Network
- CIM
- CGMES
- Network topology
- Substations
- Lines
- Transformers
- Bays
- Connectivity
NeuroCluster Labs · Energy & grid research
Foundation models for the European power grid.
Grid Foundation Models
Research programme·In developmentEurope's electricity system is becoming more interconnected, distributed and dynamic.
- Renewables fluctuate.
- Demand electrifies.
- Power flows become bidirectional.
- Infrastructure ages.
- Connections increase.
- Congestion grows.
The intelligence operating the grid must evolve with it.
NeuroCluster Labs is researching sovereign foundation models designed specifically for Transmission System Operators, Distribution System Operators and the European power system.
Why grid foundation models
A power grid is a living system.
Today's AI models primarily learn relationships between words and images.
AssetthenTopologythenElectrical statethenWeatherthenConstraintthenForecastthenInterventionthenOutcome
That requires more than an LLM.
It requires models capable of understanding the grid as a changing physical network.
What the model can observe
The grid speaks in many data languages.
Real-time operations
- SCADA
- EMS
- PMU / WAMS
- Alarms
- Events
- Measurements
- Switching state
Assets
- EAM
- SAP
- Asset registers
- Maintenance history
- Failures
- Condition data
- Work orders
Environment
- Weather
- Vegetation
- Temperature
- Wind
- Lightning
- Flood risk
- Geospatial context
Inspection
- RGB imagery
- Thermal imagery
- LiDAR
- Drone inspection
- Satellite imagery
Planning & market
- Generation
- Demand
- Market data
- Outage schedules
- Connection requests
- Capacity forecasts
Knowledge
- Operating procedures
- Engineering standards
- OTL
- Taxonomies
- Lessons learned
- Incident reports
The research model family
Different models. One shared representation of the grid.
Try a question
Components active
7 / 7
Illustrative routing. The programme researches how a grid question can activate only the specialised components it needs.
GridGraph
Learns network structure, connectivity, asset relationships and topology.
GridTime
Learns temporal behaviour from measurements, events, demand, generation and weather.
GridVision
Understands visual observations of grid infrastructure and relates defects to the correct asset.
GridLanguage
Understands procedures, engineering documents, work orders, incidents and technical knowledge.
GridReasoner
Combines grid context with calculations, rules, simulations and evidence.
GridAgent
Interacts with approved tools and enterprise applications within defined authority boundaries.
GridWorld
Research into models that represent grid state and evaluate possible future states following operational interventions.
Shared grid representation
Not isolated AI applications — together, a foundation intelligence layer for the grid.
The grid representation
From fragmented systems to a machine-readable grid state.
Fragmented systems
SCADA / PMUplusCIM / CGMESplusGISplusSAP / EAMplusWeatherplusInspectionplusMarketplusDocuments
Grid ontology
- Asset
- Terminal
- Substation
- Connection
- Constraint
- Measurement
- Event
- Inspection
- Work order
- Weather
- Risk
Grid state
- Q1What exists?
- Q2What is connected?
- Q3What is happening?
- Q4What changed?
- Q5What is constrained?
- Q6What may happen next?
Grid models
Graph · Time series · Vision · Language · Reasoning
Operational intelligence
Understand · Predict · Simulate · Recommend
Operator authority
Review · Approve · Reject · Escalate
Controlled action
Operations
Understand what is happening now.
Potential research and application areas include:
- Now
- Minutes
- Hours
- Days
- Months
- Years
Operational situation awareness
Combine network state, events, topology and operational context.
Event correlation
Understand whether multiple signals represent independent events or one developing system situation.
Congestion intelligence
Connect forecasts, topology, weather and network constraints.
Contingency support
Assist engineers analysing potential system states and validated contingencies.
Outage restoration
Bring relevant topology, assets, procedures and operational evidence together during restoration planning.
Operator assistance
Retrieve and structure evidence without removing operator authority.
Planning
Understand what may happen next.
- Now
- Minutes
- Hours
- Days
- Months
- Years
Load forecasting
Short-, medium- and long-term grid demand.
Renewable generation forecasting
Weather-aware generation expectations.
Capacity intelligence
Understand where available capacity exists and why.
Connection impact analysis
Support assessment of new generation, storage and demand.
Network reinforcement
Compare future scenarios and investment options.
Cross-border intelligence
Analyse interconnected network and system context.
Asset intelligence
Let the grid model understand the physical infrastructure.
Asset condition should not be a single field in an asset-management system.
