基于LightGBM-GA-Seq2Seq的照明负荷预测研究

Lighting load prediction study based on LightGBM-GA-Seq2Seq

照明负荷是建筑能耗的重要组成部分,在提高照明品质和降低照明能耗的双重要求下,照明负荷预测的关键性逐渐凸显。随着信息技术的发展,负荷数据采集平台日益完善,负荷数据规模不断扩大,同时也伴随着数据复杂化。为了进一步精确预测照明负荷,分析与获取负荷的行为模式愈发重要,基于此提出一种集成学习照明负荷预测方法。通过K-means和Light Gradient Boosting Machine(LightGBM)算法对预测日的模式标签进行分析与获取,并与照明负荷以及气候特征一起作为Sequenceto Sequence(Seq2Seq)模型的输入以预测照明负荷。通过分析实际案例,本研究所提出方法相较于未采用行为模式标签的传统方法,显著改善了归一化均方根误差(NRMSE)和决定系数(R2)两项性能指标,分别提升了18%和8%,验证了该方法在照明负荷预测领域具备可行性与优越性。

Lighting load is an important part of building energy consumption,and under the dual requirements of improving lighting quality and reducing lighting energy consumption,the criticality of lighting load prediction has gradually come to the fore.With the development of information technology,the increasing improvement of load data collection platform makes the scale of load data expanding,and also accompanied by data complexity.In order to further predict the lighting load accurately,it is more and more important to analyze and obtain the behavioral patterns of the load,based on which this study proposes an integrated learning lighting load prediction method.The pattern labels of the predicted day are analyzed and acquired by K-means and Light GradientBoostingMachine(LightGBM)algorithms,and together with the lighting loads and climate characteristics,they are used as inputs to the Sequenceto Sequence(Seq2Seq)model to predict the lighting loads.By analyzing the real cases,the proposed method significantly improves the performance indexes of normalized root mean square error(NRMSE)and coefficient of determination(R2)by 18%and 8%,respectively,compared with the traditional method without behavioral pattern labels,which demonstrates the feasibility and superiority of the proposed method in the field of lighting load prediction.