基于BP神经网络的高含硫油井硫化物应力腐蚀预测

UTILIZE BP NERVE NETWORK TO PREDICT SULFIDE STRESS CORROSION IN HIGH SOUR OIL WELL

付亚荣 马永忠 吕德福 孙英祥 申玉健

Fu Yarong Ma Yongzhong La Defu Sun Yingxiang Shen Yujian

华北石油管理局采油五厂

基于BP神经网络技术具有较强的收敛性及自适应、自组织学习能力、较好的容错性,并行处理强、识别预测迅速准确、稳健性好的特点,以高含硫油井在含水2.4%~19.0%之间的实际硫化物应力腐蚀(SSC)速率作为训练样本,应用BP网络进行训练,达到精度要求后,对原样本进行回判模拟,再对只知输入信息而输出信息未知的样本进行预测。证明BP神经网络技术能够正确地预测高含硫油井的SSC,且精度高于GM(1,1)预测结果。其预测结果可用来指导油田的开发生产。

The BP nerve network technique has the following characteristics:the astringency is strong, the capability of self-adaptation and self-learning is strong, the error tolerance ability is better, the parallel processing capability is strong, identification and prediction is rapid and accurate and so on. This paper takes the actual sulfide stress corrosion rate (water cut is 2.4%~19.0% for the high sour oil well) as the training samples, uses the BP nerve network to train. When the precision requirements is achieved, the original samples are judged and simulated, and then prediction is conducted for the samples for which the input data are known but the output data are unknown. The results indicate that the BP nerve network technique can correctly predict the sulfide stress corrosion in the high sour oil wells, and the precision is higher than that of GM. The predicted result can be used to direct the development of the oilfields.

10.3969/j.issn.1000-7393.1999.06.022