AI driven operational load forecasting for short-term flexibility procurement
AI-driven operational load forecasting for near-real-time flexibility procurement enables distribution network operators to shift from annual, assumptions-based flexibility planning to daily, data-driven procurement. As distributed energy resources, electric vehicles and heat pumps increasingly connect to low and medium voltage networks, distribution network operators face a growing challenge in transitioning from passive network managers to active system operators.
Flexibility services – in which asset operators shift demand as a paid service – can be a lower-cost alternative to physical network reinforcement. However, because flexibility has traditionally been procured many months in advance, asset operators lack the near-term certainty needed to commit, which limits both participation and the practical value of flexibility for the network. AI-based load forecasting addresses this by ingesting historical load data from network substations, weather forecasts and other input variables to produce day-ahead and week-ahead forecasts at the substation level, including a quantified uncertainty range. This allows operators to define flexibility needs with greater precision and closer to the time of need, when asset operators can more reliably assess and stack their capabilities. It can broaden the pool of eligible participants and make flexibility procurement a more effective operational tool.
The approach also reduces the need for large teams of engineering staff to manually validate data and produce load forecasts, helping automate and scale a process that must expand substantially as low-carbon technologies accelerate connection to low-voltage systems.
Main Outcomes of the Solution
- Affordability: Lower reinforcement and operating costs by enabling distribution network operators to use market-based flexibility as an alternative to physical network upgrades. More accurate forecasting and procurement can improve asset utilization and reduce unnecessary network investment.
- Reliability & Flexibility: Improved visibility of short-term local network constraints, allowing operators to procure and dispatch flexibility where and when it is needed while supporting the integration of low-carbon technologies such as electric vehicles and heat pumps.
- Institutional Efficiency & Transparency: Reduced dependency on labor-intensive manual forecasting processes, enabling distribution network operators to scale forecasting capabilities across increasingly distributed and complex networks.
Key Deployment and Scaling Factors
- Build forecasting models on network-based substation load data rather than individual device-level data feeds, reducing privacy concerns, lowering data acquisition costs and limiting dependency on asset owners sharing granular operational data
- Ensure sufficient data quality, telemetry coverage and weather-data integration, because short-term forecasting depends on reliable historical load data, timely network measurements and relevant external variables
- Design models to produce both a point forecast and an associated uncertainty range, giving flexibility procurement teams the confidence intervals needed to support operational decision-making
- Integrate forecasting outputs directly into flexibility tender, dispatch and settlement processes so that asset operators receive clearer, shorter-horizon signals about when and where services are needed
- Align forecasting-led procurement with regulatory flexibility-market rules, cost-recovery frameworks and system-operator responsibilities, because near-real-time flexibility procurement requires clear rules for when flexibility can substitute or defer reinforcement
- Share model methodologies and learnings across distribution network operators through industry innovation programs to support replication and avoid duplication of development effort
Case study
Northern Powergrid and Faculty (Part of Accenture), United Kingdom
Northern Powergrid, the electricity distribution company for Yorkshire, the North East and northern Lincolnshire, developed the Artificial Forecasting project with Faculty Science (now part of Accenture), Oak Tree Power and ev.energy, with UK Power Networks participating as a collaborating network. The project was funded under the Ofgem Strategic Innovation Fund Round 2 Discovery phase with a £160,000 project budget and ran for three months from April 2023.
The project developed machine-learning models to produce day-ahead and week-ahead load forecasts at extra-high-voltage to high-voltage and high-voltage to low-voltage substation levels. The models used historical network load data, weather forecasts and associated uncertainty outputs to support operational flexibility procurement closer to real time.
The project aimed to replace an annual, labor-intensive forecasting process with a semi-automated capability that could be run more frequently without requiring engineering staff to manually validate data at every substation. Project documentation stated that forecasting coverage would need to expand by up to 50 times compared with traditional approaches to capture the transformation points where low-carbon technology connections are creating new flexibility and constraint-management requirements.
According to project assessments, the approach could support approximately £51 million in net present value reinforcement cost avoidance through 2033 and up to £60 million in operating cost savings over the following decade by improving the targeting and timing of flexibility procurement. The project documentation also estimated approximately 700 tons of CO2-equivalent emissions reductions associated with peak demand reduction enabled by improved forecasting and flexibility dispatch.
The project further explored how near-real-time forecasting could improve flexibility procurement by allowing distribution network operators to define flexibility requirements closer to the time of need, when asset operators can more reliably assess available capacity and participate in flexibility markets.
References:
