基于多级数据驱动模型的集中式空调系统优化策略自生成算法  

Self-generating Algorithm of Optimization Strategies for Central Air-conditioning System Based on Multilevel Data-driven Model

作　　者：毛晨楠 叶明树 李铮伟[1] MAO Chennan;YE Mingshu;LI Zhengwei(School of Mechanical Engineering,Tongji University,Shanghai 200092,China;Xiamen Jinming Energy-Saving Technology Co.,Ltd.,Xiamen 361010,Fujian,China)

机构地区：[1]同济大学机械与能源工程学院,上海200092 [2]厦门金名节能科技有限公司,福建厦门361010

出　　处：《建筑科学》2024年第12期164-172,共9页Building Science

摘　　要：在当今“双碳”目标的背景下,空调系统优化控制策略以其对建筑运行阶段能耗降低的重要作用一直以来颇受行业重视。相较于传统的基于经验的规则策略,监督式优化控制方法可以在变工况下实现系统更高效地运行。而监督式优化算法消耗较多算力,通常依赖于边缘计算机或云服务器。为了避免因网络或计算设备问题导致监督式优化算法的失效,在本地控制器需要1套备用的控制策略。为了解决这个问题,本文提出了基于多级数据驱动模型的集中式空调系统优化策略自生成算法,该算法基于监督式优化算法优化结果,通过两级数据驱动模型,生成规则性优化策略。本研究以某空气源热泵系统为例,根据历史数据建立系统性能预测模型,选择Q学习作为监督式优化算法并部署于系统预测模型,得到优化结果。利用所提算法生成优化策略并验证节能效果。结果表明,相较于原始策略,所提方法生成的策略在维持一定舒适度的前提下,提升COP约5.09%~7.19%,与Q学习的优化效果相当。In the context of carbon peaking and carbon neutrality goals,optimization strategies for air-conditioning systems have always been highly valued by the industry for its key role in reducing energy consumption during the operation phase of a building.Compared with the traditional empirical rule-based strategy,the supervised optimization control method can achieve more efficient system operation under variable operating conditions.The supervised optimization algorithm,however,consumes more computing power and usually relies on edge computers or cloud servers.In order to avoid the failure of the supervised optimization algorithm due to network or computing device problems,a set of back-up control strategies is needed at the local controller.To solve this problem,this paper proposed a self-generating algorithm for air-conditioning system optimization strategies based on a multilevel data-driven model,which generates regular optimization strategies according to the optimization results of the supervised optimization algorithm through a two-level data-driven model.This paper conducted a case study of an air source heat pump system,built a system performance prediction model based on historical data,selected Q-learning as the supervised optimization algorithm and deployed it in the system prediction model to obtain optimization results.The proposed algorithm was used to generate optimization strategies and verify the energy saving effect.The results showed that compared with original strategies,the strategies generated by the proposed method improved COP by about 5.09%to 7.19%while maintaining a certain level of comfort,which is equivalent to the optimization effect of Q-learning.

关 键 词：优化策略 监督式优化控制 规则性策略 空调系统 Q学习 

分 类 号：TU831[建筑科学—供热、供燃气、通风及空调工程]