Project Summary

Our Predict4Resilience (P4R) Alpha phase will look to prototype a "weather fault
prediction tool" including the following:

• Prototype the Fault Forecasting Engine which is the fault predicting statistical
  model using weather forecasts and historical data, such as weather variables
  and satellite imagery.
• Carry out further user engagement with other DNOs and potentially other
  infrastructure operators to capture wider user needs and ensure the project
  continues in a direction to be commercialised and maximise uptake and value.
• Build Wireframes/Mock-Ups of the user interface to inform the Beta Phase
  design and development.
• Write a blueprint which evolves along the implementation of Agile Methodology
  to guarantee a fast start at Beta Phase.
• Develop a refined business case incorporating additional user needs from wider
  user engagement.

This addresses the Innovation Challenge by taking a data driven approach and
combining state-of-art weather forecasting and novel statistical methods to
revolutionise the process of electricity network fault prediction.

Currently, control room engineers rely on basic weather forecasts combined with
experience and intuition when making decisions about where to allocate resources
to restore supply. These decisions are not data-driven and are subject to human
bias. When a storm is approaching the UK, decision-makers are hesitant to
release their engineers to other districts, as they may be exposing their network to
a higher risk of lengthy outages. This leads to an overall slower response rate, as
resources tend to be allocated after faults have occurred.

The P4R application could be used by Control Rooms and Emergency Response
teams to enhance operational management, decision making and preparedness in
advance of any severe weather event. These users should be provided with a
forecast of the expected number of faults based on a weather forecast. Concise,
specific and visual data will enable the users to comprehend the information and
develop an effective response plan. In the Discovery Phase, we have developed a
deep understanding of users' optimal needs, which will be built into sprints during
the Alpha Phase

Through the proof of concept of P4R in terms of technical capabilities and
business value of fault prediction in general in the Discovery Phase our
understanding of the problem has evolved -- for example, we have learned that:

• A 5-7 day operating window is sufficient for control room engineers to make
  operational decisions and
• The priority is the accuracy (i.e. fault prediction) in this window.

We will build on this in Alpha to deliver a fit-for-purpose solution.

To deliver the objectives of the Alpha Phase, SPT have established partnerships
with:

• SP Distribution will provide user knowledge to inform the model and will
  witness the functionality testing of the prototype in various sprints. We will also
  engage with other licensees to ensure interoperability.
• SIA Partners: with extensive experience in control room operations, they will
  lead the technical development of the prototype platform with an agile approach
  and analyse the best commercialisation pathway for the tool.
• The University of Glasgow (UofG): as one of the leading UK universities in
  the fields of statistics, engineering, and energy forecasting, they will contribute
  to the modelling and building the prototype.
• The MET Office: utilising their expertise, data sets and learnings they will
  provide guidance to UofG and Sia Partners in the deployment of weather
  forecasting data to statistical and machine learning models.
• NG-ET will provide additional network knowledge to support the development of
  an interoperable solution.

ARUP is no longer part of the consortium, as it was mutually recognised they were
not best placed to continue the prototype development (now led by SIA Partners).
A full handover of learning is already underway.

Innovation Justification

We must minimise outage periods and the associated Customer Minutes Lost
(CML) as part of any maintenance or fault; with the currently available information,
estimating the impact of an event on maintenance and local fault volumes heavily
relies on user judgment. P4R needs SIF funding to support our level of ambition
within 6 months, accelerating our predictive capabilities..

No other past innovation projects have considered probabilistic fault
prediction and related decision-support, leaving a significant gap in DNOs'
predictive capability. Such capability offers many advantages:

1. Increased accuracy by leveraging advanced weather forecasts, new data
sources, and machine learning,

2. Short- to medium-range forecasting with uncertainty quantification, enabling
new modes of risk management and increasing resilience though early
warnings up to one week ahead, and

3. Consistent forecast data easily made available to all internal and external
stakeholders supporting open data and establishing cohesive practices.

This capability will benefit the DNOs with better prediction of faults, improved
responses to faults, decreased CML, and reduced unnecessary abortive costs of
cancelled planned works, creating a more resilient network.

There is still a journey to develop this solution. As for any innovative data-driven
project, there is a need for consistently produced accurate outcomes validated by
users. In addition, the quality of the output is highly dependent on the consistency
and accuracy of available data. In the current price control, there are no
alternative funding mechanisms to SIF that could support the adoption of this
project given its, risks, IP specifications, collaborative requirements and scale of
innovation.

