【作者】 李文科； 钟向宏； 赵劲松；

【Author】 LI WenKe1, ZHONG XiangHong2 ,ZHAO JinSong1 (Department of Chemical Engineering, Tsinghua University, Beijing 100084; MaoMing Petrochemical Company Maoming ,GuangDong,525000)

【机构】 清华大学化工系； 茂名石化；

【摘要】 多牌号产品生产过程经常涉及到牌号过渡,如果采用传统的单一方法,对于其生产过程的故障检测和诊断无法得到满意的结果。本文将人工神经网络(ANN)用于牌号识别,提出了牌号识别和主成分分析(PCA)相结合的方法,即利用历史数据建立各个牌号的神经网络模型,在线数据经过BP神经网络(BPNN)识别,确认牌号类型后,调用对应牌号的PCA模型进行故障检测和诊断。结果表明,BPNN不仅可以准确识别牌号,识别率较规格界限法更高,而且能用于牌号过渡过程的判断。另外,与不进行牌号识别,而仅采用单一牌号正常样本或者所有牌号正常样本混合建立的PCA模型相比较,采用牌号识别后进行故障检测时的精度更高,证明了该方法的有效性。更多还原

【Abstract】 Multi-grade production process are often involved in grade transition, satisfactory results can not be obtained if a single method of fault detection and diagnosis is used in the process. In this paper, with artificial neural network (ANN)being used for grade recognition, a approach of grade recognition combined with principal component analysis (PCA) is proposed, namely, historical data is used to establish neural network model of various grades, on-line data is recognized to confirm the grade type by using BP neural network (BPNN),and then PCA models are called corresponding to the grades to perform fault detection and diagnosis. The results show that, BPNN not only can accurately recognize the grade type, and recognition rate is higher than the specification limits method, but also can be used to determine the grade transition process. In addition, compared to PCA-based fault detection without grade recognition using PCA model established by adopting normal samples of a single grade or mixed normal samples of all grades, PCA-based fault detection after grade recognition is more accurate, and effectiveness of the approach is proved.更多还原