National Grid ESO Man Electric Scooter Glow
4. Year-round probabilistic analysis
ETYS 2020

4. Year-round probabilistic analysis

The flows on the system over the next 10 years are uncertain. This is caused by a range of generation sources that can be variable and intermittent within the different boundaries presented in chapter 3. When high wind, solar and other embedded sources are at their peak output at the same time, it can lead to severe network constraints and in some cases, this can be very expensive.

Probabilistic analysis looks at year-round conditions. It captures more snapshots, building on what we already do in the previous Chapter 3, by assessing many generation and demand background conditions that the system could face.  

This could lead to more informed network investment and operational planning decisions, by considering the potential risk and the cost involved to mitigate it. 

In ETYS 2019, we showed how the probabilistic approach can provide more information on the transmission network’s requirements and performance. This year, we have extended this and our tool capabilities to now include: 

  • power flow control devices such as Quadrature-Booster (QB) 

  • optimisation techniques, for setting of these devices for both pre-fault (preventative) and post-fault (corrective) conditions  

  • an improvement to our statistical analytics

  • addition of data mining capabilities to help us understand year-round requirements, drivers, and opportunities.  

We are also exploring the development pathway for our probabilistic tool and in future publications we will showcase some improvements that we hope to implement.

We’re working with academia to understand how use this analysis this under the many generation and demand scenarios that could happen across a year.

We’re hoping this allows us to investigate a broader range of options that could improve year-round boundary performance. We’ll provide more detail in the case study section of this chapter.

How do we do this?

The analysis looks at year-round conditions and we can identify the most likely conditions that the system must be able to accommodate, as well as some rare extreme conditions. 

Although these rare conditions may only occur for a relatively short duration over a year, it is important to understand the additional requirements that such an outcome would impose on the NETS. 


1. Prepare snapshots

We start the analysis by looking at a snapshot of the system and seeing what generation, demand and assets are connected on the system at that time. From this, we can see what is happening on the system, what constraints are active and the drivers behind them. 

For example, if we were looking at a snapshot of a summer afternoon, we could see large amounts of embedded solar with minimal demand on the system and identify what constraints are reaching their limit. 

In our analysis we look at 10 different snapshots for every hour, which represents approximately 87,600 scenarios across the course of the year

Similarly, if we were looking at a snapshot with a large amount of wind generation instead we could see if different issues arise on the system as a result.

2. Identifying acceptable and unacceptable outcomes

An “acceptable” outcome is where the network does not see any thermal overloads after a credible fault has occurred in a snapshot or a scenario and an “unacceptable” outcome is when the network sees at least one thermal overload after a credible fault condition has been applied to the network.

To identify acceptable and unacceptable outcomes, we assess many power transfer scenarios across a boundary, based on how likely it is to occur and then summarise this in a probability distribution chart on the right.

From the graph, the blue line shows the number of acceptable outcomes that have occurred, and the red line shows the number of unacceptable outcomes for a particular level of power flow.

We can also break the graph into a few regions: 

greenunrestricted – no issues
blackwithin boundary capability limits (from Chapter 3) but we see some  unacceptable outcomes in certain rare scenarios. The scenarios are rare enough that, there would be a low likelihood having this scenario on the system in a year 
amberabove boundary capability limits (from chapter 3) and relatively higher likelihood for an unacceptable scenario. At this point, the event is manageable but more severe than the black region; constraints in this region may last for moderately low durations with a rare chance of seeing constraints of high magnitudes.
redwhen power transfer is not possible without further investment. This is because attempting to transfer power within this range is likely to result in constraints that last for very long periods as well as of very high magnitude. 

So, we can see that at a low power flow, it is more likely that we have more acceptable scenarios where the power flows are low. When we increase the power flow, there is a combination of acceptable and unacceptable scenarios. The unacceptable conditions will need to be managed through action should they occur and there could be an opportunity to propose solutions to manage these conditions economically and efficiently in planning timescales. 

The objective is to find ways to make the green shaded area bigger and reduce the amber region (where there are partial restrictions to transfer power). 

To turn the amber region green, we might need a range of additional capacity. If the duration of additional capacity is short, it may be possible to consider solutions that last for a short duration such as flexible-generation dispatch actions. We call this non-network solution. We can also consider network build solutions and compare them to non-network solutions to find the one better suited to resolving the issues.  

3. Determine any additional year-round system needs

Classifying the shaded regions in the previous step gives us an understanding of a boundary’s year-round transfer requirements. To better understand the regions, we introduced a new term - Boundary Congestion Probability (BCP).

The BCP is a ratio between the non-green and green region (from step 2). If the value of the BCP is small, it tells us that, generally, boundary congestions will be experienced for short durations and conversely if the value of the BCP is too high, it tells us that a boundary could be congested for longer durations. 

This information means we can choose a BCP value and use it to define specific boundary requirements. Defining requirements this way helps us to see the potential and the need for more solutions beyond the deterministic capability of the boundary. 

We could also look at the size of the additional capacity and how long it would be needed to overcome the constraint. We’ve plotted this below to show how additional requirement may vary throughout the year.


