Example Scenarios

How utilities apply ML Playground guides to solve real grid problems.

These scenarios illustrate the practical path from completing ML Playground guides to deploying models that improve grid operations. Each shows how a utility's existing domain expertise, combined with the guide's ML techniques, leads to measurable operational improvements.

Note: These are representative example scenarios based on common utility challenges and realistic outcomes. They illustrate the type of results achievable when applying ML Playground techniques to utility data.

Midwest Municipal Utility Reduces Outage Duration by 35%

Municipal Utility ~85,000 Customers Guide 01 + Guide 04 3 Months to Pilot

Challenge

A Midwest municipal utility experienced frequent weather-driven outages but relied entirely on reactive dispatch. Crews were deployed after outages were reported, leading to long restoration times during storm events. The reliability team knew which feeders were problematic but had no systematic way to predict where outages would occur before a storm hit.

Approach

Two distribution engineers completed Guide 01 (Outage Prediction) using SP&L data, then adapted the Random Forest model to their own historical outage and weather data. They used Guide 04 (Predictive Maintenance) to identify high-risk transformers for proactive replacement.

  • Trained a weather-based outage prediction model on 5 years of outage history
  • Merged transformer age and condition data to prioritize equipment replacement
  • Built a daily "storm readiness" dashboard consumed by the dispatch team

Results

  • 35% reduction in average outage duration during storm events through proactive crew positioning
  • 22% fewer equipment-failure outages after replacing the top 50 highest-risk transformers
  • Model accuracy: 71% on held-out test data, sufficient for operational crew pre-staging decisions

Timeline

Month 1: Completed guides, adapted model to local data. Month 2: Validated predictions against actual storm events. Month 3: Pilot deployment with dispatch team during spring storm season.

Southwest Co-op Accelerates DER Interconnection by 60%

Rural Electric Cooperative ~45,000 Members Guide 03 + Guide 07 4 Months to Deployment

Challenge

A Southwest cooperative faced a surge of rooftop solar interconnection requests. Their hosting capacity analysis relied on manual engineering reviews that took 2–4 weeks per application. The backlog frustrated members and slowed solar adoption across the service territory.

Approach

The planning engineer completed Guide 03 (Hosting Capacity) and Guide 07 (DER Scenario Planning), then built an automated screening tool using the cooperative's SCADA data and network model.

  • Built a feeder-level hosting capacity model using transformer ratings, peak load, and existing solar penetration
  • Created automated pass/fail screening for interconnection applications under 25 kW
  • Used scenario planning to identify feeders approaching hosting limits under projected solar growth

Results

  • 60% faster interconnection processing for residential solar (from 3 weeks to 5 days average)
  • 85% of applications auto-screened without manual engineering review
  • Proactive upgrades planned for 8 feeders approaching hosting capacity limits within 3 years

Timeline

Month 1–2: Completed guides and adapted models to local network data. Month 3: Built screening tool with internal IT support. Month 4: Deployed for new interconnection applications.

Regional IOU Detects Revenue-Grade Meter Anomalies

Investor-Owned Utility ~320,000 Customers Guide 08 + Guide 02 6 Months to Production

Challenge

A regional IOU with full AMI deployment was sitting on billions of interval data points but using them primarily for billing. The revenue assurance team suspected meter tampering and technical losses but had no automated way to identify anomalous consumption patterns across 320,000 meters.

Approach

A data analyst on the revenue assurance team completed Guide 08 (Anomaly Detection) and Guide 02 (Load Forecasting), then built a pipeline processing their AMI data.

  • Built load forecasting baselines by customer segment using historical interval data
  • Applied Isolation Forest anomaly detection to flag meters with consumption patterns deviating from expected baselines
  • Cross-referenced anomalies with transformer-level energy balance checks

Results

  • $2.1M annual revenue recovery from identified meter anomalies and technical losses
  • 340 meters flagged for field investigation in the first quarter, 78% confirmed as genuine issues
  • Ongoing monitoring: automated weekly anomaly reports replaced manual quarterly audits

Timeline

Month 1–2: Completed guides, established data pipeline from AMI system. Month 3–4: Trained and validated models on historical data with known anomalies. Month 5–6: Production deployment with field validation of flagged meters.

Start Your Own Path

Every scenario above started with an engineer completing a guide. Your domain expertise is the hard part—the ML Playground handles the rest.

Explore the ML Playground or Get In Touch