This repository contains code and resources related to the implementation of Physics-Based Regularized Bayesian Neural Networks (PBR-BNN) for precise occupant detection in the context of building energy optimization.
Accurate occupancy detection is crucial for optimizing building energy management. Challenges arising from limited data, sensor noise, and intricate dynamics often undermine precision. In this project, we propose a two-step approach to address these challenges:
-
Feature Enhancement: We incorporate supplementary features derived from building information to enhance data quality and predictive capacity.
-
Physics-Based Regularization: A physics-based regularizer is integrated into a Bayesian Neural Network Model, ensuring adherence to constraints and reliable uncertainty estimation, thus elevating prediction precision.
The replication package requires Python 3.8+ and a bunch of dependencies listed in the requirements.txt file. Install the requirements.txt before running each notebook.
pip install -r requirements.pip
-
dataset: Utilizing open-access data from an office building experiment encompassing occupancy profiles, electricity consumption, and indoor environmental data [33]. This folder entails all the data required to replicate the notebooks. -
GUI: Contains the code for the web application, which can also be accessed at scis_occupancy and Occupant Detector. -
Figures: Contains the images from this research. -
Notebooks: Contains all notebook files. Ensure to have install the requirements before usage. -
saved_model: Contains saved model for each test cases.
The algorithms trained and tested on this project include:
- Naïve Bayes
- Gradient Boosting Machine
- Kernelized Support Vector Machine
- Bayesian Neural Network
- PBR-BNN
A. Yahaya, A.B. Owolabi, D. Suh,
Enhancing Building Energy through Regularized Bayesian Neural Networks for Precise Occupancy Detection,
Journal of Building Engineering, 2025.
https://doi.org/10.1016/j.jobe.2025.112777