基于MI+PSO-LSTM的能耗预测模型  

Energy consumption prediction model based on MI+PSO-LSTM

作　　者：谌东海 王伟 赵昊裔 明新淼 CHEN Dong-hai;WANG Wei;ZHAO Hao-yi;MING Xin-miao(The Urban Planning and Architectural Design Institute,CISPDR Corporation,Wuhan 430010,China;School of Machinery and Automation,Wuhan University of Science and Technology,Wuhan 430080,China)

机构地区：[1]长江勘测规划设计研究有限责任公司城市规划与建筑设计院,湖北武汉430010 [2]武汉科技大学机械自动化学院,湖北武汉430080

出　　处：《计算机工程与设计》2022年第10期2889-2896,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61902285)。

摘　　要：为实现智能楼宇的高效节能,提出一种短期建筑能耗组合预测模型。为在保证模型预测精度的同时减少模型训练时间,通过MI选取对能耗预测有效且关键的特征参数。利用粒子群优化算法(PSO)对长短时神经记忆网络(LSTM)的超参数进行优化,使选择的特征与网络拓扑结构相匹配,提高LSTM模型预测的精度和鲁棒性。实验结果表明,与传统ARIMA、KNR、单一LSTM模型相比,提出的MI+PSO-LSTM模型具有更高的预测精度和更稳定的预测性能。To realize the high-efficiency and energy-saving of smart buildings,the energy consumption forecast of the building was studied,and a short-term energy consumption combination forecasting model was proposed.To reduce model training time while en-suring the accuracy of model prediction,MI was used to selcct effective and key feature parameters for energy consumption prediction.The particle swarm optimization algorithm(PSO)was used to optimize the hyperparameters of the long and shortterm neural memory network(LSTM)to match the selected features with the network topology and improve the accuracy and robustness of the LSTM model prediction.Experimental results show that,compared with traditional ARIMA,KNR,and single LSTM models,the MI+PSO-LSTM model has higher prediction accuracy and more stable prediction performance.

关 键 词：能耗预测 互信息 特征选择 超参数 长短时记忆网络 粒子群优化算法 单步预测 

分 类 号：TP130[自动化与计算机技术—控制理论与控制工程]