It can be constructed from:
- Age
- Operating history
- Telemetry
- Inspection
- Environment
- Previous failures
- Maintenance
- Network criticality
= Composed
Asset condition estimate
Composed from every input, and traceable back to each of them.
This enables research into
- Asset health estimation
- Predictive maintenance
- Defect detection
- Failure-risk modelling
- Maintenance prioritisation
- Vegetation risk
- Corrosion detection
- Thermal anomalies
- Infrastructure ageing
Dynamic capacity
Static limits describe infrastructure. Intelligence describes its current capability.
Weather, asset condition, network topology and system state can influence available capacity.
Grid foundation models can support research into:
Dynamic Line Rating
WeatherthenConductorthenThermal statethenAvailable capacity
Topology optimisation
Network statethenConstraintsthenAlternativesthenSimulated impact
Flexibility
DemandthenGenerationthenStorage
Flexibility evaluated against network conditions.
Constraint forecasting
Predict where and when system constraints are likely to emerge.
AI.grids and the European ecosystem
Designed for an emerging European grid-model ecosystem.
Europe is now developing sovereign AI foundation models specifically for power-system operations through initiatives including AI.grids.
NeuroCluster Labs supports the same direction: open European research, interoperable datasets, sovereign compute and practical validation with grid operators.
Our focus is particularly on the layer needed to bring those capabilities into operational environments:
Grid ontologyplusModel orchestrationplusEnterprise contextplusPolicyplusEvidenceplusHuman authorityplusControlled action
The goal is interoperability, not another isolated AI stack.
European grid foundation models
Open European research
- Including AI.grids
NeuroCluster grid model layer
NeuroCluster Labs
- Model routing
- Domain adaptation
- Evaluation
- Multimodal fusion
- Grid ontology
NeuroCluster platform
Operational intelligence
- Identity
- Policy
- Evidence
- Agents
- Human authority
- Audit
GridSight
For the energy market
- Operations
- Asset intelligence
- Capacity & planning
TSO / DSO systems
The operator's estate
- EMS
- SCADA
- GIS
- EAM
- Planning
- Maintenance
- Control room
AI.grids is an independent European initiative. Naming it describes the ecosystem this programme is designed for; it does not imply participation or endorsement.
Safe by design
Intelligence can advise. Authority remains explicit.
The programme separates:
Read
Models can retrieve and interpret authorised information.
Requires
- Authorised access
Recommend
Models can produce forecasts, analysis and proposed actions together with evidence and uncertainty.
Requires
- Evidence
- Uncertainty
Act
Actions require explicitly assigned authority and policy.
Requires
- Assigned authority
- Policy
For operationally critical decisions, human authority remains part of the system architecture.
Model evaluation
A grid model should prove what it understands.
Generic language benchmarks tell us very little about operational grid capability. NeuroCluster Labs therefore proposes grid-specific evaluation across areas including:
Topology understanding
Can the model correctly reconstruct and reason about network connectivity?
Temporal reasoning
Can it connect events occurring across different timescales?
Constraint reasoning
Can it identify which system constraint matters to a decision?
Asset reasoning
Can it connect observations and maintenance history to the correct physical asset?
Forecast performance
How accurate and calibrated are predictions?
Simulation consistency
Do recommendations remain consistent with validated engineering tools?
Evidence grounding
Can every material conclusion be traced to its evidence?
Authority compliance
Does the system remain within its assigned operational permissions?
The public benchmark programme these evaluation areas feed into. Industrial Intelligence Benchmark
Sovereign grid AI
Critical infrastructure requires controlled infrastructure.
Designed for deployment across:
01
European sovereign cloud
02
Dedicated TSO environment
03
On-premises AI infrastructure
04
OT-adjacent protected environment
05
Air-gapped deployment where required
Operational data remains within the boundaries defined by the system operator.
The relationship with GridSight
Models understand the grid. GridSight makes that intelligence operational.
- 01
NeuroCluster Labs
Develops and evaluates grid intelligence.
- 02
Grid Foundation Models
Understand network, time, assets, vision and technical knowledge.
- 03
NeuroCluster
Adds ontology, agents, governance and evidence.
- 04
GridSight
Turns intelligence into real TSO and DSO applications.
ResearchthenIntelligencethenDecisionthenControlled action
GridSight — for the energy marketBuild the European grid model with us.
Partners can contribute through datasets, synthetic environments, model research, compute, evaluation or operational use cases.
We are looking to collaborate with
- TSOs
- DSOs
- Universities
- Research institutes
- European AI Factories
- Energy technology companies
- Equipment manufacturers
- Weather and satellite-data providers