Improving Energy Modeling Accuracy for Complex Systems

42.1% error reduction with 85% less data. At Ento, we use a lot of traditional black-box Machine Learning to model building energy consumption, and they're great for many use cases. But they have their limits. When we're dealing with: - Plenty of indoor sensor data - Limited historical data - The need to actively control a building’s HVAC system ... plain black-box approaches often fall short. That’s why I’ve been following key trends around blending data-driven methods with physical modeling: 🔹 Transfer Learning: Use data from similar buildings to improve models. 🔹 Digital Twins: Blend data-driven methods and physical simulations. 🔹 Physics-Informed AI: Embed physical laws into the learning process to improve results. Just last month, three papers in these fields came out from leading researchers: - GenTL: A universal model, pretrained on 450 building archetypes, achieved a 42.1% average error reduction when fine-tuned with 85% less data. From Fabian Raisch et al. - An Open Digital Twin Platform: Han Li and Tianzhen Hong from LBNL built a modular platform that fuses live sensor data, weather feeds, and physics-based EnergyPlus models. - Physics-informed modeling: A new study proved that Kolmogorov–Arnold Networks (KANs) can rediscover fundamental heat transfer equations. From Xia Chen et al. Which of these 3 trends do you see having the biggest real-world impact in the next 2-3 years?

We’re excited to share our latest research on a topic at the intersection of electrical engineering and building systems to support building to grid integration: 🔌💨 Coupling Induction Machines with HVAC Systems for Integrated Simulation and Control. In this work, we developed a Computationally Efficient and Accurate Induction Machine (CEAIM) model and integrated it with HVAC components like pumps, heat pumps, and chillers. This allows us to study how electrical behavior directly affects thermal and fluid performance—a key step in simulating grid-interactive, energy-efficient buildings. ✅ Validated with experimental data ✅ Up to 1,000× faster than existing induction machine models ✅ Achieved R² values between 0.98 and 1 for power, speed, and torque predictions The CEAIM model is implemented in #Modelica, enabling scalable, equation-based modeling. Our case study shows how this integrated approach can improve accuracy and reduce computational load—especially important for smart grid and load-shedding analyses. Led by SBS Lab Ph.D. student Viswanathan Ganesh, this is a joint effort of Penn State University, Berkeley Lab, Oak Ridge National Laboratory and National Renewable Energy Laboratory (Zhanwei He, Jianjun Hu and Sen Huang). Thanks to the Gordon D. Kissinger Graduate Research Fellowship for Viswanathan Ganesh, this paper is published as Open Access paper: https://lnkd.in/eXyWDEVC #HVAC #BuildingSimulation #SmartGrid #EnergyEfficiency #ElectricalEngineering #IntegratedModeling

🚀 New preprint alert! Proud to share our latest work: "Thermodynamically Consistent Latent Dynamics Identification for Parametric Systems" 📄 https://lnkd.in/gPCQHYiZ In this paper, we propose tLaSDI, a novel framework for reduced-order modeling that fuses thermodynamic principles with machine learning to model complex, parametric dynamical systems with #interpretability, #consistency, and #speed. 🔍 Key innovations: + #pGFINNs: A new class of GENERIC-informed neural networks that enforce the first and second laws of thermodynamics in latent space dynamics. + #Physics-#informed #active #learning: An adaptive sampling strategy that drastically improves accuracy and efficiency using a physics-informed error indicator. + #Massive #computational #gains: Up to 3,528× speed-up with only 1–3% error, plus 50–90% training cost reduction over prior state-of-the-art. + #Insightful #latent #dynamics: Latent variables reflect #free #energy #conservation and #entropy #generation, offering physically meaningful interpretation of learned models. 🧪 Benchmarks, demonstrating both predictive accuracy and thermodynamic fidelity, include: + Burgers’ equation + 1D/1V Vlasov–Poisson equation 🤝 With amazing collaborators: Xiaolong He, Yeonjong Shin, Anthony Gruber, Sohyeon Jung & Kookjin Lee #neural #network #ML #AI #simulation

Energy models are supposed to predict building performance. But here’s the uncomfortable truth: most of them don’t. In theory, LEED energy modeling is a powerful tool. It helps teams simulate design decisions, compare systems, and optimize efficiency before construction begins. Yet, once the building is occupied, the numbers often tell a different story. Why? Because models are only as real as the assumptions behind them: - Schedules that don’t match how people actually use the space. - Equipment efficiencies that fade with time (Sometimes specified systems are not procured > leading towards inefficient performance). - Climate data that were impacted with the rapid urban heat shifts. The result? A performance gap where predicted energy savings look impressive on paper, but not on the utility bill during operations. So what’s the solution? It starts with making our models smarter before we ever break ground: - Use refined inputs based on real operational data from similar buildings, and data sheets performance and not theories or textbook assumptions. - Apply practical operating schedules that reflect how the building will truly function. - Engage the commissioning team early to validate design intent, specs, and system selection. - Revisit the energy model throughout design development, not just at the end for LEED submission. - Document key performance assumptions clearly so they can guide procurement and installation. #LEED #EnergyModeling #BuildingPerformance #Sustainability #Environment #ESG #GreenBuilding #SustainableDesign #Commissioning #PerformanceGap