基于长短期记忆神经网络的工厂冷水机组短期功率预测  

Short-Term Power Prediction of Plant Chiller Based on Long Short-Term Memory Neural Network

作　　者：王宜卿 叶明树 任义成 陈焕新[1] 程亨达 WANG Yiqing;YE Mingshu;REN Yicheng;CHEN Huanxin;CHENG Hengda(School of Energy and Power Engineering,Huazhong University of Science and Technology,Wuhan 430074,Hubei,China;Xiamen Jinming Energy Saving Technology Co.,Ltd.,Xiamen 361000,Fujian,China)

机构地区：[1]华中科技大学能源与动力工程学院,湖北武汉430074 [2]厦门金名节能科技有限公司,福建厦门361000

出　　处：《制冷技术》2025年第1期55-60,共6页Chinese Journal of Refrigeration Technology

基　　金：国家自然科学基金(No.51876070)。

摘　　要：提出了一种基于长短期记忆网络(LSTM)的预测模型,实现对某法拉电子厂房空调冷水机组功率的短期预测,并通过与多元线性回归模型进行对比验证其准确性。为了进一步提升模型的预测性能,对网络结构进行了优化以获得最优的预测模型。结果表明:当LSTM模型隐藏层数为2,隐藏层神经元数为120时,模型预测精度最高,其均方根误差(RMSE)为5.644,决定系数(R^(2))为0.921,说明LSTM模型能够较好完成对冷水机组功率的预测。A prediction model based on long short-term memory(LSTM)is proposed to realize the short-term prediction of the chiller in a farad electronic plant,and its accuracy is verified by comparing with the multiple linear regression model.In order to further improve the prediction performance of the model,the network structure is optimized to obtain the optimal prediction model.The results show that when the number of hidden layers of LSTM model is 2 and the number of neurons in the hidden layer is 120,the prediction accuracy of the model is the highest,with RMSE(root mean square error)of 5.644 and R^(2)(determination coefficient)of 0.921,which indicates that LSTM model can predict the power of the chiller well.

关 键 词：冷水机组 长短期记忆网络 功率预测 决定系数 

分 类 号：TB611[一般工业技术—制冷工程] TQ051.5[化学工程]