Advanced Multi-Objective Optimization of Energy Management Systems Using Deep Neural Networks and HPC for Real-Time, Multi-Step Energy Forecasting (AIMED-HPC)

Short description of the business experiment

The AIMED-HPC project enhances real-time, AI-driven energy forecasting for large-scale building management. It addresses computational bottlenecks in RNN-based models (e.g., BiLSTM, MLPs) by integrating High-Performance Computing (HPC). The goal is to deliver 1–3-day energy predictions at 15-minute granularity, optimising Robotina’s EMS.

Key tasks include parallelised AI training, neural architecture search, and hyperparameter tuning. Deployment utilises Kubernetes, TensorFlow Lite, and knowledge distillation for real-time edge predictions.

The project enables cost-efficient energy management, supporting scalable SaaS solutions for global markets, energy trading, and real-time demand optimisation across thousands of buildings.

Organisations involved: 

End User: Robotina d.o.o.

Technology expert: Rudolfovo

HPC provider: University of Ljubljana, Faculty of Mechanical Engineering, LeCAD laboratory