Learnings

Outcomes

In addition to work package (WP) reports, the key outputs of the project have been a pair of prototype analysis tools:

• Stability Automation Tool (SAT): includes a suite of modules to read in market dispatch data and set up appropriate scenarios in DIgSILENT to run steady state and dynamic studies. The tool includes an automated stability identification module to check for system stability using time domain results without the need for any manual intervention. The output from the tool provides an understanding of the long-term capability of a boundary considering rotor angle stability as the constraint.
• Stability Classification Tool (SCT): a machine learning model which accepts data from SAT to train a classifier based on random forest algorithm. Several steps are necessary to process the output from SAT to create a final dataset which is suitable for training and testing of the machine learning model. These steps are handled by the tool and do not require manual intervention. Once the model is trained, it can be used for horizon scanning for stable/unstable scenarios without needing any resource intensive RMS studies.

These two tools are packaged together in a single platform called Stability Tool which can be accessed through a web-browser based user interface.

Lessons Learnt

Review of benefits case

The developed tools have proved the concept of the practicality of doing probabilistic stability analysis by utilising automation and Machine Learning techniques. This is a step forward towards the capability of running year-round stability analysis and to be able to run for more boundaries and contingencies in the network.

As with any new tool, the stability tool has constantly evolved throughout the project, as more features have been added to it and other application areas explored. Therefore, rather than viewing the tools developed in this project as the final, one-off delivery of a piece of software, NGESO could continue to further develop these to include new functionality and to accommodate new applications as and when required by the business. This has been made possible through the use of Github version control system which keeps a track of the development and any changes made by a contributor. This will allow NGESO to tap into a larger pool of developers to work in parallel, as well as implement an automated testing framework to identify any bugs in the source code due to periodic developmental changes.

The stability tool has been developed to streamline the long-term boundary capability assessment study which considers rotor angle stability as a constraint. The tool has automated several processes which are otherwise done manually at present, and which require significant time and effort. The tool has introduced several new methods, both in terms of analysis approach and managing large amount of data produced from the studies. These methods are new to the power industry and therefore, require extensive testing and validation as well as some possible changes to the NGESO IT systems.

During the development of the tool, we identified several challenges which are discussed in detail in this report. All these challenges have already been addressed in the project except one on scenario profiling.

Next steps

A fundamental limitation of year-round dispatch for the ETYS model from existing market dispatch simulators is the power factor set point of the generators. Market dispatch simulators are usually based on a linearised DC load flow, which naturally means that the dispatch decision is primarily guided by the market dynamics and the active power operational limits of the machines. There is no mechanism to guide the reactive power dispatch of the generators. This means that even if a dispatch is valid with respect to the operational MW rating of a machine, the operating point can nevertheless end up being at either of the extremes of a machine’s capability curve.

For most of the studied dispatch scenarios, it is observed that some machines are operating at highly under-excited condition i.e., absorbing a very high amount of reactive power. We suspect that this is a result of applying the dispatch from a linearised load flow to the dynamic AC simulations within the ETYS model. This dispatch leads to excessive line charging vars (capacitive reactive power flow from long circuits) in pockets of the network causing nearby generators to operate in under-excitation mode. This situation, coupled with the salient pole representation of all synchronous machines in the GB network, means that a large percentage of machines are operating close to their steady state stability limits. Therefore, even a small disturbance can lead to unstable behavior of the system.

Without managing the reactive power dispatch of the generators, the studied scenarios from POUYA will cause unnecessary voltage and stability issues under intact operation, not to mention problems of non-convergence. Unless every scenario is profiled to an acceptable dispatch level under intact operation, simulation results from contingency analysis will not be able to identify actual problems in the network.

Proposed solution for scenario profiling

The first step to reduce voltage issues would be to include power factor as a constraint in the optimal power flow (OPF) solution in POUYA. This will ensure the reactive power dispatch from generators are within the machine capability limits. This however means that the current implementation of OPF will need to be changed to a non-linear formulation. This will prevent synchronous generators from operating close to their stability limit. However, this does not guarantee acceptable voltage profile across the network and a separate profiling exercise will still need to be carried out.

Another approach could be to take actions after the dispatch data have been generated. This can be achieved by profiling voltage across different substations ensuring that shunt compensation devices are the first choice to provide reactive power support while synchronous generators are the last option and only provide the remaining amount and stay close to their nominal operating point.

This has been used in the past within NGESO for profiling winter peak scenario. Since then, it has been developed further in this project for use with year-round scenarios by profiling in parallel the substations which are electricity far apart. This has significantly improved the time taken to profile individual scenarios. In addition, after successfully profiling a scenario the state variables of the network are saved in a file so that successive scenarios can utilise this information. However, even after implementing these changes, a major limitation of this approach is the amount of time taken to profile a scenario which has very different demand and generation dispatch characteristics than already profiled ones.

A potential solution to this problem could be a clustering approach. To utilise the network state information more efficiently and improve on the time taken to profile a scenario, it could be useful to segregate the scenarios first based on certain network features which are important with respect to the purpose of the study. As an example, for stability studies the effective inertia of the network is an important consideration. Based on BMU dispatches and inertia information of individual units in a BMU, zonal inertia factors can be calculated, and these can serve as a feature to create clusters from the year-round data. Additional features can be included such as demand levels in individual zones and so on.

This approach can be included within the stability tool, or it can be implemented as a separate process and only profiled scenarios can be exchanged with all multi-scenario tools used within NGESO.

Dissemination

An abstract with the interim findings was accepted for paper submission in a Cigre Paris 2022 session, taking place from 28 August to 02 September 2022