基于深度强化学习的大型公共建筑智慧节能方法研究

As basic components and information nodes of smart cities, intelligent buildings have put forward higher requirements to energy efficiency. Large public buildings feature high usage rate, large total energy consumption, and large quantity of environmental equipment, which are increasingly becoming an important concern of building energy-saving..The traditional building energy consumption models obtain the defects that they cannot dynamically optimize the energy with the changes of surrounding environment. To address the problem, this project aims to build a dynamic energy consumption model which extracts the features and characteristics of personnel behavior patterns via deep learning, and analyzes the influence of behavior patterns and environmental factors on building energy consumption. Compared to the traditional models which lack the ability to reflect the variation of people’s behavior or surroundings, the proposed model ccombines the dynamic energy consumption model with deep reinforcement learning. Through the depth analysis and mining of the actual energy consumption data, we construct an optimization strategy of online energy-saving depth perception to the environment changes, which will ensure the best energy-saving effects with a comfortable living environment retained..Simulations will be performed to evaluate the correctness of the proposed model. After that, the feasibility and efficiency of the dynamic model as well as the energy-saving strategy will be assessed in practical cases. This project will provide theoretical reference and practical case for the research of new energy-saving methods for buildings. Meanwhile, it is a new attempt to expand the application field of machine learning.

智能建筑作为智慧城市的基本构成单元和信息节点，对节能效果提出了更高要求。大型公共建筑以人群密度变化大、能耗总量大、环境设备数量多且异构等特点，而成为建筑节能关注的重点。.针对传统的建筑能耗模型无法跟随环境变化而动态优化的缺陷，本项目采用深度学习方法对人员行为模式进行特征提取与表征，通过分析人员行为模式与环境因素变化对建筑能耗的影响，建立结构可调的动态能耗模型。在此基础上，将深度强化学习方法与动态能耗模型相结合，通过对实际能耗大数据的深度分析与挖掘，构建与人员行为模式相互补、对环境变化能够深度感知的在线节能优化策略，以确保在满足环境舒适度的条件下达到智慧节能的效果。.项目研究拟采用仿真工具对动态模型及节能策略进行评估，并将其应用于实际工程案例中，对节能优化策略的有效性进行检验，从而为智能建筑系统的智慧节能方法研究提供理论参考与实践案例，同时也为拓展机器学习的应用领域而进行新的尝试。

智能建筑作为智慧城市的基本构成单元和信息节点，对节能效果提出了更高要求。大型公共建筑以人群密度变化大、能耗总量大、环境设备种类及数量多等特点，而成为建筑节能关注的重点。针对传统的建筑能耗模型无法进行动态优化的问题，采用深度学习方法对人员行为方式进行特征提取与表征，通过分析人员行为方式与环境因素变化对建筑能耗的影响，建立了结构可调的动态能耗模型。在此基础上，将深度强化学习方法与动态能耗模型相结合，通过对实际能耗大数据的深度分析与挖掘，构建了与人员行为模式相互补、对环境变化能够实时感知的在线节能优化策略，为如何满足环境舒适度的条件下达到节能的最佳效果探索出新的方法。. 本项目研究采取理论研究与实际应用相结合的方式加以展开，并借助于仿真工具对动态模型及优化节能策略进行验证，最后将其应用于实际工程案例中，对节能优化策略的有效性加以检验，从而为智能建筑系统的智慧节能方法研究提供理论参考与实践案例，同时也为拓展机器学习的应用领域进行了新的尝试。. 通过本项目的实施，项目组成员围绕其核心研究内容发表论文36篇，其中SCI收录14篇，EI检索17篇，中文核心期刊12篇；申请中国专利13项，其中发明专利8项；软件著作权4项；编写强化学习专著1部；培养硕士研究生18名，其中已毕业研究生11名。