Good morning. Which segments are at high risk today and tomorrow?
AGENT · GRID RISK FORECAST
Fault risk on the electricity grid is forecast before the outage event.
Grid Risk Forecast produces daily fault risk forecasts on specific electricity or gas grid segments. It combines forecast weather data, asset maintenance history, asset age, and operating conditions. It identifies the highest-risk segments for the next 24-72 hours.
Grid Risk Forecast at work.
Forecast updated at 05:00. 23 segments at high risk for the 48-hour window: hilly zone (Arezzo and Siena provinces). Pattern: assets with an average age of 28 years plus strong wind Thu-Fri (gusts 80-100 km/h). Proposed allocation: priority 1 (8 segments with overdue maintenance), priority 2 (15 segments with regular history). List in the channel.
Confirmed. I'll activate the field team for the 8 priority-1 segments this morning.
Confirmed. The event is in the audit registry.
Why it exists.
Preventive maintenance of utility networks is a structural operational priority. European electricity networks carry assets that are partly aging; extreme weather events affect infrastructure reliability; outage events impact end customers and carry significant restoration costs. The manual approach to risk management relies on periodic maintenance schedules and reactive response after the event.
How it works each day.
Grid Risk Forecast activates daily. For each grid segment in the customer's network it reads current operating conditions (data aggregated from the infrastructure), the 72-hour weather forecast, the segment's maintenance history, and asset age. It combines the factors using the configured forecast models and identifies segments with fault probability above the configured threshold.
The decision stays with the operations manager.
For high-risk segments: an alert to the operations team with suggested interventions. The intervention decision stays with the grid operations manager under the utility's procedures.
Grid operations manager and maintenance manager.
Grid operations manager
The manager recovers the capacity to intervene preventively on the segments that are actually at risk. Maintenance capacity dimensions on real data, not fixed periodic schedules. Post-event handling decreases in favour of structured prevention.
14-day forecast 48 sections
MV-North section high
MV-East section medium
Assets > 25 years 12 replacements
Weekly maintenance plan
Maintenance manager
The maintenance manager sees the allocation of the field team for the day in a structured way. Intervention priorities emerge from the combined factor analysis, not from manual review of programmes.
SCADA · live 2 anomalies
Substation SE-104 voltage out of range
Line LV-22 current spike
NIS2 classify material · 24h timer
Alert to CISO · ticket opened
Utility operations leadership
The agent is vertical for the utility sector (electricity networks, gas networks). Operations leadership has structured visibility on network conditions and aggregate risk patterns, useful for calibrating maintenance plans.
Ongoing outage ETA 35'
Zone Northeast cluster B
Field team 3 crews · in transit
Customer SLA reset · 20'
Regulator report in preparation · 4h
12,000 grid segments, forecast at 05:00 every morning.
The agent reads 12,000 segments, weather, and maintenance history.
For a regional utility, the agent is scheduled every morning at 05:00. It reads the conditions of 12,000 grid segments distributed across the region. It compares against the weather forecast (strong wind event expected Thursday-Friday), the maintenance history of each segment, and asset age.
23 high-risk segments, ordered by priority.
It identifies 23 high-risk segments for the 72-hour window. Pattern: segments in hilly zones with aging assets and forecast strong winds. The agent notifies the operations manager in the dedicated Slack channel with the list ordered by risk and the proposed field team allocation: visual inspection and preventive reinforcement for priority-1 segments (8 with overdue maintenance history), close monitoring for priority-2 segments (15 with regular history).
The manager validates; the agent records.
The operations manager validates the proposals and activates the field team for the highest-priority segments. The full event stays in the runtime audit registry. Prerequisite: availability of historical fault data from the utility for initial model calibration — calibration requires at least 12 months of event history.
Declarative rules, SCADA and weather services in delivery.
The Grid Risk Forecast rules are declarative. The grid operations team and maintenance manager define in a readable format the risk baselines by asset type (overhead cables, transformers, substations), the forecast factors (weather weight, maintenance history weight, asset age weight), and the alert thresholds. The rules live in the customer's repository, versioned.
Integration with the SCADA system is delivered via a dedicated aggregation layer during the project by the Exelab team. Weather data is retrieved from services configured during delivery. Initial model calibration requires historical fault data from the utility.
- Language
- TypeScript (Node.js)
- LLM model
- customer's choice: Anthropic, OpenAI, Mistral, open source models hosted internally, AWS Bedrock for a private model
- Built-in controls used
- pii-detector, tool-rate-limit
- Native delivery channels
- Slack, Telegram, WhatsApp, OpenAI-compatible HTTP
- SCADA + asset management system integration
- dedicated adapter delivered during the project via aggregation layer
- Weather service
- configured during delivery
- Calibration prerequisite
- availability of utility historical fault data (minimum 12 months)
- Memory
- persistent per instance, pgvector + PostgreSQL FTS on historical fault patterns
- Registry
- append-only, queryable with a standard SQL client
Frequently asked questions about the agent.
No. The agent identifies the high-risk segments and proposes the field team allocation. The intervention decision stays with the grid operations manager under the utility's procedures.
Integration with the SCADA system is delivered via a dedicated aggregation layer during the project by the Exelab team. Technical feasibility depends on the customer's SCADA version and security policies (air-gap or strict segmentation are possible). The specific configuration is defined during discovery.
Initial calibration requires the utility's historical fault data: typically 12-24 months of documented events (fault type, segment, weather conditions at the time). Availability of this data is one of the prerequisites assessed during discovery. The model improves over time as operational data accumulates.
The typical pattern for Grid Risk Forecast is 16-22 weeks. Discovery and historical data collection 3-4 weeks, model and rule configuration with the operations team 4-5 weeks, SCADA and weather service integration 6-8 weeks, calibration and hand-off to the operations team 3-4 weeks.
From a 30-minute conversation to the squad in production.
A 30-45 minute conversation to understand how Grid Risk Forecast would configure to the utility's case. Which network size, which SCADA system, which availability of historical data for calibration.