Project Summary
As DNOs develop their distribution system operator functions, the current annual
process used to forecast load at extra-high-voltage/high-voltage needs to become
increasingly granular, at the monthly, weekly, daily and hourly level, to support
flexibility dispatch and defer or avoid reinforcement. Moreover, the increasing
prevalence of low-voltage monitoring data enables new use cases to support
network planning and the extension of flexibility markets at ED3. The Artificial
Forecasting project addresses these unmet needs by building innovative AI
solutions to expand load forecasting capability at primary (EHV-HV) and
secondary (HV-LV) substations, optimising flexibility procurement and enabling
DSO functions across the sector.
Innovation Justification
How does your Project continue to demonstrate novel and ambitious
innovation in energy networks? Why is it suitable to be funded by SIF rather
than other sources?
Challenge:
Novel technical, process and market approaches to deliver an equitable and
secure net zero power system. The project addresses theme 3, effectively
managing peak demand and stability through increased flexibility, by developing AI
solutions to optimise flexibility procurement. By sharing information with FSPs, it
aims to stimulate growth in DNO-procured flexibility, providing increased
opportunities for flexibility tenders to manage peak demand. It satisfies the
eligibility criteria on funding and contribution/timelines, including an FSP with
direct customer interactions (EV.Energy) and technology provider/significant
operator of (commercial) assets that provision flexibility (Oaktree).
Lessons from previous innovation:
Throughout Discovery/Alpha, project partners built a strong understanding of the
forecasting solution requirements in the context of flexibility service definition and
procurement. Lessons have included how best to overcome data quality
challenges, methodology selection and optimising the route to deployment.
DNOs have carried out AI innovation work (NIA_ENWL_020 AI and Machine
Learning, UKPN NIA_UKPN0070 Envision, and SSEN_0 Transition). Engaging
with peer DNOs highlighted barriers in scaling these concepts, the urgent need to
integrate solutions into flexibility decision-making, and the opportunities for novel
machine-learning techniques.
Innovative aspects:
Novel forecasting methods: The models developed in this study (e.g. Temporal
Convolutional Neural Networks) are novel relative to existing studies, and have
been selected based on data needs accompanied by domain-specific
performance metrics. These can deliver significantly improved accuracy relative
to existing methods.
Load disaggregation: The solution undertakes a novel approach to forecasting
HV load, separately modelling gross demand and distributed generation.
Existing studies treat net demand as a single time-series, or estimate distributed
generation based on simple heuristics; methods that will become inaccurate as
distributed generation grows.
DNO/FSP coordination: Sharing forecast outputs externally with FSPs is an
innovative initiative providing market participants with the information to stack
their capabilities; this could also allow FSPs to develop innovative offers that
factor in e.g. carbon impact versus peak load reduction.
Scalability: The techniques developed are scalable across all primary and
secondary substations; extending beyond the limited PoCs in previous work
(e.g. SSEN Transition included c.10 substations). This is particularly impactful at
LV where flexibility procurement and forecasting capabilities are more nascent.
Stakeholder Engagement:
Success of this solution has been reliant on stakeholder input. Together with DNO
SMEs, FSP partners have informed how technical specifications can meet the
needs of market participants (e.g. performance metrics). At Beta we will engage
with DNOs through the Energy System Catapult Forecasting Forum and will
continue to engage with European partners at CIRED. To optimise sharing of
forecast outputs and ensure requisite user testing, we plan to engage FSPs
through NPg's in-house flexibility teams. In 2025, we plan to engage with the
Central Market Facilitator to coordinate the integration of the solution, facilitating a
wider industry rollout.
Current TRL/IRL/CRL:
TRL5; significant step in technology validation from testing multiple forecasting
models.
IRL3; increased due to Alpha activity improving understanding of solution
integration.
CRL5; progress in understanding the value proposition in Alpha.
Size and scale:
Conducting work with the existing partner set over two years provides the rigour
required for a deployment of a robust forecasting solution while minimising time to
impact.
Funding:
The approaches developed in this project are being developed at a scale, with an
inherent risk of failure, that are incompatible with BAU resources. Moreover, the
opportunity to collaborate (via SIF) with partners who are leaders in flexibility
service provision and data science enables rigorous solution development, over
and above BAU.
