基于BP人工神经网络的钻柱动态行为预测方法  

Prediction method of drilling string dynamic behaviors based on BP artificial network

作　　者：王潜龙[1] 冯全科[1] 屈 展[2] 

机构地区：[1]西安交通大学能源与动力工程学院 [2]西安石油学院

出　　处：《石油机械》2001年第12期6-8,10,共3页China Petroleum Machinery

基　　金：国家自然科学基金(59774006)资助项目

摘　　要：为了掌握常见的钻柱动态行为及产生原因,应用BP三层人工神经网络模型,采用频谱分布特征、轴心轨迹和振动特征等共15个输入征兆来研究预测4种典型的钻柱动态行为。8组输入征兆的预测输出结果表明,这种人工神经网络模型通过样本训练,能够根据钻柱输入征兆完全、不完全信息,进行联想、记忆和聚类等自组织学习,实现了网络输出准确率高,且网络输出结果与实际测试资料可以互相解释印证,说明这种神经网络的预测结果可以指导实际钻井作业,同时又可由实际钻井资料收集样本模式不断反馈,从而完善网络模型。To analyze common dynamic behaviors and influencing factors, based on three layers BP artificial network, and by using 15 input symptoms such as spectrum distribution patterns, axes trajectory and vibrating patterns, 4 typical movements of the drilling string are predicted and analyzed. The output predicting results with 8 groups of input data show that, by training specimen, the artificial network model can study self-systematically by means of association, memory, classifying and so on according to the input information of the drilling string. This improves the accuracy of the output and output results can be validated with the actual testing data. The predicting results can be used in drilling engineering and the network model can be improved by training new actual drilling data.

关 键 词：钻柱 动态预测 BP人工神经网络 模型 石油钻井 

分 类 号：TE921.1[石油与天然气工程—石油机械设备]