基于双向长短期记忆神经网络的冷库冷负荷预测算法OA
Cooling load forecasting algorithm of cold storage based on bidirectional long short-term memory neural network
针对小型氟利昂冷库冷负荷难以准确预测、制冷能耗浪费的问题,提出了一种融合Transformer与双向长短期记忆神经网络(BiLSTM)的改进型冷库冷负荷预测模型(Transformer-BiLSTM-PLO-TSA).该模型以BiLSTM与全连接层代替Transformer原始解码层,引入极光优化算法(PLO)完成模型参数寻优,并嵌入时空注意力机制(TSA)挖掘时序数据的时空关联特征,提高模型的鲁棒性.实验结果表明,所提出的PLO优化时空注意力融合模型Transformer-BiLSTM-PLO-TSA的RMSE为2.57,MAE为1.54,决定系数为0.94,在精度和稳定性方面优于BiLSTM、Transformer、Transformer-LSTM、Transformer-BiLSTM、Transformer-BiLSTM-TSA 等对比模型.
To address the challenges of accurate cold load prediction and energy consumption waste in small-scale Freon-based cold storage systems,an improved cold load forecasting model integrating Transformerand bidirectional long short-term memory neural networks(BiLSTM)(Transformer-BiLSTM-PLO-TSA)is proposed.This model replaces the original Transformer decoding layer with BiLSTM and fully connected layers,incorporates the polar light optimization algorithm(PLO)for parameter optimization,and integrates a spatio-temporal attention mechanism(TSA)to extract temporal-spatial correlation features from time-series data,thereby enhancing model robustness.A validation study demonstrated that the Transformer-BiLSTM-PLO-TSA model achieved RMSE of 2.57,MAE of 1.54,and R2 of 0.94,outperforming competing models such as BiLSTM,Transformer,Transformer-LSTM,Transformer-BiLSTM,and Transformer-BiLSTM-TSA in both accuracy and stability.
王澄溪;姜珊;孙道鹏;周鹏
大连工业大学信息科学与工程学院,辽宁大连 116034大连工业大学信息科学与工程学院,辽宁大连 116034大连冰山集团工程有限公司,辽宁大连 116100大连亚尼自控技术有限公司,辽宁大连 116031
信息技术与安全科学
冷负荷预测双向长短期记忆网络时空注意力机制极光优化算法
cold load forecastingBiLSTMtemporal-spatial attentionpolar lights optimization
《大连工业大学学报》 2026 (4)
285-292,8
评论