At present,the rod parting research mainly focuses on the mechanism analysis after parting,with less involvement in parting prevention.The fault diagnosis of downhole working conditions mostly relies on indicator diagrams,which cannot diagnose the influence of wellbore friction.To solve these problems,based on the me-chanical model of the sucker rod,the differences between the theoretical calculation results and the field values of the polished rod loads of the fault wells were analyzed,and a modified mechanical model was proposed.The BP neural network method was used to conduct prediction,regression and verification on the influential factors,and the matching of the modified results with the field results was achieved.Finally,a diagnosis was conducted on the abnormal downhole working conditions.The research results show that the trained BP neural network model has a simple structure,is easy to operate,and has a diagnostic efficiency and high prediction accuracy,with a goodness of fit of over 88.96%.The calculated results of the modified mechanical model are in good agreement with the field measured values.The bias between the calculated results of 14 fault wells and the field measured results is all with-in 0.31%.The research conclusions provide theoretical support for the diagnosis of downhole rod parting faults.
关键词
抽油杆断脱/悬点载荷/力学模型/BP神经网络/故障诊断/影响因子
Key words
rod parting/polished rod load/mechanical model/BP neural network/fault diagnosis/influ-ential factors
维普
万方数据