Abstract:Heating, ventilation, and air conditioning (HVAC) energy consumption now accounts for a major portion of energy use for buildings. Therefore, finding the optimal energy-saving control strategy for HVAC systems to optimize energy consumption has become crucial in realizing energy savings, emission reductions, and green buildings. Traditional methods for HVAC parameter control require complex physical model calculation; continuous and coupled parameters are handled poorly. Developing deep reinforcement learning (DRL) methods provides new ideas for HVAC energy consumption optimization. Herein, a DRL-based energy consumption optimization framework for HVAC systems is proposed. First, an HVAC system energy consumption prediction model based on a convolutional neural network–long short-term memory (CNN-LSTM) network is suggested to approximate the real world. This model solves the efficiency problem of energy consumption prediction while also providing highly accurate predictions of HVAC energy consumption. We propose an enhanced deep deterministic policy gradient (E-DDPG) energy consumption optimization algorithm for HVAC systems based on an improved training strategy to obtain the best real-time energy consumption control strategy for HVAC systems. Finally, experiments using real-world building HVAC control data sets were conducted to evaluate our models. The experiments show that the CNN–LSTM model for HVAC system energy consumption prediction outperforms baseline models while reducing training time by 42.9%. Compared to the baseline algorithm, the E-DDPG algorithm using an improved training strategy requires 20% fewer iterations for convergence, has a 14.8% narrower fluctuation interval during the training process, and improves the energy efficiency ratio of HVAC systems by 49%.

Data-driven control of room temperature and bidirectional EV charging using deep reinforcement learning: Simulations and experiments

Real building implementation of a deep reinforcement learning controller to enhance energy efficiency and indoor temperature control

Energy-Efficient Thermal Comfort Control in Smart Buildings via Deep Reinforcement Learning

Energy-Efficient HVAC Control Based on Reinforcement Learning and Transfer Learning in a Residential Building

Data-driven control of micro-climate in buildings: An event-triggered reinforcement learning approach

Modelling building HVAC control strategies using a deep reinforcement learning approach

Distributed Multi-Agent Deep Reinforcement Learning Framework for Whole-building HVAC Control

Intelligent multi-zone residential HVAC control strategy based on deep reinforcement learning

Hybrid-Model-Based Deep Reinforcement Learning for Heating, Ventilation, and Air-Conditioning Control

An experimental evaluation of deep reinforcement learning algorithms for HVAC control

Intelligent control of electric vehicle air conditioning system based on deep reinforcement learning

Near-optimal Deep Reinforcement Learning Policies from Data for Zone Temperature Control

Exploring Deep Reinforcement Learning for Holistic Smart Building Control

Simulation Studies on Deep Reinforcement Learning for Building Control with Human Interaction

Energy Consumption Optimization for Heating, Ventilation and Air Conditioning Systems Based on Deep Reinforcement Learning

DeepComfort: Energy-Efficient Thermal Comfort Control in Buildings Via Reinforcement Learning

Deep Reinforcement Learning for Optimal Control of Space Heating

ED-DQN: An event-driven deep reinforcement learning control method for multi-zone residential buildings

MF^2: Model-free Reinforcement Learning for Modeling-Free Building HVAC Control with Data-Driven Environment Construction in a Residential Building

Experimental analysis of data-driven control for a building heating system