4. Assess various options and address requirements

We’ve used the “standard classification and regression tree“ technique or the CART analysis method to understand the size and the duration of a constraint.

The CART method follows a data mining process to identify key dispatch conditions that may influence the likelihood of an unacceptable outcome within the shaded overlapping region, in the earlier plot. 

We classify the key dispatch conditions by either technology type or region and use a binary high/low dispatch output criteria to understand dispatch conditions around the boundary that generate the lowest probability of an unacceptable outcome. 

This is illustrated in the diagram on the right where the technologies in region A and C produce the lowest probability of an unacceptable outcome compared to the combination of outcomes from technologies in regions A, D and B that results in a higher probability of an unacceptable outcome. 

By analysing the diagram, as an example, we can see that it is more desirable to operate the network with technologies from region A and C than from D and B to lower the probability of achieving an unacceptable outcome. 

For example, technology ‘A’ of a given capacity placed in region ‘X’ of the boundary can be more useful to alleviating congestion than technology ‘B’ of a given capacity in region ‘Y’. However, effectiveness is one factor amongst many, such as availability and cost. We are developing our analysis and techniques to calculate effectiveness outcomes from various technologies and locations to have a better representation of the magnitude of additional requirements. 

Therefore, we can show which regions or dispatch combinations can assist in lowering the outcome of an unacceptable transfer across the boundary and thereby lowering the BCP index—which as earlier stated measures the % of time in a season/year the boundary experiences congestion. 

Thermal Probabilistic Case Study - Winter Season

We’re seeing diverse generation mixes connected to the GB transmission network. These generation technologies are spread across the network and driven by differing uncertain inputs for which we apply a probabilistic season-round winter analysis. In this case study, we’re primarily concerned with understanding winter season-round operating requirements. 

To capture the requirements, we’ve collected hourly historical data of:

  • regional wind 
  • solar profiles, 
  • plant availability (both forced and random outage data), 
  • hydro and pumped storage typical loading patterns. We’ve applied this data to reflect our winter conditions. 
  • using our modelling of the European market dispatch we’ve generated typical interconnector dispatches as well as energy storage charging and discharging cycles. 

Like last year, we have validated our probabilistic approach by comparing our generation and demand dispatches against the historical data input. While this is internally verified, we cannot publish these results due to the risk of exposing third party confidential information. 

For our winter analysis, we generated around 22 000 scenarios of generation and demand dispatches. From this, considering both intact and contingencies network conditions, we produced around 100 000 network flows, on average, per individual boundary. 

Following last year’s publication, we’ve improved on our network analysis by applying pre- and post-fault actions using power flow control devices (e.g., QB) in addition to 6hr post-fault rating and HVDC flow control. These represent the types of actions already taken in operational timescales to maximise network capability.

The graphs on the right show the effect before and after the improvements were made. We see that without applying pre- post-fault actions the network begins to see unacceptable outcomes at around 2400MW. After pre- and post-fault actions are applied we see unacceptable outcomes starting to arise at around 2800 MW.

Example Case: B6

To show you how this might work practically, the following section presents results from using probabilistic analysis on the B6 boundary.

On the graph above, we can see that the black region, the area in which an unacceptable outcome was present below the boundary capability limit, covers the overlap between plots almost entirely. The black region equates to a BCP index of 6.66% or 145.9 hours. To turn this region green, solutions providing additional capacity for that duration need to be considered. 

Identifying additional requirements is not limited to just the black region only. We can see from the acceptable plot that there is potential for increasing power transfer beyond the peak capability level, however doing so will increase the BCP index accordingly due to the unacceptable outcomes increasing. 

By evaluating BCP values beyond the boundary capability limit we can consider new opportunities for solutions and evaluate whether it may be more appropriate to use a solution that provides short duration capability rather than year-round capability or a combination of the two.

The graph below shows how the BCP index will be distributed between duration and capacity. As stated earlier in the document, at present we are unable to represent the magnitude of additional requirement. However, we can see the maximum duration requirements will be needed for. For boundary B6, we can see that no single magnitude will exceed a duration of 40 hours when transferring power at a time.

To better understand how certain network conditions influence the probability of an unacceptable outcome, we can look at scenarios on a CART diagram.

When the network sees embedded generation output across the North East of England and Central Scottish regions, there is the lowest probability of an unacceptable power transfer outcome across B6. On the contrary the highest chance of an unacceptable transfer outcome occurs when there is high embedded generation output coupled with low wind from Central Scotland and low interconnector and storage output around the B6 region. This analysis allows us to identify solutions to accommodate additional requirements to securely transfer power across the B6 boundary.

We’ve done this similarly for boundaries B2 and B4 and their results are summarised in the table below:

BoundaryBCP IndexCART Summary
B20.92% or 20.1 hours over the winter season at chapter 3 capability valueA combination of high wind in the Argyll and Bute region, low wind in the North East of Scotland and relatively high hydro output results in the best network performance. Low wind output across the central highland and Argyll and Bute regions in Scotland results in more constrained boundary performance
B42.29% or 50.1 hours over the winter season at chapter 3 capability valueLow output from central Scotland, low hydro output and low wind output in the Argyll and Bute region, results in relatively poor network performance. Low output from Central Scotland but a high hydro output results in relatively better boundary performance.

