基于Transformer的空调能耗预测模型构建与参数优化
Modelling and Parameter Optimization of the Energy Consumption Prediction for HVAC System Based on the Transformer Neural Network
刘兴成1
作者信息
- 1. 中国电子科技南湖研究院,浙江 嘉兴 314001
- 折叠
摘要
针对空调系统能耗预测建模过程中的数据质量、模型输入参数筛选等问题,研究基于Transformer 神经网络的空调系统能耗预测模型构建和参数优化方法,结果表明:可以通过广义极端学生化偏差方法对数据中的离群值进行检测修正,从而提升数据质量;通过余弦相似度对输入参数进行两两相关性检验来消除各参数间的多重共线性,实现对输入参数的初步筛选;采用随机森林算法计算初选参数对空调能耗预测结果的影响来判断冗余参数,进而完成对输入特征参数的最终筛选;建立的空调能耗预测模型对数据测试集的预测结果均方根误差RMSE为38.831 kW,相关系数R2 为0.952,表现出了良好的预测性能.
Abstract
To address the data quality and the input parameters screening of the HVAC system energy consumption prediction,this research investigates the potential parameter optimization approach of the HVAC energy consumption prediction model based on the Transformer neural grid.The result indicates that a Generalized Extreme Studentized Deviate(GESD)algorithm can be used to find outliers,correct data,and thus improve data quality.The initial screening of the input parameters is conducted through a cosine similarity association test,which will reduce the multicollinearity between the parameters.To review and screen the final input feature parameters,the accuracy of the prediction results is determined using the Random Forest(RF)algorithm,which helps to eliminate extraneous elements.The final energy consumption prediction model of HVAC system shows good prediction performance on test dataset,with an R2of 0.952 for the test set of data and a root mean square error(RMSE)of 38.831 kW.
关键词
空调系统能耗预测/Transformer神经网络/数据质量/模型参数优化Key words
energy consumption prediction of HVAC system/Transformer neural networks/data quality/model parameter optimization引用本文复制引用
基金项目
上海市青年科技英才"扬帆计划"项目(21YF1460000)
出版年
2024
NETL
NSTL
万方数据