Counterfactual solutions:
The alternative to this approach, i.e. expanding existing approaches cannot deliver
the requisite scalability, given the extent of human interventions required and
frequency of forecasting necessary to optimise flexibility procurement.
Benefits
Pre-innovation baseline: Our solution represents a critical enabler of NPg's DSO
functions, providing new capability over and above business-as-usual (BAU). The
baseline therefore reflects BAU expenditure items set out in NPg's ED2 DSO
strategy from 2024-2028. BAU expenditure is phased using the rate of asset
investment.
To estimate ED3 BAU expenditures, a conservative uplift of 5% is used to
estimate aggregate expenditure relative to ED2 for non-reinforcement items. A 5x
aggregate multiplier is projected for EHV/HV and HV/LV reinforcement costs in
ED3 relative to ED2. This reflects the considerable interventions required as the
network evolves over the next decade.
The baseline therefore covers the period 2024-2033, for which reasonable
foresight can be provided, and includes line items associated with:
Network reinforcement costs
DSO strategy investment in skills and systems
Benefits/Cost savings from DNO-contracted flexibility
Investment in flexibility market stimulation
Tracking: Baseline expenditures/savings through ED2/ED3 act as an aggregate
metric to assess realised benefits. For network reinforcement and flexibility
decisions, NPg also assesses the relevant opportunity costs (cost of the next-best
option) at the time of procurement, which can enable tracking of costs/savings
attributable to the Artificial Forecasting solution (e.g. when used to refine dispatch
windows).
The core technical performance of AI load forecasts will be benchmarked using
metrics including mean absolute percentage error, incorporating learnings through
automated retraining. An intangible success criteria represents their adoption and
trust by NPg within flexibility procurement decision-making, following extensive
user testing at Beta.
Quantified benefits to date (Option 1):
i) Reduced costs of operating the network
Reinforcement: Accurate primary and secondary load forecasts enable more
efficient flexibility procurement and targeted reinforcement spend, estimated at
3% of EHV/HV and HV/LV reinforcement costs at ED2, and 6% at ED3 (derived
from prior efficiencies observed from the implementation of PI historian) (c.
£65m total 2024-33, pre-discounting).
Flexibility: The forecasting solution enables dynamic, semi-automated
procurement, with revision of flexibility service windows closer to the time of
use. Semi-automation combined with stimulation of wider FSP participation is
projected to reduce costs/increase savings delivered through DNO-contracted
flexibility by 25% (c.£10m).
ii) Avoided costs
Resources: Expanding current manual approaches in the absence of integrated
AI forecasting solutions would require considerable labour costs, estimated at
£0.5m annually conservatively. In practice, the resources required to expand
existing methods on a substation-by-substation basis at EHV/HV and HV/LV
could be considerably higher.
iii) New revenues for FSPs:
By increasing deliverability of flexibility contracts, and by reducing barriers to
procurement, the solution can stimulate wider growth in use of market-based
solutions. Based on the proportion of flex-suitable sites identified through NPg's
community DSO project, with assumptions on maximum demand, attribution
rates and unit prices, c.£1m of additional FSP revenue would be attributed to
this solution.
iv) Carbon benefit:
Flexibility not only implies load reduction, but many providers such as Oaktree
Power target emissions optimisation, not simply cost. Using the volume of flex
attributed to this solution in iii), GHG conversion factors and estimating a 10%
attribution, c.700 tCO2e would be avoided in aggregate until 2033, attributable
to this solution.
This will be supported by qualified benefits including the evolution of flexibility
products (e.g. 30-min granularity allows for better programme stacking), together
with wider AI/data readiness.
The whole-life NPV is c.£60m for Option 1. Option 2 considers wider rollout to 3
additional DNOs following completion of the Beta Phase (whole-life NPV c.
£250m).
Benefits realised to date include a centralized understanding of HV/LV
monitoring data, and how AI solutions can be applied to network decision-making.
By examining readiness for this solution, NPg have gained significant learnings
regarding 'live' data availability, quality and deployment readiness.
