One Grid Model, Better Decisions: A Semantic Backbone for Grid Operators, Vendors and Integrators

Enlit Europe meets in Vienna this November under three words: integration, resilience and competitiveness. For a grid operator that means connecting more renewable generation, flexible demand and AI data centres, and keeping the system stable, with lines, substations, and control systems that are largely already built. The European Commission puts the grid investment Europe needs this decade at 584 billion euros, and obtaining permits for high-voltage grid projects can take eight to ten years before construction starts. Until new lines are built, the green transition runs on the grid that is already there.

Europe’s grid must carry more, and explain more

Most of the answer is physics, hardware and investment. A growing part is information: knowing, at the moment of decision, what an asset is, what it is connected to, which limits and obligations apply to it, how much the weather lets it carry, which flexibility, from home batteries to EV charging, can relieve it, and what changes if it trips.

The final report on the Iberian blackout of 28 April 2025, published by ENTSO-E’s expert panel in March 2026, shows what happens when that information is missing. Many renewable plants disconnected before reaching the voltage thresholds set at their grid connection points, some with overvoltage protection set below the regulatory limits. Incomplete data kept investigators from establishing the cause of some disconnections, and the panel’s 22 recommendations include standardised data collection so that future investigations find complete and consistent records.

The settings, the obligations, the network models, and the measurements all exist. They sit in different systems, described in different words, and cannot be joined fast enough when it counts. After three decades on data in oil and gas, I recognise the problem. The grid starts better placed: transmission operators already exchange their models in a shared language every day.

A semantic backbone: one description of the grid that every application reads

Grid operators already share a language. The Common Information Model (CIM) describes networks, assets, and market data, and the Common Grid Model Exchange Standard (CGMES) is the part of CIM that European transmission operators use to exchange grid models. ENTSO-E publishes CGMES in open web standards, with most of the rules for a valid model alongside the definitions.

A semantic backbone puts that language to work across the business. It holds the grid model, asset data, documents, and links to measurements in one connected description, built on the published standards and extended with the operator’s own terms. Measurements stay in SCADA and the time-series historians, and transactions in the business systems. The backbone holds their definitions, how they connect to the grid model, and which system keeps each. Applications read the same meaning, rules are checked before data moves on, and every answer can be traced to the facts and the rule behind it.

The backbone answers in two directions. Forward, for engineering: from a planned change to everything it affects, such as a new protection setting, a revised line rating, or a connection offer. Backward, for operations: from an event to its cause, such as an alarm, a failed model merge or an unexpected disconnection.

Graphwise brings it to three parts of the grid business first:

• Grid model exchange. It loads CGMES as published and checks each model against the rules before it is submitted.
• The grid’s digital twin. It connects the network model with assets, documents, and measurements, so engineers can see what a planned change affects and operators can trace an event back to its cause.
• Safe and optimal operations. AI assistants can help operators keep the grid secure while using its full capacity, as long as each action they propose is checked against the operator’s rules first. The groundwork is plain-language access to the grid model, which Graphwise worked on with Statnett, Norway’s transmission system operator, in the Talk2PowerSystem project.

It can come built into software the operator already buys, through an integrator, or from us directly.

Six decisions a semantic backbone supports

The backbone serves planning, operations, and the market desk. Graphwise’s grid work so far concentrates on the first three; the other three build on the same model. Distribution operators meet the same decisions in connection requests, flexibility, and local congestion.

A backbone does not compute power flows, clear markets, or forecast wind. It gives those calculations a model they can trust, and it records which model version, rating, and limit were in force when a decision was taken.

Grid model exchange 

Every model contributed to the common European grid model must pass the published rules, and each new version of the standard reopens the work. With validation switched on, GraphDB, Graphwise’s semantic graph database, checks each change as it is committed, not only the whole model. Versions of the standard can sit side by side, so a new release is loaded as data next to the old one. Graphwise has set out the approach for CGMES 3.0 in an earlier post.

A semantic digital twin

A digital twin shows what is happening; a semantic twin adds what things are, what depends on them and which rule applies. Without that, many twin programmes stall after the pilot: the model underneath cannot answer a new question.