Specifically, innovation is required in:

• Weather-related fault prediction which has not benefited from advances in
  digital technologies. We will leverage newly digitised asset health data with
  high-resolution ensemble numerical weather predictions to produce a worldleading
  capability.
• Probabilistic fault predictions which enable risk-based decision-making.
  Furthermore, the nature of faults, being relatively rare resulting in sparse data,
  requires specific statistical modelling, drawing on extreme value theory. We will
  develop an appropriate modelling capability to be used in combination with
  ensemble weather forecasts. This will accurately predict the likelihood of faults
  occurring across an entire electricity network for the first time.
• Medium-range forecasting - using ensemble numerical weather prediction to
  quantify the probability of future weather occurring in the days ahead. The
  further ahead we predict, the more uncertainty weather forecast become.
  However, ensemble NWP quantifies this uncertainty, providing information on
  probability of different weather situations arising.

The accuracy with which faults will ultimately be forecasted by this capability is
unknown, and therefore a risk. However, in the Discovery Phase, a literature
review was carried out and a proof-of-concept fault forecasting method was
implemented which verified the feasibility of the above innovations. In addition, our
models have shown forecasts are highly accurate in days 1-5, a key timescale for
operational planning. It was also found that the method successfully predicted
significant events resulting in large numbers of faults. This exercise has de-risked
the project and identified key areas for development improving forecasting.

In addition, from engagement with our users and literature review, we have seen
that forecast improvement and business value are attainable in the case of fault
prediction, although the nature of faults (low-probability events occurring primarily
during severe weather) presents a new technical challenge for forecasters.

Resilience continues to be of high importance to our sector and the project aligns
with Government policies regarding the future of UK energy infrastructure,
including the National Infrastructure Commission's Resilience Framework. Not
developing this capability will leave electricity networks vulnerable to weatherrelated
faults at a time when the frequency and severity of adverse weather
events is increasing due to our changing climate.

Benefits

This data-driven approach of the Weather Forecast System (WFS) can transform
human-centric decision-making practices and improve maintenance decisions.
The enhanced asset management and emergency response capability will deliver
benefits to consumers through fewer/shorter interruptions, to the environment
through reduced emissions, to the network through avoided costs, and lastly to the
wider industry through transferrable learnings.

During Discovery, the core benefit identified by SPEN's Control Room engineers
was savings in CML through early access to a credible fault caused by wind and
gale. On this basis alone, a quantitative analysis of the associated savings,
presented in the Business Case appendix, is based on the development cost
required and the net savings across 10 years to calculate:

• The benefits/costs ratio: 1.76
• The Net Present Value: £430k
• The Payback period: 6 years

The CML cost savings will only increase substantially when other weather-related
faults, such as those caused by snow, sleet and blizzards, are included, thus
further reducing the payback period.

In addition, as a social benefit to consumers, reduced CML translates to faster reconnection
times and better communication when faults occur which can result in
reduced stress during an outage. Based on the DNO-wide Social Value guidance,
provision of information and faster connection will help consumers cope with
outages. In alpha, we will use the SROI tool to quantify this further benefit.

The baseline for annual savings of developing P4R is assumed to be 5% in the
business case. During the Alpha phase, we will work to validate this assumption
and measure how it will impact the duration of outages for consumers. This will
help us assess the benefit to consumers with financial metrics.

Benefits to the Network

Credible forecasted location and number of faults will result in avoided costs to the
network on a few fronts including more strategic response team mobilisation and
avoiding the abortive cost of planned maintenance. In addition, early fault
forecasts will improve operational practices and decrease the reliability of backup
generators. Lastly, P4R will enable more proactive maintenance to protect and
increase the lifespan of assets through the understanding of weather-related
impacts.

These benefits have not been measured so far due to the limitations of the
Discovery phase. The proof of concept in the Alpha phase will help us introduce
operational metrics of Control Room engineers and the response team to measure
the operational efficiency and consequently savings to the network. However, this
will not be quantifiable until the Beta phase.

Benefits to the Environment
The foreseen environmental benefits of P4R will result from reduced CO2 from
avoided unnecessary logistics and diesel generators as well as reduced scope II
and III emissions from purchasing new equipment and materials due to prolonged
life of assets, With the operational metrics defined during Alpha we will have a
more accurate estimation of associated CO2 reductions.

Other benefits

We have also foreseen additional societal and cross-sectoral benefits that are not
quantifiable. Including :

• Embedding learnings in the continuous development of SPEN's Control Room
  and Resource Dispatch (transferable to T&D Asset Owners, and the Oil and
  Gas Industry)
• Providing similar services to other utilities operators and cross-sector
  infrastructure operators
• Potential reduction of H&S incidents and staff exposure to extreme weather
  hazards

The P4R WFS is closely aligned with the six aspects of National Infrastructure
Commission (NIC)'s proposed Resilience framework: anticipate, resist, absorb,
recovery adapt and transform.