军事知识服务平台
Transformer-FNN地铁站空调冷负荷预测
Transformer-FNN model for cooling load forecasting in subway stations
摘要 地铁站空调系统能耗高,冷负荷预测是优化控制的基础.针对传统模型难以兼顾长期时序依赖与局部非线性特征的问题,提出基于XGBoost特征筛选的Transformer-FNN混合预测模型.采用XGBoost评估多源传感器变量重要性并筛选关键输入,利用Transformer提取长期时序特征,以FNN增强局部非线性映射,并用ADAM优化参数.基于佛山市某地铁站实测数据验证,结果表明:与FNN、LSTM与LSTNet相比,所提模型在多指标维度下表现出更优的泛化性与鲁棒性,可为地铁站空调系统的节能控制与能源优化调度提供负荷预测支持.
Abstract:
The air-conditioning system of a subway station consumes substantial energy,and cooling-load forecasting un-derpins optimized control.To address the inability of conventional models to capture both long-term temporal dependence and local nonlinear characteristics,an XGBoost-based feature-selection Transformer-FNN hybrid model is proposed.XGBoost e-valuates the importance of multisource sensor variables and selects key inputs,Transformer extracts long-term temporal features,FNN strengthens local nonlinear mapping,and ADAM optimizes model parameters.The model is evaluated using measured data from a subway station in Foshan.Compared with FNN,LSTM,and LSTNet,the proposed model achieves lower MAE and RMSE and higher R2 under both single and two chiller operating conditions.It can provide reliable load forecasting support for the energy-saving control and optimal energy scheduling of air-conditioning systems in subway stations.
刊 名 低温与超导 PKU
年,卷(期): 2026, 54(7)
分类号: TP273
机标分类号: TU831.2 TP391 TU201.5
在线出版日期: 2026年8月20日
基金项目: 62371032:国家自然科学基金4232021:北京市自然科学基金2019-K-149:住房和城乡建设部科学技术项目GJZJ20220803:北京建筑大学高级主讲教师培育计划项目,