Sign in Subscribe
Artificial neural networks

Research Uncovers Efficient Neural Network Approach for Predicting Energy Performance in Commercial Buildings

Strategic Insights

— 2 min read

A new report from researchers at the University of Perugia reveals a simplified neural network approach to predict energy and thermal performance in commercial buildings, enhancing the applicability of artificial intelligence techniques. By utilizing the EnergyPlus dynamic simulation software, the study demonstrates a computationally lightweight and scalable solution for performance prediction. The research findings show that the proposed approach can accurately reproduce the functional relationship between input and output data, particularly in the southern region of Italy.

Key Takeaways:

  • The research focused on optimizing energy consumption in non-residential buildings with complex geometries, a challenging task due to high computational costs.
  • A neural network-based approach was proposed to characterize the thermal-energy relationship in commercial buildings, providing an efficient and scalable solution for performance prediction.
  • The study utilized EnergyPlus dynamic simulation software to generate consumptions trends for a building over a year in different locations and trained neural network models using this data.
  • Uncertainty analyses were conducted to evaluate the behavior effectiveness of artificial neural networks (ANNs) in different weather conditions.
  • The research demonstrated that the proposed approach can accurately predict energy and thermal performance in commercial buildings, with an average root mean square error (RMSE) of 0.5 °C for the southern region and 1.0 °C for the northern region.
  • The proposed approach is computationally lightweight, with an inference time of below 5 ms, making it suitable for load dispatch optimization or microcontroller applications for building automation systems.
  • The study's findings have significant implications for building energy management and automation, as the proposed neural network approach can provide accurate and efficient predictions of energy and thermal performance in commercial buildings.

Statistics:

  • 61.2% of the models trained for the southern region achieved an RMSE below 0.5 °C, indicating a high level of accuracy in predicting energy and thermal performance.
  • Despite the challenges in the northern region, the model for these cities still achieved an acceptable accuracy of 1.0 °C, demonstrating the robustness of the proposed approach.
  • The study utilized EnergyPlus dynamic simulation software to generate consumptions trends for a building over a year in different locations, covering a total of 5400 data points.
  • The proposed neural network approach has a computational complexity of O(n log n), making it computationally efficient and scalable for large-scale applications.

Sources:

  • Development of Recurrent Neural Networks for Thermal/Electrical Analysis of Non-Residential Buildings Based on Energy Consumptions Data. Energies, 2025,18(12):3031. (Energies - http://www.mdpi.com/journal/energies).
  • Journal of Engineering, July 7, 2025, p 2469.

Read more