Online transfer learning strategy for enhancing the scalability and deployment of deep reinforcement learning control in smart buildings

Applied Energy, Год журнала: 2023, Номер 333, С. 120598 - 120598

Опубликована: Янв. 10, 2023

Язык: Английский

---

State of the art review on the HVAC occupant-centric control in different commercial buildings

Journal of Building Engineering, Год журнала: 2024, Номер 96, С. 110445 - 110445

Опубликована: Авг. 13, 2024

Язык: Английский

---

A comprehensive review of predictive control strategies in heating, ventilation, and air-conditioning (HVAC): Model-free VS model

Journal of Building Engineering, Год журнала: 2024, Номер 94, С. 110013 - 110013

Опубликована: Июнь 21, 2024

Язык: Английский

---

Comparative study on deep transfer learning strategies for cross-system and cross-operation-condition building energy systems fault diagnosis

Energy, Год журнала: 2022, Номер 263, С. 125943 - 125943

Опубликована: Ноя. 3, 2022

Язык: Английский

---

Enhancing HVAC control systems through transfer learning with deep reinforcement learning agents

Smart Energy, Год журнала: 2024, Номер 13, С. 100131 - 100131

Опубликована: Янв. 26, 2024

Traditionally, building control systems for heating, ventilation, and air conditioning (HVAC) relied on rule-based scheduler systems. Deep reinforcement learning techniques have the ability to learn optimal policies from data without need explicit programming or domain-specific knowledge. However, these data-driven methods require considerable time effective prior Performing transfer using pre-trained models avoids underlying scratch, thus saving resources. In this work, we evaluate as a method of pretraining fine-tuning neural networks HVAC control. First, train an RL agent in simulation environment obtain foundation model. We then fine-tune model two separate environments such that one simulates same under different weather conditions while other conditions. perform experiments with reward functions their effect learning. The results indicate agents outperform controller show improvements range 1% 4% when compared trained scratch.

Язык: Английский

---

A systematic review of reinforcement learning application in building energy-related occupant behavior simulation

Energy and Buildings, Год журнала: 2024, Номер 312, С. 114189 - 114189

Опубликована: Апрель 20, 2024

Язык: Английский

---

Chiller energy prediction in commercial building: A metaheuristic-Enhanced deep learning approach

Energy, Год журнала: 2024, Номер 297, С. 131159 - 131159

Опубликована: Апрель 4, 2024

Язык: Английский

---

Energy and comfort aware operation of multi-zone HVAC system through preference-inspired deep reinforcement learning

Energy, Год журнала: 2024, Номер 292, С. 130505 - 130505

Опубликована: Янв. 29, 2024

Язык: Английский

---

An innovative heterogeneous transfer learning framework to enhance the scalability of deep reinforcement learning controllers in buildings with integrated energy systems

Building Simulation, Год журнала: 2024, Номер 17(5), С. 739 - 770

Опубликована: Фев. 20, 2024

Abstract Deep Reinforcement Learning (DRL)-based control shows enhanced performance in the management of integrated energy systems when compared with Rule-Based Controllers (RBCs), but it still lacks scalability and generalisation due to necessity using tailored models for training process. Transfer (TL) is a potential solution address this limitation. However, existing TL applications building have been mostly tested among buildings similar features, not addressing need scale up advanced real-world scenarios diverse systems. This paper assesses an online heterogeneous strategy, comparing RBC offline DRL controllers simulation setup EnergyPlus Python. The study tests transfer both transductive inductive settings policy designed manage chiller coupled Thermal Energy Storage (TES). pre-trained on source transferred various target characterised by system including photovoltaic battery storage systems, different envelope occupancy schedule boundary conditions (e.g., weather price signal). approach incorporates model slicing, imitation learning fine-tuning handle state spaces reward functions between buildings. Results show that proposed methodology leads reduction 10% electricity cost 40% mean value daily average temperature violation rate controllers. Moreover, maximises self-sufficiency self-consumption 9% 11% respect RBC. Conversely, achieves worse either or settings. (DRL) agents should be trained at least 15 episodes reach same level as TL. Therefore, effective, completely model-free can directly implemented real satisfying performance.

Язык: Английский

---

Novel transformer-based self-supervised learning methods for improved HVAC fault diagnosis performance with limited labeled data

Energy, Год журнала: 2023, Номер 278, С. 127972 - 127972

Опубликована: Май 31, 2023

Язык: Английский

---