PerPetum

28/10/2026

Optimizing solar plant performance with machine learning

We helped PerPetum bring fragmented plant, weather, and market data into a single operational platform that forecasts solar production, detects performance issues, and automatically adjusts output when market conditions demand it.

PerPetum Energy is a regional leader in the construction and management of industrial photovoltaic installations in Belgium and abroad.

The business challenge: managing solar performance across fragmented data

Operating a growing portfolio of solar plants requires a reliable view of how each installation is performing. For PerPetum, the necessary data was spread across different inverter manufacturers, custom data loggers, irradiance sensors, weather services, and energy-market sources.

Missing or inconsistent measurements added another challenge. PerPetum needed to bring this information together and simultaneously account for the operational realities of solar plants, including planned downtime, underperforming inverters, and periods when production is deliberately curtailed.

Monitoring photovoltaic plants with machine learning

The ambition: anticipating issues and market conditions

The company wanted to move beyond retrospective monitoring and use its data to anticipate what would happen next.

That meant forecasting solar production and electricity prices, detecting potential performance issues earlier and giving operators the ability to respond automatically when market conditions made continued production less attractive.

The approach

1. Creating a reliable data foundation

We designed a modular data and analytics architecture with a dedicated data warehouse that consolidates inverter measurements, weather observations and forecasts, and market data from sources including Elia and ENTSO-E. This created a consistent foundation for monitoring, analytics and automated control across PerPetum’s solar portfolio.

2. Forecasting production and detecting anomalies

On top of this data foundation, we developed machine learning models for real-time production estimation and multi-day solar forecasting. The models can also estimate unconstrained production when actual output has been curtailed, keeping forecasts useful even when measured production no longer reflects the plant’s full potential.

Anomaly detection algorithms flag underperforming plants and inverters, with SMS alerts helping technicians identify issues that require intervention.

3. Turning insights into operational action

We expanded the platform with day-ahead electricity-price forecasts, custom calculation tools, financial reporting, and Excel exports.

An operator-configurable PLC interface also translates market and operational conditions into automated plant-control actions. When electricity prices turn negative, for example, output can be reduced automatically rather than continuing to produce at a loss.

The outcome: improving performance and protecting margins

PerPetum’s teams now have a unified view of plant performance, with forecasts and anomaly detection helping them identify potential problems earlier and prioritize maintenance before production losses accumulate.

More accurate production and price forecasts support better nomination decisions and help reduce exposure to balancing costs. Automated curtailment protects margins during negative-price periods, while integrated financial calculations show more clearly how technical performance translates into commercial results.

The platform supports PerPetum in monitoring and forecasting hundreds of megawatts of solar power across Belgium and abroad. It also provides a scalable foundation for new data sources, analytics, and control strategies as its portfolio evolves.