Customer story · Energy
Energy grid operators
A load forecasting copilot for energy grid operators
Probabilistic load forecasts with an explainable copilot. Operators see ranges rather than a single number, ask questions in plain language and approve any recommended action.
Executive summary
The situation
Grid planning needs to account for uncertainty, weather and expected peaks. Operators must also be able to question a forecast when data or model behaviour changes.
What was built
Grid load and weather inputs feed P10/P50/P90 forecasts, peak alerts, drift detection and degradation monitoring. Explainability and a chat copilot support questions and what-if scenarios, with local models as an option.
Why it matters
Makes uncertainty and forecast drivers part of the operating conversation. Monitoring surfaces when review is needed; recommended actions wait for approval rather than executing on the copilot’s authority.
- Primary users
- Grid operators
- Core systems/data
- Grid load, weather and forecasting models
- Human control point
- Operator approval of recommended actions
- Nexus role
- Forecasting, monitoring and approval routing
The challenge
A single forecast number hides the risk.
Expected load matters, but so does the uncertainty around it. Weather and changes in the input data affect how a forecast should be interpreted. An expected peak calls for context, not only a number on a chart.
Operators need to examine forecast drivers and recognise when drift or model degradation warrants review. The task was to bring prediction, explanation and monitoring together while leaving decisions about operational actions with people.
What we built
Forecast ranges, monitoring and a copilot.
Nexus produces probabilistic forecasts from grid load and weather data, with P10, P50 and P90 estimates rather than only a single point. Peak alerts highlight periods warranting attention. An explainability layer shows the factors behind the output, and a chat copilot lets operators explore questions and what-if scenarios in plain language.
Drift detection and degradation monitoring sit alongside the forecast pipeline. They surface when the setup needs review; they do not autonomously change the model. Recommended operational actions pass through human approval before execution. Locally hosted models can be used for the forecasting and copilot setup.
How the workflow runs
Step by step.
01
Ingest
Grid load and weather data are collected. Probabilistic forecasts are produced as P10 / P50 / P90 ranges.
02
Alert
Expected peaks raise alerts for operators. Drift and degradation checks surface when review is needed.
03
Explain and explore
Operators see forecast drivers and ask the copilot what-if questions. Recommended operational actions wait for human approval.
What the business team sees
Forecasts operators can question.
Ranges sit beside peak alerts and explanatory factors in the operator workspace. The aim is to let operators examine why the answer looks that way, not merely display it. Monitoring adds context to copilot conversations, while an action recommendation remains separate from permission to carry it out.
- P10 / P50 / P90 forecast ranges
- Peak alerts
- Forecast drivers
- Drift and degradation status
- Copilot for what-if questions
How Nexus works underneath
Prediction and permission are different responsibilities.
Nexus connects ingestion, forecasting, alerts and monitoring. Explainability makes the output approachable without removing uncertainty. Drift and degradation signals prompt review rather than autonomous changes to the setup, and the approval gate keeps recommended operational actions under human control.
Architecture
How the layers fit together.
Business team
Operator workspace and copilot
Forecast ranges · alerts · drivers · what-if questions
Nexus orchestration
- 01Data ingestion
- 02Probabilistic forecast
- 03Peak alerts
- 04Drift monitoring
- 05Explainability
- 06Approval gate
Connected systems and data
- Grid load data
- Weather data
- Locally hosted models
Human control and audit
Recommendations wait for operators.
The copilot can recommend operational actions, but execution requires human approval. Monitoring flags review needs rather than acting autonomously.
What this changes
Uncertainty is available for discussion. Forecast ranges provide more context than a single-point estimate.
Drivers can be examined. Explainability and what-if questions can help operators evaluate forecasts before acting.
Monitoring prompts review. Signals surface when attention is needed, while people retain authority over changes.
Your process
Discuss a similar workflow
Tell us about the documents, systems and decisions involved. We will discuss where Nexus orchestration fits and where your team stays in control.