The expanding and interconnected electricity network poses challenges for National Grid Electricity Transmission (NGET) in monitoring and assessing network resilience. Currently, tracking resilience involves multiple metrics on separate Power BI pages, making it difficult to view overall resilience and hindering timely decision-making during storms or major events. This fragmented approach can lead to suboptimal maintenance and repair decisions. To address these challenges, the proposed project aims to incorporate probabilistic risk analysis to further improve resilience assessments, enhance decision-making accuracy using risk scores, and identify weak areas in the network. Additionally, simplify monitoring through a holistic map view. Ultimately, the project seeks to improve network resilience by streamlining risk identification and establishing minimum resilience levels for substations. FRAME II will be the second phase following the Proof-of-Concept (POC) phase in NIA2_NGET0075 Framework for Risk Analysis and Modelling of Events (FRAME I) .
Benefits
- Facilitates faster decision-making during network events and extreme weather conditions.
- Enables more accurate decision-making by utilising holistic resilience information.
- Supports improved investment justification and asset management decisions.
- Enhances overall network resilience by identifying weak areas and establishing minimum resilience levels for substations.
- Simplifies the identification of network areas at increased risk during storms, particularly due to weak Delayed Auto Reclose (DAR) functionality.
Learnings
Outcomes
- Delivered a fully integrated multi-layer risk modelling framework combining empirically derived machine-learning models with expert-defined resilience rules to generate calibrated asset-, circuit-, and substation-level risk scores. This enabled consistent, repeatable, and transparent risk assessment across the transmission network, supporting NGET in prioritising interventions and targeting resources towards the highest-risk areas.
- Developed and validated advanced predictive models for both ground assets and OHL circuits, using tree-based machine-learning techniques trained on historical faults, failures, and technical limitations, enhanced with Bayesian post-hoc calibration to align predictions with real-world operational conditions. This provides NGET with improved foresight of emerging asset risks and supports more proactive maintenance and asset management decisions.
- Implemented a comprehensive substation resilience and consequence scoring methodology. This aggregated asset-level risk through hierarchical grouping (asset, group, circuit, substation) and incorporated asset criticality, group failure effects, substation design, DAR status, and demand to produce interpretable 1–5 risk and consequence scores. This enables NGET to identify substations with the greatest resilience challenges and better prioritise resilience improvement programmes.
- Deployed a secure, web-based analytical and visualisation platform that transforms complex modelling outputs into intuitive, user-oriented network and substation views. This enables engineers and planners to rapidly identify high-risk substations, improve visibility of network resilience issues, and support evidence-based operational and investment decision-making.
Technology Readiness level (TRL)
Although the project initially targeted completion at TRL 8, internal capacity and expertise constraints meant that additional time was required to progress the proof of concept (PoC) into a minimum viable product (MVP), particularly for the development of the user interface (UI) and user adoption activities. As a result, the project concluded at TRL 7, having successfully demonstrated the technology in a live operational environment and proven its functionality under real network conditions. Progression to TRL 8, including full transition into Business as Usual (BaU), is planned over an additional year and will be achieved once the technology is fully qualified, deployment-ready, and supported by the necessary processes and operational arrangements for routine use.
Lessons Learnt
- The empirical risk modelling approach shows strong potential to predict asset-level risk and subsequent substation resilience, demonstrating the potential for using Machine Learning to identify risk and supporting operational decisions. Adoption and tracking over a longer term will be needed to refine accuracy, validate predictions against real-world outcomes, and fully realise its benefits for proactive risk management. Class imbalance for the problem requires careful handling. Rare failure/technical limitation events necessitated under-sampling and Bayesian calibration techniques, which were used to aid modelling and align predictions more closely with real-world operating conditions.
- Despite challenges with dataset imbalance, the models effectively prioritised rare but critical failure events. The use of Bayesian calibration proved valuable in aligning predictions more closely with real-world operating conditions. A hybrid approach is beneficial - Machine Learning outputs can be enhanced with engineer-provided operational resilience nuances; engineering rules anchored model outputs in practical network behaviour.
- The development of FRAME highlighted opportunities to improve data consistency and quality. Key risk indicators, such as asset age, weather metrics, and historical events, were successfully identified. Addressing issues like inconsistent asset identifiers, maintenance records and naming conventions will unlock even greater predictive accuracy and operational insights.
Dissemination
Dissemination was carried out through an oral presentation at the Risk and Resilience Day in Edinburgh on 19 March 2026
Risk and Resilience Day | March 2026 event in Edinburgh
