Posted on Leave a comment

GRL approach for smart building energy and comfort optimization

Reference :

Haidar, N., Tamani, N., Ghamri-Doudane, Y., & Boujou, A. (2023). Selective reinforcement graph mining approach for smart building energy and occupant comfort optimization. Building and Environment228, 109806. https://doi.org/10.1016/j.buildenv.2022.109806

Overview:

Optimizing building energy usage is crucial because it significantly impacts the environment. Unfortunately, current techniques’ inability to accurately forecast occupant behavior often leads to ineffective HVAC management, causing inhabitants to feel uncomfortable or waste energy. The author proposes using information technology to gather data on energy use and tenant behavior in buildings by installing sensors. The authors present an optimization technique based on graph mining that combines selective reinforcement learning with a model for predicting occupant behavior. Real-time occupancy sensors help identify prediction flaws, enabling remedial action to be taken.

Limitation in the previous HVAC management techniques:

  • Traditional methods suggested by some papers limited due to their accurate fixed operation schedule.
  • Despite the fact that people have a major impact on HVAC energy consumption, there is a dearth of research on the detection and analysis of occupant behavior in building systems.
  • Conventional methods either aim at predicting occupant movements to selectively heat/cool rooms or leave HVAC systems running in every room, wasting energy while the rooms are empty.
  •  On the other hand, inaccurate forecasts may make residents uncomfortable.
  • Existing models might not be able to handle these mistakes well enough.
  • Although Reinforcement Learning (RL) has demonstrated potential in a number of domains, there hasn’t been much incorporation of RL into building systems for energy optimization.
  •  It’s possible that RL wasn’t completely utilized in earlier research to increase prediction accuracy and energy efficiency.

To tackle these issues the author proposes:

The author presents a graph mining-based optimization technique that combines a selective reinforcement learning method with an occupant behaviour prediction model to analyse user behaviour, identify prediction mistakes, and fix the model.

Methodology:

Three primary algorithms make up the methodology of this paper: Selective Prediction Reinforcement (SPR), Real-time Room Occupancy based on Prediction Reinforcement (OPR), and Occupant Movement Prediction (OMP).

Simplified diagram of the proposed method from the study by Haidar, N(2023)

OMP evaluates room occupancy based on past movement data and forecasts future movements. It uses a graph-based technique to describe motions and trains a model using the Graph Learning algorithm. OPR identifies and fixes forecast flaws by contrasting anticipated and actual occupancy from sensors. SPR optimizes the prediction model, giving more weight to certain predictions based on occupancy length. It maximizes energy usage by identifying rooms with brief occupancy and refraining from turning on HVAC systems.

Conclusion:

  • By combining OMP, OPR, and SPR, we can reduce HVAC energy consumption by up to 80.1% while maintaining occupant satisfaction levels of up to 97%.
  • Future scope: To evaluate and improve the presented technique, we need further trials using larger datasets. It is important to take into account various building kinds while fine-tuning and optimizing the prediction model and system parameters.

 

 

Sakthivel R

I am a First-year M.Sc., AIML student enrolled at SASTRA University

I possess a high level of proficiency in a variety of programming languages and frameworks including Python. I have experience with cloud and database technologies, including SQL, Excel, Pandas, Scikit, TensorFlow, Git, and Power BI

Leave a Reply

Your email address will not be published. Required fields are marked *