摘要
为解决传统建筑能耗管理方法预测精度低、实时性差、可扩展性弱等问题,本文提出一种基于机器学习集成模型的分布式建筑能耗监测预警平台。平台采用分层架构设计,感知层通过物联网设备实时采集多源数据;数据层完成缺失值填充、异常值修正、离散编码与标准化预处理,并基于特征重要性评估筛选日类别、时间节点、干球温度、相对湿度、太阳辐射强度及供热通风与空气调节系统能耗功率6项核心特征;应用层集成可视化与预警功能。为避免出现预警错误或遗漏,本文提出双层集成预测模型:第一层由人工神经网络与轻量级梯度提升机并行生成初步预测结果,第二层通过线性回归加权融合双模型输出,显著提升模型性能。模型验证表明,平台可实现实时数据采集、动态阈值预警及多建筑集群并行管理,为节能决策提供全流程技术支持。
In order to solve the problems of low prediction accuracy,poor real-time performance and weak scalabil⁃ity of traditional energy consumption management methods,this paper proposes a distributed building energy con⁃sumption monitoring and early warning platform based on machine learning integrated model.The platform adopts ahierarchical architecture design,and the perception layer collects multi-source data in real time through IoT devic⁃es;the data layer completes missing value filling,outlier correction,discrete coding and standardized preprocess⁃ing,and screens six core features of day category,time node,dry bulb temperature,relative humidity,solar radia⁃tion intensity and energy consumption power of heating ventilation and air conditioning system based on feature im⁃portance evaluation.Application layer integrates visualization and early warning functions.In order to avoid earlywarning errors or omissions,this paper proposes a two-layer integrated prediction model:the first layer generatespreliminary prediction results in parallel by artificial neural network and lightweight gradient hoist,and the secondlayer fuses the dual model output through linear regression weighting,which significantly improves the model perfor⁃mance.The model verification shows that the platform can realize real-time data acquisition,dynamic thresholdwarning and multi-building cluster parallel management,and provide full-process technical support for energy-sav⁃ing decision-making.
作者
徐基平
XU Jiping(Shandong Urban Construction Vocational College,Jinan 250103,China)
出处
《粘接》
2025年第10期218-221,共4页
Adhesion
关键词
建筑能耗预测
轻量级梯度提升机
人工神经网络
集成模型
监测预警平台
building energy consumption prediction
lightweight gradient hoist
artificial neural network
integrat-ed model
monitoring and early warning platform