A transmission operator has done this with GraphDB: an electric grid assets knowledge graph that sets sensor, monitoring and maintenance data against the network on CIM, with IFC and ISO 15926-14 beside it, and surfaces dependencies none of those systems held alone. The same questions then run both ways, forward to what a fault threatens and back to where it began.

A semantic twin can also hold the procedure for changing a protection setting as steps, with their equipment, conditions and limits, so a proposed change is tested against the rules before anyone signs. For the vendors of twin platforms, their twin plus a semantic backbone is a semantic digital twin.

Safe and optimal operations 

In the Talk2PowerSystem project with Statnett, engineers ask a grid model in plain language instead of writing graph queries. The platform can show the query it ran and the data it used, which is what makes an answer checkable. That is the groundwork for assistants in the control room. Before an assistant acts on what it finds, it needs what a switching programme already has: a check against the rules, a named person where the rules require one, and a record of what was done. The backbone keeps those rules in the same model as the grid, so an agent from any vendor meets them in one place. Where an action cannot be taken back, a plausible answer is not a check.

Asset and connection compliance

Protection settings, ride-through capability (staying connected and operating through a voltage dip or surge), and inverter firmware sit with different parties in different systems, while the obligations sit in connection agreements and grid codes. The backbone extracts the requirements from the documents, links each setting to its asset, connection point and obligation, and flags a setting outside its limit before an event. Graphwise offers this as compliance intelligence.

More from existing lines 

Dynamic line rating lets a line carry more when the weather allows; National Grid reports an average gain of 8 per cent in capacity. The rating system calculates the figure. The backbone puts it on the right line in the shared model, with its source and time, so planners, operators and neighbouring operators all use the same number.

Congestion, balancing, and trading 

An optimiser works with the model it is given. CIM also has a market part, whose bid and activation documents identify the resource behind them, so each can be linked to the asset and its place in the network. Trading desks keep their own copies of counterparties, delivery points and units, so several versions of one fact coexist. A backbone gives each one identity, with its source, owner, and permitted use, before an AI agent works with it.

What the grid can take from oil and gas

Oil and gas has worked on the same problem for thirty years, and I worked on it from the operator side, in research, and at a software vendor. From 2009 to 2012 I managed IOHN, Integrated Operations in the High North, in which twenty-two oil companies, suppliers, research bodies and industry associations built shared ontologies for drilling and operations on ISO 15926.

Offshore, the two industries already meet: platforms on the Norwegian shelf run on power from shore, and a floating wind farm, Hywind Tampen, supplies about 35 per cent of the power the Snorre and Gullfaks fields use. Four lessons carry to the grid.

People have been the bridge between systems

Grid control rooms work across energy management, outage, geographic, metering, and asset systems that each keep their own model of the network, often with their own identifier for the same transformer and their own name and value for its rating. A control centre for wind farms often connects to a different control system at each farm, usually the turbine supplier’s. Operators and engineers make the links, and that work grows with every wind farm, solar park, and battery that connects.

Twenty years ago oil and gas brought its disciplines together in integrated operations centres onshore, but the systems stayed apart, so in an upset everyone went back to their own desk and came back with a partial answer. An oil company at least owns the systems on its installations, legacy control systems included. A grid operator does not own those of the plants that connect to it, so the description is what has to be shared.

A person reading a badly labelled record works out what was meant; an AI agent acting on it does not. What people knew has to be written down where the agent can read it.

An operator that owns its model can test any assistant on it

A grid operator owns its network model, so an AI assistant that reasons over it can prove itself on the operator’s terms: put a fault into the model and see whether the assistant finds it. On the Tyra gas field redevelopment, the operator I worked for had that means of checking a control room assistant, because the plant’s full dynamic model came with the construction contract and belonged to the operator, not to the assistant’s supplier.

Meaning is lost at every handover

The grid hands over constantly: from planning to operations, from one operator to its neighbour, and from a plant’s owner to the grid operator each time a wind farm, solar park, or battery connects. The blackout report shows what is lost when those handovers drop the facts. The subsurface taught me the same: a reservoir model loses its assumptions as it passes from geophysicist to geologist to engineer, and closed-loop reservoir management, which keeps measurements, models and decisions in one loop, was built to win them back.

