摘要
针对站场油气储罐计量仪表的检测误差,以现场366座储罐正常计量数据为基础,设置0.5%~2.5%的误差率,构建数据驱动方法数据集,综合考虑误报率和窗口大小(25~125),针对7类算法(WIN-G、WIN-M、WIN-L、RuLSIF、KL-CPD、VAE和LSTM-VAE)进行综合性能评价,用现场检测的误差数据和正常数据进行算法验证,结果表明:在窗口大小100的条件下,LSTM-VAE算法性能最佳,正确报警率高于0.95。
Aiming at the detection error of the metering instrument for the oil and gas storage tank in the station,having the normal metering data of 366 tanks in the field based to set a 0.5%to 2.5%error rate and construct the data set of the data-driven method,including having both false alarm rate and window size(25 to 125)considered to comprehensively evaluate the performanceion of seven algorithms(WIN-G,WIN-M,WINL,RuLSIF,KL-CPD,VAE and LSTM-VAE)and make use both error data and normal data detected in-situ to verify the algorithm.The results show that,as for the window size of 100,the LSTM-VAE algorithm has the best performance and the accurate alarm rate is higher than 0.95.
作者
陈永久
陈思
王智慧
梁冰
CHEN Yong-jiu;CHEN Si;WANG Zhi-hui;LIANG Bing(China Petroleum Engineering&Construction Corp.North China Company)
出处
《化工自动化及仪表》
CAS
2024年第3期422-426,437,共6页
Control and Instruments in Chemical Industry
