基于时段敏感权重组合的LightGBM和LSTM冷负荷预测方法  

LightGBM and LSTM Cooling Load Prediction Method Based on Time Period Sensitive Weight Combination

作　　者：陈适铭 CHEN Shiming(Zhuhai Pilot Technology Co.,Ltd.,Zhuhai,Guangdong 519000,China)

机构地区：[1]珠海派诺科技股份有限公司,广东珠海519000

出　　处：《自动化应用》2024年第20期111-114,共4页Automation Application

摘　　要：冷负荷预测是中央空调系统节能控制的基础。为进一步提升冷负荷预测的精度,提出了加权组合轻梯度提升机(LightGBM)模型和长短期记忆(LSTM)网络的预测方法,并在不同时段分配不同权重。首先,对冷负荷数据、室外温度、室外湿度进行数据预处理,分别按照LightGBM模型和LSTM网络的输入格式进行输入训练;其次,对验证集上的评估结果进行超参数调整,再将验证集划分为不同时段,使用最优化算法获得各时段最优的组合权重;最后,使用实际冷负荷数据进行算例分析。结果表明,所提方法能在不同时段有效利用2种模型的优点,具有较高的预测精度。Cold load prediction is the foundation of energy-saving control for central air conditioning systems.To further improve the accuracy of cooling load prediction,a weighted combination of LightGBM model and Long Short Term Memory(LSTM)network prediction method were proposed,and different weights were assigned at different time periods.Firstly,preprocess the cooling load data,outdoor temperature,and outdoor humidity,and train them separately according to the input formats of LightGBM model and LSTM network.Secondly,hyperparameter adjustments are made to the evaluation results on the validation set,and the validation set is divided into different time periods.Optimization algorithms are used to obtain the optimal combination weights for each time period.Finally,use actual cooling load data for case analysis.The results indicate that the proposed method can effectively utilize the advantages of the two models at different time periods and has high prediction accuracy.

关 键 词：冷负荷预测 分时段 加权组合预测 轻梯度提升机 长短期记忆网络 

分 类 号：TU831[建筑科学—供热、供燃气、通风及空调工程]