基于栈式稀疏自编码器的抽油机故障诊断研究

Research on Fault Diagnosis of Pumping Unit Based on Stacked Spare Auto-encoder

为了及时发现抽油机故障,减少生产成本,提高生产效率,通过分析不同形状的抽油机示功图来及时准确地判断抽油机工作状况很有必要。传统人工识别方法不能实现抽油机工况实时诊断,而传统智能算法识别准确度低,故提出一种基于栈式稀疏自编码器的抽油机示功图识别方法,用于抽油机故障诊断。该方法通过栈式稀疏自编码器自动提取示功图数据深层可分性特征,然后利用学习到的特征结合对应的样本标签通过支持向量机进行有监督训练与分类。将采集的中原油田实测示功图对该方法进行实验,结果表明该方法具有较高的示功图识别速度和识别准确度。该方法为快速准确地进行抽油机故障诊断提供了参考。

In order to find out the fault of the pumping unit, reduce production cost and raise production efficiency, it is necessary to judge the working condition of the pumping unit in time and accurately by analyzing the different shape' s indicator diagrams of pum-ping units. The traditional artificial recognition method can' t realize the real-time diagnosis of the pumping unit working condition. The traditional intelligent algorithm has low recognition accuracy. Therefore, a method based on stacked sparse auto-encoder ( SSAE) for indicator diagram identification was proposed for fault diagnosis of pumping unit. In this method, the deep and separable features of indicator diagram data were automatically extracted by stack sparse self-encoder, then the learning features combined with the corre-sponding sample labels were used to carry out supervised training and classification through support vector machine. The measured indicator diagrams of Zhongyuan Oilfield were used to experiment with this method. The experimental results show that the method has high recognition speed and accuracy. The proposed method can help to diagnose the faults of pumping unit quickly and accurately.