Abstract: After the start of the heating furnace in the upstream station of the hot oil pipeline, the unsteady trend of the oil temperature in the downstream inlet station will affect the hot⁃washing operation of the pipeline before the pigging and the emergency treatment for abnormal shutdown. Accurate prediction of unsteady oil temperature trend changes can effectively improve the safe operation of pipelines and energy conservation and consumption reduction. A deep learning framework is adopted to take the unsteady process of upstream and downstream oil temperature as the mapping relationship from sequence to sequence, and the classical seq2seq algorithm of Recurrent Neural Network (RNN) was used to establish a predictive model of unsteady oil temperature. Through the correlation analysis, the parameters such as the unsteady oil temperature, pipeline flow and ground temperature in the upstream outbound station are selected as the model influence factors. In the SCADA system database of the HY hot oil pipeline, the relevant parameter historical data is downloaded as sample data, and after the data is preprocessed, the model is trained and tested. The error between the predicted value and the true value is ±0.2 °C, the correlation coefficient R and the root mean square difference RMSE are 0.96 and 0.99. At the same time, the verification model can effectively predict the unsteady oil temperature of the downstream station, which has high accuracy and good generalization.

Key words: Hot oil pipeline, Unsteady oil temperature, seq2seq, Forecast model

摘要： 热油管道上游站场启停加热炉后，下游进站油温的非稳态变化趋势影响管道清管前启炉热洗、异常停炉应急处置等工作，非稳态油温趋势变化的准确预测可有效提高管道安全运行与节能降耗工作的开展。为此，采用深度学习的框架，将上下游油温的非稳态过程看作序列到序列的映射关系，建立基于attention机制的seq2seq非稳态油温预测模型。通过相关性分析选取上游出站非稳态油温、管道流量、地温等参数作为模型影响因子，并计算获得相关度。在HY热油管道的SCADA系统数据库中下载相关参数历史数据，作为样本数据，经数据预处理后，对模型进行训练测试。将训练完成的模型应用于实际生产，获得预测值与真实值误差为±0.2 ℃，启停加热炉过程预测值与真实值的相关系数R分别为0.96和0.99，均方根误差分别为0.11和0.09，可见模型可有效预测下游站场非稳态油温。该模型基于实际生产数据驱动，为未来管道智能化控制奠定基础。

关键词: 热油管道, 非稳态油温, seq2seq, 预测模型