为优化天然气处理厂生产计划,弥补天然气处理厂负荷率预测模型的空缺,提出一种基于贝叶斯优化(Bayesian Optimization,BO)和长短期记忆(Long ShortTerm Memory,LSTM)的天然气处理厂负荷率预测模型。LSTM模型用于捕捉因检修天数和日处理量等因素引起的时间特征,贝叶斯算法用于优化LSTM网络的结构、隐藏层层数、隐藏层神经元个数、初始学习率和正则化系数,弥补参数造成预测波动的缺陷。选取波动型和平稳型天然气处理厂负荷率对不同模型测试,结果表明,LSTM模型较其他传统预测模型的预测精度高。BOLSTM模型的预测平均绝对误差(MAE)值和均方根误差(RMSE)值均最小,预测精度最高,通用性强,较传统LSTM模型的MAE和RMSE值可提高57.8%和30.1%。天然气处理厂负荷率预测模型可为天然气处理厂的生产运行和决策提供数据支撑,具有稳定的预测精度和适应性。

To optimize the production schedule of natural gas treatment plants and fill in a gap in the load rate prediction model, this study introduces a natural gas treatment plant load rate prediction model based on Bayesian optimization and long short-term memory (BO-LSTM). The LSTM model captures temporal patterns arising from maintenance schedules and daily processing volumes. Concurrently, the Bayesian optimization refines the LSTM network's structure, hidden layers, neuron counts, initial learning rate, and regularization coefficient, mitigating prediction fluctuations due to parameter variations. The load rates of treatment plants are classified into fluctuating type and stable type. The results indicate that the LSTM model outperformed other traditional forecasting models regarding prediction accuracy. Furthermore, the BO-LSTM model stands out, boasting the lowest mean absolute error (MAE) and root mean squared error (RMSE) values, which translates to superior prediction accuracy and robust versatility. Impressively, the BO-LSTM model achieves a 57.8% improvement in MAE and a 30.1% boost in RMSE values when compared with the conventional LSTM model. This advanced load rate prediction model offers data support for the operational and decisionmaking processes of the treatment plant, ensuring consistent prediction accuracy and adaptability.