Loading...

Design and Implementation of Machine-Learning Systems for Energy Saving in Smart Buildings

Sadeghzadeh, Hassan | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57433 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Gholampour, Iman
  7. Abstract:
  8. The comfort of occupants has been a key driver behind the adoption of smart building technologies in recent years. To achieve this goal, machine learning methods have seen significant growth, with deep reinforcement learning emerging as particularly advantageous. Unlike advanced control strategies such as model predictive control, deep reinforcement learning does not require a physical model. Instead, learning occurs through the direct interaction of an intelligent agent with the environment, without the need for predefined datasets. This approach is further strengthened by its ability to adapt to dynamic conditions and effectively operate in large state spaces. Given that heating and cooling systems account for 40-50% of a building's energy consumption and play a critical role in occupant comfort, optimizing these systems through intelligent control is of paramount importance. This study focuses on the design and implementation of an intelligent controller for heating and cooling systems using advanced machine learning techniques, with a particular emphasis on deep reinforcement learning and the Soft Actor-Critic (SAC) method. The proposed system is tailored for large-scale buildings, such as office complexes and hotels. To address the challenge of lengthy learning times for intelligent agents in new environments, this research introduces a novel feature of generalization and transferability, enabling the rapid adaptation of the control system to new buildings. This innovation significantly reduces both the cost and time of deployment, while enhancing system reliability and sustainability. Additionally, we developed performance metrics to quantitatively assess the trade-off between energy savings and occupant thermal comfort. The implementation was carried out using a modular approach, without relying on pre-existing libraries, allowing for execution in both simulation and real-world environments, with potential for future expansion. The simulation platform developed for this purpose yielded promising results. Key outcomes include an average 25% reduction in energy consumption for heating and cooling systems while maintaining thermal comfort, and successful transferability to new buildings. Furthermore, occupancy status was found to contribute an additional 10% in energy savings when factored into the control strategy. Notably, using a binary representation of occupancy, rather than exact counts, reduced complexity and costs while improving performance. Comparative analysis between the generalized intelligent controller and a non-generalized version demonstrated higher stability and superior performance for the former. In a new building, the generalized controller achieved a 15% energy savings without any initial learning, which increased to 25% after the agent had adapted to the environment over time
  9. Keywords:
  10. Intelligent Controller ; Deep Reinforcement Learning ; Machine Learning ; Energy Optimization ; Thermal Comfort ; Intelligent Building ; Energy Consumption Optimization

 Digital Object List

 Bookmark

...see more