Meaning outlives the platform if someone keeps it, and here electricity starts ahead

When IOHN closed, its ontologies survived because a standards association kept maintaining them. In oil and gas, such standards are voluntary. The European network codes oblige transmission operators to supply grid models for a common European model, and those models are exchanged in CGMES. The grid’s shared language is already written down and in daily use. What remains is to use it beyond model exchange, and to name who keeps it.

For architects: the semantic knowledge graph inside the backbone

At its core, the backbone is a semantic knowledge graph. Every fact is stored as a connection between two things: this breaker sits in this substation, this line runs between these two substations, this rating applies under these conditions, this setting came from that commissioning report. An ontology defines what each kind of thing and each connection means, and for the grid that ontology already exists in CIM and CGMES. Vocabularies and taxonomies hold the operator’s own terms and map them to the standard.

That structure does four things that are hard to get from tables or document stores:

• Meaning travels with the data. Every fact points to the definition it follows, so two systems, or a person and an AI agent, read it the same way.
• The graph derives what follows. If a breaker sits in a substation and the substation in a region, a question about the region finds the breaker without anyone listing it.
• Rules are checked as data arrives. The constraints ENTSO-E publishes in SHACL for CGMES run against the graph, and an operator can add its own.
• Every fact keeps its source and time. An answer can show where it came from and which version of the model it used.

ENTSO-E publishes the CGMES vocabulary in RDFS and most of its constraints in SHACL, and grid models are exchanged in CIMXML, a restricted form of RDF/XML. RDF, RDFS and SHACL are the W3C standards that semantic knowledge graphs are built on, and Graphwise works natively in them. The standard and its constraints therefore load as published: there is no mapping of the standard to build, and none to rework when a new version arrives.

Connecting the operator’s own systems to it is still work, done once against the published model. Applications query the graph with SPARQL, the W3C query language.

For AI, the graph is the context an agent works in. GraphRAG lets an assistant answer from connected, defined facts, not only from passages of text. GraphDB has a built-in server for the Model Context Protocol (MCP), an open protocol for connecting AI agents to data, so agents from any vendor that supports it reach the same facts.

Building it in: for software vendors and integrators

For the companies that make grid software, building the semantic backbone in changes four things:

• New versions of CIM and CGMES arrive as data, so keeping up needs less engineering.
• Conformance is checked inside the product.
• AI features can show how an answer was reached, which operators of critical infrastructure ask for.
• Engineering time stays on the grid.

Grid platform vendors increasingly sell their suites on a federated data foundation. Three questions decide whether the meaning in such a foundation belongs to the operator, and we put them to ourselves:

• Can a constraint change without a software release? In the Graphwise platform, the rules are data, so yes.
• Does the meaning leave with the operator? It is written in open W3C standards, so it does.
• When the person who must approve an agent’s action says no, does the action stop? Only if the check sits outside the agent; the backbone holds the rules and approvals such a check reads.

Graphwise already runs inside other companies’ products: Johnson Controls selected GraphDB for semantic data creation and management in Metasys, its building automation system, integrated since the launch of Metasys 11 in 2020. Energy software vendors can work with us as technology partners, building the backbone in or offering it beside their products; integrators and consultancies as solution partners. For integrators, the standard and its rules are the same in every grid project, so methods and tooling built for one operator carry over to the next.

The Graphwise Platform runs on the operator’s own infrastructure or chosen cloud. Graphwise is certified to ISO/IEC 27001 for information security and ISO/IEC 42001 for AI management, and has achieved SOC 2 Type II compliance.

Where to start

Nobody sets out to buy a knowledge graph. They set out to fix something. For an operator, start with one question the business asks often and answers slowly: which connection points a new protection setting affects, why the last model merge failed, which grid users sit behind a congested line. Bring it on your own, or with your software supplier or integrator. We answer it on the relevant part of your model, or a public test model, with and without the backbone. Graphwise’s AI Accelerator Program is designed for this step.

For a software vendor, start with the roadmap: which version of the standard comes next, where customers ask for explanations, and what building the backbone in would change. Talk to us about partnering.

Bring your question to stand B.b82 at Enlit Europe in Vienna, 10 to 12 November 2026, or send it to us before then.