Traditional software solutions that support such process optimizations typically require re-engineering by optimization and production experts for each new type of process variable prediction. While some machine learning (ML) methods are already in use, building and maintaining end-to-end ML pipelines remains a complex task. It demands close collaboration between data scientists and domain experts to engineer the pipeline—from data preprocessing and model selection to hyperparameter tuning and prediction.

The solution to this problem is automated machine learning. AutoML leverages available plant data to generate forecasts and model identification without the need for data science expertise. This forms a critical foundation for widespread energy cost reduction and more sustainable resource utilization in industrial settings.

With these advantages in mind, ABB has now combined AutoML with its model-based energy management system – the ABB Ability^TM OPTIMAX®, an ABB-developed optimization technology and one of the few commercially successful mixed-integer (non-) linear program solvers worldwide. 

OPTIMAX® deployment is highly scalable from on-premises to server-based in data centers, to native cloud deployment. It can also be deployed as a hybrid solution with monitoring and reporting, predictive optimization and forecasting modules in the cloud, and real-time control and optimization features on-premises.

Combined with AutoML, OPTIMAX® has opened the door to integrated energy management workflows →01 based on multivariate forecasting models. Such models incorporate multiple input variables covering historical recordings and forecasts, thus enabling highly accurate and cost-effective planning optimization in energy-intensive industrial processes. Historically, generating reliable forecasts of this complexity required either advanced, specialized prediction tools or deep domain expertise.

With the integration of AutoML into OPTIMAX®, correlations among diverse influencing factors can now be automatically identified, allowing for the autonomous development of robust forecasting models. This scalable approach supports flexible adaptation across a wide range of industrial applications—from optimizing load management strategies to enabling predictive maintenance. It not only reduces engineering and development effort but also strengthens the analytical foundation for informed, data-driven control of energy systems. 

While data-driven methods identify patterns and correlations from historical data and forecasts that can emulate complex physical behaviors, physics-based models provide a reliable foundation grounded in established scientific principles. The combination of both approaches—known as grey-box modeling—is an active area of research at ABB. For example, the company is developing neural networks that enhance differential equations governing thermal networks and model state equations for CO₂ systems. In addition to automation through AutoML, the Functional Mock-up interface (FMI) enables the standardized integration of such models into industrial optimization environments, such as the Dynamic Optimization Runtime in OPTIMAX. 

The practical evaluation of these new methods is currently being conducted at a chemical park in the German state of Hesse, where multivariate AutoML forecasting models are being tested under real -world conditions. Hybrid modeling approaches are gaining increasing importance, particularly in emerging market segments such as carbon capture and storage. Additionally, prototypes for meta-models are being developed that utilize synthetic data from physical simulations to initialize data-driven models. This approach has the potential to significantly enhance the quality and usability of energy management systems.