重庆理工大学学报2025，Vol.39Issue(5)：66-74,9.DOI:10.3969/j.issn.1674-8425(z).2025.03.009
多尺度通道注意力机制空调启停时间预测研究
Research on start and stop time prediction of multi-scale channel attention mechanisms for air conditioning
摘要
Abstract
Maintaining optimal production environments through continuous air conditioning operation is crucial for cigarette manufacturers to ensure product quality.However,determining efficient air conditioner operation schedules has become a significant challenge,as operators often struggle to optimize start/stop timing,resulting in energy wastage.Predicting air conditioner start/stop times is a vital component of energy consumption management and represents a prominent research focus in air conditioning energy efficiency. When environmental parameters,air conditioning systems,and indoor/outdoor conditions are known,analyzing these data enables the prediction of approximate air conditioning start/stop times.Recent research has demonstrated the feasibility of predicting these operational timings using air conditioning system data.This prediction approach differs from direct energy consumption forecasting,as it proactively guides system operation by anticipating optimal activation and deactivation times to achieve energy conservation. In comparison,traditional prediction methods typically focus either on mapping data feature relationships or implementing high-performance predictive models.However,these approaches often overlook inherent data characteristics and inter-variable dependencies,resulting in suboptimal performance when handling complex,highly coupled data.This limitation necessitates the development of a more comprehensive prediction methodology that considers both data-level internal feature relationships and modeling-level interactions to enhance feature extraction and data representation. To address these limitations,we introduces a novel prediction model incorporating a multi-scale channel attention mechanism.While time series prediction traditionally emphasizes overall temporal patterns,it often neglects individual feature interactions and channel information associations.Our proposed channel attention mechanism explicitly models inter-feature dependencies,enhancing the model's sensitivity to different channels and enabling it to emphasize crucial features while suppressing irrelevant ones.Furthermore,to preserve specific temporal characteristics(such as periodic and trend features)that might be lost when directly processing time series data through attention modules,the model preprocesses data by emphasizing periodic or trend components. The proposed approach mitigates attention computation sparsity and improves focus capture.The feature extraction process involves treating inputs as distinct features from the same sequence,processing them through shared convolutional layers,and combining weighted feature maps with original features and residual connections by employing a weighted cumulative fusion method.Experimental results demonstrate the model's superior performance,showing reductions of 16.67%,5.29%,and 20.15%in MSE,MAE,and MAPE metrics respectively when compared to existing prediction models.The improved accuracy can be attributed to the model's enhanced data exploration capabilities and multi-scale feature supplementation,which emphasizes distinctive features and enables more effective feature extraction and fusion.The implementation of these prediction results has led to significant reductions in energy consumption of workshop operations,contributing to the efforts to improve the energy efficiency.关键词
空调启停时间/数据分解/通道注意力机制/预测模型/节能优化Key words
air conditioning start/stop times/enhance feature extraction and data representation/channel attention mechanism/prediction model/energy savings分类
信息技术与安全科学引用本文复制引用
王华秋,谭佳豪..多尺度通道注意力机制空调启停时间预测研究[J].重庆理工大学学报,2025,39(5):66-74,9.基金项目
国家科技部重点研发计划项目(2018YFB1700803) （2018YFB1700803）
