【摘要】 针对油田抽油机井故障诊断方法较落后和故障信息采集不充分的问题,提出一种小波变换基神经网络故障诊断系统。它先对输入信号进行离散小波变换,把由Mallat算法得到的多尺度下的离散细节信号作为故障特征,之后将其输入到神经网络进行故障模式分类。为了进一步提高诊断的正确率,一方面对神经网络的结构进行优化,另一方面采用学习率自适应调整的共轭梯度法训练神经网络的权值。对某油田32口故障油井进行诊断,正确率在95%以上,这表明该方法的有效性。更多还原

【Abstract】 Aiming at the questions of pump-jack fault diagnosis method backward and diagnosis information lack of, it is proposed a kind of wavelet transform based neural network fault diagnosis system. The system first performs discrete wavelet transform for input signals and treats the multi-scale discrete detail signals achieved by Mallat arithmetic as fault characters. Then the transformed signals are sent to the neural network to classify the fault modes. In order to raise the correct rate of fault diagnosis, on one hand the structure of the neural network is optimized, and on the other hand conjugated gradient method is used with learning rate self-adjusting to train the network weight values. In the practice, the system is used to diagnose up to 32 fault pump-jacks, and the correct rate of diagnosis is above 95%. The results show that the proposed scheme is very effective.更多还原