贝叶斯神经网络方法在套损预测中的应用研究

Bayesian neural network approach to casing damage forecasting

油气是国家发展的重大战略资源.随着近年来对油气的开发利用,几大油气田都出现了严重的套损问题,严重影响了油气的开发,造成了巨大的经济损失.因此,寻找一种合理有效的方法来对套管状况进行预测并针对的采取预防措施,对延长油水井的工作寿命、降低损失从而提高生产效率具有重要意义.首先介绍了贝叶斯神经网络的基本原理,以及ARD技术与模型比较技术;其次,与经典BP神经网络进行对照实验,验证贝叶斯神经网络在过拟合方面的优势;根据模型证据确定最优神经网络结构,整理大庆油田南一区西西区块部分井数据进行预测分析,获得套损因素相关权重.结果显示:油井与水井预测结果的准确率分别为82.86%和73.85%,具有较高的准确性;该区域套损的主控因素为高压注水导致的断层开启导水,油田方面应采取提高固井质量、选择合理的注水压力等措施来预防套管损坏.

Oil and gas are major strategic resources for national development. In recent years,with the exploitation for oil and gas,serious casing damage occurs in several oil fields,which affects the development of oil and gas badly,inflicts grievous economic harm.As a result,it has a great significance to extend the working time of oil well,reduces of losses and thus improve production efficiency by finding a reasonable and efficient prediction method for predicting casing situation and take relative precautions. Casing damage problem is a system with many affecting factors and complex mechanism, which has the characteristics like complexity,uncertainty,ambiguity, and qualitative analysis, is difficult to qualify. Firstly,introduces the basic principle of Bayesian neural network,ARD technique and mode comparison technique. Then,the use of Bayesian neural network should parallel experiment used for classical BP neural network to verify the advantages of Bayesian neural network in overfitting. The result shows that the Bayesian neural network has higher accuracy and lower generalization error.To determine the optimum neural network structure according to the value of mode evidence. Organize the data of part wells in Xixi Block in Nanyi Block in Daqing oil field and use these data to build a model including fourteen factors for water wells and eleven factors for oil wells served to forecasting casing damage and analyzing casing damage factors. Finally, gets related weights of casing damage factors. The results indicates that the accuracies of oil wells and water wells are 82. 86% and 73. 85%,which proves that this method has high accuracy and the result is credible; the main controlled casing damage factor for this area is the water diversion of fault resulted from high-pressure water injection. The oil field should take measures to prevent casing damage,such as improving cementing quality, selecting reasonable injection pressure and so on.