摘要
供热负荷预测是指导供热运行管理和调度的前提。供热负荷预测是一个时间序列预测问题,要求利用可用的历史记录和天气信息,预测未来24 h的实时供热负荷。该文提出一种基于精心设计的串联长短期记忆(LSTM)递归神经网络的短期供热负荷预测模型。展示数据预处理的过程,并设计损失函数以提高模型的性能。此外,还将集成策略与LSTM模型结合,以增强其泛化能力和鲁棒性。在离线(历史)测试数据上,所提出的模型能够进行令人满意的预测,满足本地电厂的需求。除离线测试外,还将该模型应用于山东省某电厂的在线系统。在2018年的供暖季节中,该模型在没有人工干预的情况下连续进行预测,持续4个月。该模型在在线测试中的表现与使用历史数据进行的离线实验结果相当,取得令人满意的测试效果。
Heating load forecasting is a prerequisite for guiding heating operation management and scheduling.Heating load forecasting is a time series forecasting problem that requires us to use available historical records and weather information to predict the real-time heating load for the next 24 hours.In this paper,a short-term heating load forecasting model based on a carefully designed tandem long short-term memory(LSTM)recurrent neural network was proposed.We demonstrated the process of data preprocessing and design the loss function to improve the performance of the model.We also combined the ensemble strategy with the LSTM model to enhance its generalization ability and robustness.On the offline(historical)test data,the proposed model is able to make satisfactory predictions to meet the needs of the local power plant.In addition to the offline test,we applied the model to the online system of a power plant in Shandong Province.During the heating season of 2018,the model continuously made predictions without human intervention for four months.The performance of the model in the online test is comparable to the offline experimental results using historical data,achieving satisfactory test results.
出处
《科技创新与应用》
2025年第26期139-142,共4页
Technology Innovation and Application