Probabilistic Thermal Analysis Methodology

Our probabilistic approach uses historical profiles as inputs to a Monte-Carlo method that samples those inputs and uses the technical operational logic of generation and demand to produce realistic outputs of wind farms, solar panels, hydro units, generation units’ availability and demand dispatches.  We use these dispatches to estimate the likely power flow on individual transmission circuits or a group of circuits. A group of circuits are also known as a boundary as discussed in Chapter 3.

When Monte-Carlo is used to sample likely background generation and demand conditions, it produces hourly snapshots of generation and demand for each sample year. We then use Economic Dispatch (ED) to find out the probable dispatches of energy resources assuming an ideal electricity market. The results, which are hourly generation and demand snapshots, are evaluated by power system analysis based on DC power flow for a set of credible contingencies. Pre- and post-faults actions are applied, when applicable, to relive congestions and increase the transfer capability. The results from the power flow analysis makes us understand the impact on the GB NETS. 

Our probabilistic approach can be summarised by two key elements – the Monte Carlo sampling economic dispatch and the DC power flow network assessor element. The overall probabilistic process approach is summarised below:

Comparing with our deterministic approach

Developing our probabilistic approach has followed a learning by doing process, which involved reconsidering our deterministic process to identify the steps in the process that we could enhance and incrementally evolve toward a full probabilistic analysis process. 

The purple highlights in the probabilistic approach show the steps in our analysis where we have improved on our deterministic process. 

In the probabilistic approach, we consider about 10 samples an hour or 87,600 year-round scenarios. This leads to 87600 network flows per studied boundary when it is intact. When contingencies are considered against a boundary, we can generate 87600 network flows multiplied by the number of contingencies considered. We then apply pre- and post-fault actions to relieve congestions and increase the acceptable power transfer across the boundary. 

From this we can distinguish a network’s state into either an “acceptable” or an “unacceptable” power flow state outcome. We use these outcomes to perform analysis of network thermal requirements, from statistical and data mining approaches. 

Our improvements also enable us to use probabilistic generation and demand dispatch results to compare against our single snapshot generation and demand dispatch deterministic method. This helps us see how both the likely and rare worst-case dispatch scenarios compare between the two methods. 

Development Pathway

We are continually working to extend our tools’ functionalities. Our probabilistic work is one of our pathfinder projects, where we are learning by doing and are shaping our thinking as we apply our new tools to real data. 

We are investigating various techniques to integrate probabilistic analysis into the ETYS and the NOA. One of the options is the BCP concept introduced; another one we are working on is a concept called Dynamic Boundary Capability (DBC) which relies on risk-based techniques to calculate multiple boundary capability(ies) per season based on the background condition(s).  

Below, we have provided our development pathway. We are working to develop a bespoke joint market and network tool for GB thermal constraint analysis as well as proof of concept for integration of the year-round technical analysis into ETYS and NOA processes.

Aside from the probabilistic analysis, we are also working on other innovation projects. You can find out more information in our Network Innovation Allowance summary document which can be found here.

Some of the projects are listed below:

Advanced Modelling for Network Planning under Uncertainty (University of Melbourne)

This project was established to independently validate the economic and technical aspects of our NOA methodology, compare our process to those used in other countries and to explore the potential for new analysis tool. The project provided useful recommendations for the tools enhancement which we will publish in a separate report early next year.

Applications of convex optimisation to enhance National Grid’s NOA process (University of Strathclyde)

This NIA project focuses on developing new tools and techniques to better assess GB year-round reactive power requirements. A python-based tool is now developed with multiple techniques to do year-round voltage assessment. Further on, we are going to do extensive tests and study using GB data to make the tool fit for purpose.

Probabilistic planning for stability constraints (TNEI)

We are also running another innovation project to explore, develop and test probabilistic approaches for modelling of angular stability. This will enable year-round boundary capability calculation for stability accounting for a number of sources of variability and uncertainty. The project is expected to finalize and publish its findings by end of 2021. You can find more information about the project here.

Jump into the ETYS

Click on the sections below go to the different chapters in the ETYS.

National Grid ESO - download electricity ten year statement 2020 - man with phone.jpg

1. What is the ETYS?

Find out about the ETYS and how this fits with our entire planning process

National Grid ESO - lightbulbs hanging from roof

2. Network Development Inputs

Here is more information on how we prepare the ETYS

Bridging the gap to Net Zero Wind turbine fields in Cornwall

3. Electricity Transmission System

You can find out about the various regions of the transmission system and their capabilities

Early Competition Header

4. Year-round probabilistic analysis

You will find how we are applying year round conditions to assess the capability of the system

National Grid ESO - annual report and accounts - car on a road

5. Way forward

Find out about how we are going to improve the ETYS and how you can get involved

6. Further information and appendices

Get to the appendices, glossary and other helpful contact information

Get in touch

Email us

Email us to learn more or get involved in the EYTS


Signing up to the mailing list

Sign up to receive updates progress of the pathfinders

Tweet us

Follow us and join the conversation