model.  
Work on incorporation transmission network constraints and line outages in the COEF model has started.  
The influence of considering frequency-related constraints on operation cost and computing time is analysed. Under most cases, the computing time can be limited within 15 minutes.  
 The influence of minimum inertia limit is studied. 
 
Outcomes 
 
The COEF model can be considered as a novel tool that may provide cost optimal requirements for frequency and energy, by the application of the two-region model and the use of co-optimisation. 
  
It must be emphasised that under the specific conditions for which the tool was tested, the cost of securing the system was found to be higher, where only one region is secured, using co-optimisation. The reasons for this are not fully understood and further work is needed to determine the context in which this occurs, and if there are other situations where the costs may be lower, therefore giving the option of a strategy that uses regional co-optimisation when it is most beneficial and using the existing process when it is not beneficial. 
  
There is potential for the tool developed to be used to provide some fundamental evidence to Ofgem, DESNZ, industry etc., about the importance of co-optimisation between energy, reserve and frequency related ancillary services. Also, future market designs could be stated as future works, e.g., incentivising different stakeholders to contribute to the frequency security of the GB power system and co-optimisation process (e.g., currently there is no income for provision of inertia). In general, the COEF model is flexible enough to be used in different contexts.  
 
Comprehensive details of the outcome of the project can be found in the final report delivered by Imperial College: “Development of a Novel Software Tool for Co-optimisation of Energy, Reserve and Frequency Ancillary Services. Tool for Co-optimisation of Energy and Frequency-containment Services, Funding Source: NIA, Sponsor organisation: National Grid ESO Imperial College Team, June 2024”. 
 

Lessons Learnt

 
Initial analysis has demonstrated significant benefits of coordinating frequency regulation services and energy delivery, while considering changes in system inertia. Detailed analysis of the potential benefits is planned in the upcoming work packages. A full list of lessons learnt will be published when the project is completed.  
 
Lessons learnt, 2024 reporting 
 
There may be potential value in regional frequency modelling, as the model suggests, it may be possible to secure the whole system based on a single region. How the region to be secured is determined as being most optimal is still not fully understood. 
   
More consideration is needed on the work required to evolve the solution to a practical production system that can be used in a live environment. Further explanation on how the most optimal region to secure is determined and how this can be better presented to the user in the output results would be a meaningful output. Currently the results show that securing the system on a regional basis result in a slightly higher cost than securing on a whole system basis; more detailed analysis on whether this is the case all of the time or only in specific situations (and if so which situations) is needed.   
  
In general, the COEF model can consider the security of both regions (e.g., England and Scotland). Also, securing both regions will lead to slightly higher costs than securing on a single system basis, due to the consideration of regional frequency oscillations (this may require more frequency ancillary services). 
  
The initial intent was for the ESO to run the tool on internal systems, however, further investigation of this approach resulted in the ESO having to request that Imperial College performed the testing on their site, due to the extensive IT security checks required to implement it on the ESO system, and possibly extending the project duration by double or triple the planned duration. Additional time or provision for a standardised process for remote server access to a vendor's IT systems may be beneficial for future projects of this type. 
  
The methodology presented places the responsibility for solving the problem solely within the domain of the System Operator. The industry comprises the SO, Generators, DSOs, DNOs, Consumers, Suppliers, Traders and other participants. alternative consideration to other possible solutions which can also be used in conjunction with the tool and that can also be considered innovative, where responsibility is shared with a broader range of industry participants to achieve the same outcome. The model is capable of considering flexibility services from different resources (e.g., storage, demand-side response, synthetic inertia, etc.) and then appropriately co-optimise application of these resources.   
  
Further study would need to be undertaken on the process used to calculate the largest loss in the model. Currently the assumption in the tool is that it is a variable that changes with the largest generation or demand asset. However, this is not always the case, and can depend on other factors. 
  
The largest credible loss may be affected by certain rules and exemptions or derogations which change from time to time. Therefore, the ability of the tool to enable to user to ensure the largest loss is selected correctly requires further work. The tool assumes that largest loss is a decision variable, but it could be also be modified as part of further work in future, as an input parameter, depending on assumptions.  
  
The tool offers the functionality to model Transmission Constraints and N-1 Contingencies. This process is already been undertaken by power system engineers, prior to performing response, reserve and energy scheduling, the functionality can be potentially removed from the COEF model 
  
The GB market is a liberalised energy market, that operates primarily as a 'Self-Dispatch' market, where generators or demand sites determine when to generate or consume energy, in order to meet their contractual positions. 'Centralised-Dispatch' typically occurs after gate close, when the balancing mechanism operates to ensure that the total system generation equals demand. Further work is required to determine how the tool can be used to co-optimise using both approaches as a staggered process. 
  
The section on Synthetic Inertia is particularly interesting, because the idea presented here is very new - that batteries could be connected to synchronous condensers, specifically for the purpose of increasing system inertia. This may be further developed to determine the effect of additional synchronous condensers on the rest of the network. If increased compensation is required for voltage as a result of the additional synchronous condensers, what would be the additional costs of this; this assessment (impact on voltage levels) was not in the scope of the project. 
  
The findings in the report imply that Nuclear Units can be dispatched. Due to the nature of Nuclear Power Generation, these units are entirely Self-Dispatch and can only be instructed during an emergency. Therefore, enabling functionality in the tool to modify the output of Inflexible Nuclear Units may not be beneficial, even if it would be a lower cost than scheduling more DC, because the output results would be invalid (unusable).   
  
The report includes a sensitivity analysis showing that using a 'Regional' Frequency model may result in higher balancing costs than using the uniform frequency model, whilst having a lower inertia limit reduces balancing costs. The specific percentage changes in costs are presented in the report. Further insights from the sensitivity analysis may give some indication of the direction of possible new markets, e.g. Demand Side Inertia or Wind Inertia, however further investigation would be recommended to validate the outputs from the report. It can also provide the basis for the direction of the level of minimum inertia for future FRCR Assessments.  
  
It may also be beneficial to give further attention to the provision of data from the market about their intended day ahead positions, and how this may affect the Self-Dispatch algorithm. The current timestep of 30 minutes is sufficient for any day ahead assessments. Where the tool requires testing and validation (e.g. due to any future modifications to the algorithm), much shorter time resolution (i.e. 1 second or less) would be required to perform such testing. Therefore, the users would require the interfaces for the capability of reducing the time resolution for calculations and would also need functionality to operate the tool in 'reverse' - i.e. to inject values based on real life events into the tool to generate a simulated frequency profile for the purpose of validation and verification.  
  
The current user interface requires heavy back-end programming and therefore any user would need to be skilled in the use of Python. Control engineers (even if skilled in the use of Python) are primarily focused on delivering multiple outputs for the preparation of the day ahead plan within a 2-hour window each day, typically during the morning shift period, and the use of the tool to deliver the mandatory and ancillary service requirements would make up a fraction of that time. The tool would need to be developed to minimise the effort and time required to generate the requirements, ensuring that it does not distract them from other tasks.  
  
The tool takes approximately 10 minutes to run and produce results, however it is run 4 times, to output the complete set of results for the different options, which is a total of 40 minutes of the control engineer’s time. Further work is needed to improve the running time to comparable levels. 
  
As part of the industry days facilitated by the ESO, it may be beneficial to have sessions, training vendors to better understand the ESO data sources available for performing innovation research work