Applying AI at Vattenfall: Exploring Smart Charging with Plug-Out Time Prediction

Across our E-mobility business, we operate thousands of charging points across multiple markets. At the same time, grid congestion, particularly during peak periods such as winter evenings, places increasing pressure on how and when energy can be delivered.

Today, charging typically starts immediately when a car is plugged in, regardless of when the energy is actually needed. In many cases, this leads to unnecessary peaks in demand and missed opportunities to better align charging with grid conditions.

To manage grid constraints, charging capacity may occasionally need to be reduced. While many sessions can absorb this, a meaningful share is impacted, potentially leading to incomplete charging and reduced revenue.

At the same time, EV charging flexibility represents a significant opportunity. If we can understand when a vehicle actually needs to be ready, we can shift charging to more optimal moments, thereby supporting grid stability while maintaining a good customer experience.

The key question becomes: how can we better understand and utilise this flexibility in real time?

The plug-out time prediction initiative reflects a broader way of working within IT at Vattenfall: structured experimentation combined with clear evaluation criteria.

Rather than moving directly into production, we work in stages:

• Data exploration and preparation from charging sessions

• Model development and validation

• Controlled proof-of-concept scenarios

• Dry runs, where predictions are generated without influencing operations

This approach allows us to validate whether predictions are accurate, meaningful, and actionable before considering integration into operational systems.

Integration itself remains conditional and driven by demonstrated value rather than predefined assumptions.

Although still in progress, the initiative highlights several areas of potential impact:

• Smarter charging strategies that actively support grid stability during peak periods

• Improved customer outcomes by reducing the likelihood of incomplete sessions

• Increased ability to monetise flexibility in electricity and balancing markets

• A stronger and more adaptive InCharge offering

At the same time, important questions remain:

• How accurate can predictions be at an individual session level?

• How should uncertainty and confidence levels be reflected in decision-making?

• What is the most effective way to incorporate predictions into operational systems without adding unnecessary complexity?

These questions are part of the ongoing learning process and will shape future development.

The plug-out time prediction initiative demonstrates how we explore AI at Vattenfall: starting from a real business challenge, validating potential through structured experimentation, and making decisions based on observed value.

As a work in progress, it represents a step towards more flexible, grid-aware energy solutions, where charging behaviour can adapt both to customer needs and to the realities of the energy system.

Through initiatives like this, we aim to better utilise flexibility, reduce grid stress, and contribute to a more resilient and efficient energy system.