Project Summary
This project addresses Challenge 2: "Preparing for a net zero power system", theme 1. Novel ways to reliably support low stability systems.
As DNOs transition to DSOs, the current annual load forecasting process must become increasingly frequent (monthly, weekly, and daily), to support flexibility dispatch. The scope must also extend at least SO-fold to capture HV/LV substations, as low-carbon technologies connect to LV systems. Given networks have typically employed manual/disaggregated approaches to forecast load and
account for the diversified contributions of new loads, novel approaches are required to enable system flexibility and support network stability under these new conditions.
This project will develop innovative Al-based approaches to augment load forecasting capability. In turn, flexibility will become more realistic as a reinforcement option, and the available capacity in the network for new low-carbon loads will expand, increasing the speed, and lower the cost, of decarbonisation.
Specifically, this project will:
1. Test machine learning algorithms to produce load forecasts at EHV-to-HV transformation points, suitable for the shorter-term forecasting DSO systems require.
2. Produce HV-to-LV forecasts, and develop Al techniques for modelling the connection of load.
Capabilities delivered:
This approach will enable:
• The ability to integrate, ex-ante, demands on the network such as EV charging, local PV and heat pumps;
• The understanding needed to promote targeted reinforcement options such as flexibility, moving from annualised to e.g. daily timescales.
• A step change in the scope of load forecasting, needed as the 230/400V system
becomes the focal point of the energy system, without a step change in technical staff requirements.
User needs:
This project would benefit a wide set of users and the system at large, including:
• Connectees of low-carbon load who, given better network capacity information, would see reduced timescales and costs associated.
• Flexibility providers and controllers, who will have a better understanding of the likely value, call-off and effectiveness of flexibility services.
• Network customers who will see lower price pressure, given more effective and efficient flexibility and network reinforcement investment.
• Other electricity distribution companies, who benefit from the knowledge sharing mechanisms inherent to the SIF.
Partners:
• Northern Powergrid is the electricity distribution system company for Yorkshire and the Northeast.
• Faculty Science Limited specialises in the implementation of custom Al systems for critical national infrastructure.
UKPN, the electricity distribution system company for South East England, the East of England and London.
Innovation Justification
As DNOs transition to DSO, the current annual process used to forecast load is required to become increasingly granular, at the monthly, weekly, daily and hourly level to support flexibility dispatch. Additionally, this process will need to be extended at least 50-fold to capture HV-to-LV transformation points as low-carbon technologies connect to LV systems.
At present load forecasting is labour-intensive requiring engineering input to understand the validity of data and spot outliers. Novel techniques are required to allow machines to undertake this work. Load modelling has traditionally been done either using numerical methods based on engineering models, or using an estimate for a typical network load mix built up over many years. As new loads arise, flexibility introduces new probabilistic elements and the load mixes change rapidly. Moreover, given this occurs at different rates across different networks, existing methods become less accurate. This project is ultimately experimental in nature (given the nature of data science), however the availability of data from network substations provides every opportunity for this to represent a successful Al use case.
Other attempts to improve load forecasting, like UKPN's Envision (NIA_UKPN0070), have endeavoured to use dataflows from individual low carbon technology installations and the information systems associated with them. These are making some progress, but are presently more cumbersome and costly relative to a centralised solution using network load information. Our 'network based' solution also minimises data privacy issues. We intend to contrast the results of our approach with those of UKPN's Envision, a potentially higher accuracy, but more data hungry and granular, approach.
The method proposed avoids using circa 50 dedicated engineering staff per DSO to undertake load forecasting for flexibility relative to traditional methods. Such staff are both expensive and scarce. It should also be significantly less expensive than acquiring data feeds for all low carbon technology installations and avoids the uncertainty associated with individuals and organisations withholding the data at some future time
Project Benefits
The benefits selected in Benefits Part 1 (Q5) were:
• Financial - future reductions in the cost of operating the network
• via fewer staff required to forecast and model than the counterfactual
• via lower and better targeted flexibility payments
• Financial - cost savings in consumers' network part of the bill
• via lower and better targeted network reinforcement requirements leading to lower DUoS bills
Referencing the benefits selected in Benefits Part 1 , the metrics and indicative quantitative measurement with their associated timeframes are:
• Metrics:
1. Staff employed in load forecasting - currently circa £2m pa expenditure
2. Level of flexibility payment per kilowatt reduction and magnitude of kilowatt reduction required
3. Annual connections reinforcement and discretionary reinforcement expenditure
• Measurements (values based on NPg; GB distribution values would be circa 7x NPg's):
1. Reduction in ED2 opex by up to c.£2m pa (value realised from time of transfer to BAU and will be ongoing)
2. Neutral or slight reduction in flexibility (opex) payments relative to ED2 forecasts in DSO sections (value and timing dependent on evolution of flexibility market)
3. 5-10% reduction in reinforcement capex - ED2 reinforcement forecasts of
£348m would give c.£17-34m over a given 5 year period (actual value realised from time of transfer to BAU but will be ongoing)
