| 熊伟,何彦霖,宋伟,张厚望,尹爱军.极端梯度提升与随机森林融合的天然气露点预测方法[J].装备环境工程,2022,19(6):133-140. XIONG Wei,HE Yan-lin,SONG Wei,ZHANG Hou-wang,YIN Ai-jun.Prediction Method of Natural Gas Water Dew Point Based on the Fusion of eXtreme Gradient Boosting and Random Forest Regression[J].Equipment Environmental Engineering,2022,19(6):133-140. |
| 极端梯度提升与随机森林融合的天然气露点预测方法 |
| Prediction Method of Natural Gas Water Dew Point Based on the Fusion of eXtreme Gradient Boosting and Random Forest Regression |
| |
| DOI：10.7643/issn.1672-9242.2022.06.000 |
| 中文关键词: 三甘醇脱水装置 天然气水露点 极端梯度提升(XGBOOST) 特征提取 随机森林(RF)中图分类号:TB115 文献标识码:A 文章编号:1672-9242(2022)06-0133-08 |
| 英文关键词:triethylene glycol dehydration unit gas water dew point extreme gradient boosting (XGBOOST) feature extraction random forest (RF) |
| 基金项目:重庆市科技重大主题专项重点研发项目(cstc2018jszx-cyztzxX0032)；中国石油重庆气矿科研项目(K20-15) |
|
|
| Author | Institution |
| XIONG Wei | Chongqing Gas District, Southwest Oil and Gasfield Company, Chongqing 400021, China |
| HE Yan-lin | College of Mechanical and Vehicle Engineering, Chongqing University, Chongqing 400044, China |
| SONG Wei | Chongqing Gas District, Southwest Oil and Gasfield Company, Chongqing 400021, China |
| ZHANG Hou-wang | Chongqing Gas District, Southwest Oil and Gasfield Company, Chongqing 400021, China |
| YIN Ai-jun | College of Mechanical and Vehicle Engineering, Chongqing University, Chongqing 400044, China |
|
| 摘要点击次数: |
| 全文下载次数: |
| 中文摘要: |
| 目的 解决目前水露点数据多为人工采用测量仪器测得，时效性低且成本高昂的问题。方法 建立一种基于极端梯度提升(XGBoost)和随机森林(RF)的天然气水露点预测方法。采用XGBoost方法对所有监测工艺参数进行分析，筛选出主要影响水露点的关键工艺特征参数，以排除无关特征参数对预测的干扰。建立RF预测模型，输入关键特征集参数，实现对水露点的实时预测。以重庆气矿某脱水监测系统监测数据与生产数据为例，对所提预测方法进行对比分析验证。结果 相较于XGBoost、SVM等预测方法，RF模型具有最佳的预测性能，且经过XGBoost特征选择后，RF预测结果的MAE值降低了0.016 9 ℃，RMSE值降低了0.014 6 ℃。结论 基于极端梯度提升与随机森林融合的水露点预测方法具有更优的预测精度与鲁棒性，对指导脱水现场生产具有积极作用。 |
| 英文摘要: |
| Aiming at the problems that the current water dew point data are mostly manually measured with measuring instruments, the timeliness is low at the time with the high cost, this paper establishes a prediction method natural for gas water dew point based on extreme gradient boosting (XGBoost) and random forest (RF). This paper analyzes all the monitored process parameters by using the XGBoost method, and filtrates the pivotal process characteristic parameters that mainly affect the water dew point to eliminate the interference of irrelevant typical parameters to the prediction; the RF prediction mode is established, and the typical characteristic parameters are inputted to realize the real-time prediction of the water dew point. Taking the monitoring data and production data of a dewatering monitoring system in the Chongqing gas mine as an instance, this paper compares and analyzes the proposed prediction method. Compared with the other prediction methods, such as XGBoost and SVM, RF model has the best prediction performance, and after XGBoost feature selection, the MAE value and RMSE value of RF prediction results are reduced by 0.016 9 ℃ and 0.014 6 ℃ respectively. The results show that the water dew point prediction method based on the fusion of eXtreme Gradient Boosting and Random forest regression has better prediction accuracy and robustness. What's more, it has a positive effect on guiding dehydration on-site production. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
