基于领域泛化的制冷系统性能预测模型研究OA
Research on Performance Prediction Models for Refrigeration System Based on Domain Generalization
本文提出一种基于多域动态自适应泛化网络的领域泛化方法,旨在构建适用于多联机系统的通用预测模型.为验证所提方法的性能,采用5 套多联机系统的历史数据集开展实验;对各数据集完成数据清洗筛选与特征维度对齐处理后,选取其中一套系统作为目标域,剩余系统作为源域.采用网格搜索法对不同的平衡系数λ 进行模型训练.结果表明:功率模型决定系数 R2均大于0.95,峰值为 0.977,制冷量模型决定系数R2均大于0.9,峰值为0.946.相较于经验风险最小化基准方法,所提方法的功率与制冷量预测模型在R2指标上均取得提升.
In this paper,the domain generalization method of multi-domain dynamic adaptive generalization network is proposed to establish a general model of domain generalization in variable refrigerant flow system.To evaluate the performance of this method,experiments are conducted using historical datasets from five variable refrigerant flow systems.After cleaning and dimension alignment for each dataset,one system is selected as the target domain and the remaining system as the source domain.Models are trained with different balance coefficient λ,yielding power model coefficient of determination R2 all above 0.95,with the highest at 0.977,and refrigerating capacity model coefficient of determination R2 all above 0.9,with the highest at 0.946.Compared with the empirical risk minimization method,both power and refrigerating capacity prediction models show improvement in R2.
郑飞;刘颖;王杰;晋欣桥;杜志敏
上海交通大学制冷与低温工程研究所,上海 200240上海交通大学制冷与低温工程研究所,上海 200240上海交通大学制冷与低温工程研究所,上海 200240上海交通大学制冷与低温工程研究所,上海 200240上海交通大学制冷与低温工程研究所,上海 200240
通用工业技术
多联机系统通用模型领域泛化性能预测
Variable refrigerant flow systemUniversal modelDomain generalizationPerformance prediction
《制冷技术》 2026 (3)
21-27,69,8
科技部重点研发计划(No.2021YFE0107400),国家自然科学基金(No.52276011).
评论