中文摘要：

及时发现并排除抽油机故障对于降低生产成本、提高油井产量具有重要作用。由于抽油机故障种类多,测量的悬点示功图形态多样且容易受环境噪声的影响,测量数据的区分性特征难以提取,造成通过传统神经网络进行故障诊断时的准确率较低。为提高抽油机故障诊断的精度,提出一种基于深度信念网（DBN）和支持向量机（SVM）混合模型的抽油机故障诊断方法。采用深度信念网从样本示功图图像中学习特征,支持向量机根据特征判断抽油机的故障类别。深度信念网和支持向量机的结构参数均使用网格寻优的方法进行优化。实验结果表明,DBN和SVM方法避免了复杂的人工提取数据特征的过程且具有较高的识别准确率和识别速度,同时与其它方法相比具有更好的性能。

英文摘要：

Discovering and eliminating the pumping unit fault timely plays an important role in reducing the production cost and improving the production of the oil well. There are so many kinds of fault types and the measured indicator diagrams are varied in shapes and vulnerable to be influenced by ambient noises. It is difficult to extract the feature of the measured data for distinguishing,so that the accuracy rate is lower when using the traditional neural network for fault diagnosis. In order to improve the accuracy of pump unit fault diagnosis,a method to diagnose the fault diagnosis was proposed based on the hybrid model of deep belief network（ DBN） and support vector machine（ SVM）. Firstly,the deep belief network was used to extract features of the indicator diagram images. Then the support vector machines was used to identify the fault types based on the extracted features. Structure parameters of both DBN and SVM were optimized using the grid optimization method. The experiment results show that,it can avoid the process of complicated artificial extracting the characteristics of the data in this method. It has a higher recognition accuracy and recognition speed than other methods.