SIMBa

System Identification Methods leveraging Backpropagation - by Muhammad Zakwan

Imagine a smart building that can think ahead, like a thoughtful host who adjusts the thermostat just before guests arrive, making sure the room is always cozy, never too hot or too cold. To achieve this, the building needs to predict how temperatures will change over the next few hours or even days. But if its predictions rely only on experience instead of real physics, things can quickly go wrong: energy gets wasted, comfort suffers, and the system starts making decisions that simply don’t make sense in the real world.

This is where system identification comes into the picture. It’s like teaching a building how to behave based on data. Traditionally, there are two main approaches. On one hand, there are purely physics-based models, grounded in classical first principles such as Newton’s laws of motion and the fundamental laws of thermodynamics. While these models can be highly accurate in theory, they often require detailed knowledge of every component in the system, rely on simplifying assumptions, and can become impractically complex when applied to real buildings with countless interacting variables. On the other hand, there are traditional data-driven methods, often called black-box models, which try to fit data into general mathematical forms. These can be clunky, requiring lots of manual tweaking and still struggling to capture the true complexity of real-world systems.

This is exactly why we developed SIMBa, our open-source, Python-based toolbox. SIMBa takes a smarter approach: it integrates the fundamental laws of physics directly into the learning process, while also using advanced algorithms to fine-tune the building’s model so it aligns with reality. In other words, it combines the best of both worlds: the flexibility of data-driven methods and the reliability of physics-based principles, all within a single framework.

The result? A building that stays sharp over time, making accurate predictions and smart decisions, keeping everyone comfortable while saving energy. And the best part? This isn’t just limited to buildings. The same approach can help robots move more smoothly, drones fly more efficiently, and power grids operate more reliably. It’s the perfect blend of cutting-edge AI and good old-fashioned physics, working together to make technology smarter, more efficient, and far more dependable.