摘要
压缩机是输气管道最主要的耗能设备,其流量计量结果的准确与否直接关系到管道全线的管理水平。在现场测试实验的基础上,构建了基于随机森林的数据驱动软测量模型,采用网格搜索和交叉验证对影响模型精度的超参数进行寻优,最终构建最优预测模型,并与支持向量机(SVM)模型、朴素贝叶斯(NB)模型和经网格搜索确定超参数的支持向量机(GS-SVM)模型的预测结果进行对比。结果表明,通过基尼指数确定的燃料气消耗量和大气压力对压缩机流量的相关性较小,其变量应予以删减;当决策树的个数n为300,分裂特征数m为5时的预测效果最佳;随机森林模型的均方根误差(RMSE)、平均绝对百分比误差(MAPE)均最小,说明了随机森林模型对于压缩机流量这类复杂非线性数据集的回归效果较好,具有一定的先进性和科学性。
Compressor is the main energy consuming equipment of gas transportation pipeline.The accuracy of its flow measurement results is directly related to the management of the whole pipeline.On the basis of field test experiments,a data-driven soft sensing model based on random forest is constructed.The hyperparameters that affect the accuracy of the model are optimized by grid search and cross-validation.The optimal prediction model is constructed finally,and the prediction results are compared with those of SVM,NB,GS-SVM and other models.The results show that the correlation between fuel gas consumption and atmospheric pressure on compressor flow is small through Gini index,and its variables should be eliminated.When the number of decision tree n is 300,and the number of split features m is 5,the prediction effect is the best.The RMSE and MAPE of the random forest model are both the smallest,which indicates that the model has a good regression effect for complex nonlinear data sets such as compressor flow,and has certain advanced and scientific nature.
作者
孙海芳
胡天亨
马亚欣
刘忠刚
梁昌晶
Sun Haifang;Hu Tianheng;Ma Yaxin;Liu Zhonggang;Liang Changjing(Kazakhstan-China Gas Pipeline Project of Sino-Pipeline International Company Limited,Beijing,100029,China;Sichuan Tianyurui Group Co.Ltd.,Chengdu,610000,China;China Municipal Engineering Northwest Design and Research Institute Co.Ltd.,Lanzhou,730000,China;No.5 Oil Production Plant of Huabei Oilfield Company,CNPC,Xinji,052360,China)
出处
《石油化工自动化》
CAS
2024年第1期21-24,共4页
Automation in Petro-chemical Industry
