基于RS-LVQ的同井注采系统故障诊断研究

RS-LVQ-based Fault Diagnosis of the Injection-production System

针对有杆泵同井注采系统的非常规结构,提出了一种反推计算方法,建立并求解了其故障诊断模型;然后应用不变矩特征法提取注入泵功图的7个不变矩特征,建立了基于RS-LVQ的有杆泵同井注采系统井下故障诊断系统。在保持分类能力不变的前提下,应用SOM神经网络对原始特征数据进行离散化处理,并应用RS理论对其进行属性约简;建立LVQ故障诊断子系统后,输入约简决策表进行学习训练。实例分析结果表明:SOM神经网络解决了决策系统中连续属性值的离散化问题;RS属性约简不仅可以提高诊断效率,而且简化了故障模式识别的LVQ网络结构;LVQ作为后端,因其有很强的非线性映射能力,保证了诊断结果的精度。因此,该诊断系统能够正确而且高效地对有杆泵同井注采系统进行故障诊断。

Aiming at the unconventional structure of the sucker rod pump injection-production system,an inverse calculation method was proposed. The fault diagnosis model was established and solved. Invariant moment feature method was used to extract seven invariant moments of the injection pump indicator diagram,so as to establish the downhole fault diagnosis system for sucker rod pump injection-production system based on RS-LVQ. Keeping the classification ability unchanged,the SOM neural network was used to discretize the original feature data and attribute reduction was carried out by using RS theory. LVQ fault diagnosis subsystem was established and was input with the reduction decision table for learning and training. The analysis results of case study showed that,SOM neural network addressed the problem of discretization of continuous attribute values in decision system. RS attribute reduction could not only improve the diagnostic efficiency but also simplify the LVQ network structure of fault pattern recognition. LVQ,as back end,ensures the accuracy of diagnostic results owing to its strong non-linear mapping capability. Therefore,the developed diagnostic system can correctly and efficiently troubleshoot the sucker rod pump injection-production system.