School of Computer Science, North China Institute of Science and Technology
Hebei Internet of Things Monitoring Technology Innovation Center

近年来,随着大学宿舍空调的普及,学生宿舍的用电量也逐渐增加。空调作为宿舍耗能的主要组成部分,分析其用电特性及预测用电量,有助于用电安全管理和宿舍线路规划,也可以帮助同学们合理规划空调使用,预防电路过载导致安全隐患。本文提出了一种基于Bi-LSTM循环神经网络的高校宿舍空调短期用电预测方法。通过采集高校宿舍空调用电数据,并建立用于短期用电预测的数据集,使用Bi-LSTM网络提取用电量样本中的时序信息,并在网络中加入正则化,避免出现网络退化问题。在此基础上,利用该模型对未来一天的用电量进行预测。实验结果表明,该模型相对于传统的预测方法,在预测准确率和稳定性方面均有明显的提升。本文首先对数据集进行了预处理和特征提取,包括对数据的清洗、聚类、归一化处理以及特征选择等。随后,使用预处理后的数据训练了多种预测模型并进行对比。最终选择基于Bi-LSTM循环神经网络的模型来预测高校宿舍空调的短期用电。在实际应用中,该方法可以帮助高校宿舍管理部门更加准确地预测未来一天的空调用电量,从而合理安排用电计划和调整设备运行状态,达到节能减排和优化设备使用效率的目的。

In recent years, with the popularity of air conditioning in university dormitories, the electricity consumption of student dormitories has gradually increased.As the primary component of electricity consumption,analyzing the electricity consumption characteristics of air conditioning and predicting the electricity consump051tion can help with electricity safety management and dormitory wiring planning.In addition, it can also helpstudents to plan the use of air conditioning rationally, prevent safety risks caused by circuit overload.We propose a method for short-term electricity consumption prediction in college dormitory air conditioning based onBi-LSTM recurrent neural networks.By collecting data on electricity consumption of air conditioning in university dormitories and establishing a dataset for short-term electricity consumption prediction, we use a Bi-LSTM network to extract the temporal information of electricity consumption samples, while incorporating regularization into the network to avoid network degradation problems.Based on this, we use our model to makepredictions of electricity consumption for the next day.Experimental results show that the proposed model hasa clear improvement in prediction accuracy and stability compared to traditional prediction methods.We firstpreprocessed and extracted features from the dataset, including cleaning the data, clustering, normalizing thedata, and selecting features.Then, we trained various prediction models with the preprocessed data and compared them.Finally, we chose the Bi-LSTM recurrent neural network model to predict short-term electricityconsumption.In practical applications, this method can help dormitory management departments inuniversities to predict the future electricity consumption of air conditioning for the next day more accurately.Inturn, the administrators adjust equipment running status and rationalize the planning of electricity use to achieve energy conservation, emission reduction and optimization of equipment utilization efficiency.

中央高校基本科研业务费资助项目(3142021009);河北省物联网监控技术创新中心绩效后补助项目(21567693H);廊坊市科学技术局资助项目(2022011075)

张云雷,李子昂,马骧尧,李冬艳 .面向安全的高校宿舍空调用电预测方法研究[J].华北科技学院学报,2023,20(6):105-114

ZHANG Yunlei,LI Ziang ,MA Xiangyao ,LI Dongyan.Research on the prediction method of electricity consumption for air conditioners in college dormitories for safety[J].Journal of North China Institute of Scienceand Technology,2023,20(6):105-114