Application Of LSTM Neural Network Model In Fault Detection Of Pumping Units

At present,deep learning technology is widely used,and has achieved remarkable results in various fields.The LSTM model is one of them.Its use of its strong timing prediction ability has been favored by many scholars.LSTM is a variant of recurrent neural network,which makes up for the problems of gradient vanish and gradient explosion of RNN,and lack of long-term memory ability.It can really use time series information effectively.Based on the advantages of the LSTM neural network model for predicting problems,this paper applies it to the actual problem of real-time fault detection in pumping units,and solves the problem of oilfield production loss caused by downhole faults.Firstly,the theory of neural network feature extraction algorithm and LSTM neural network model are studied,which lays a theoretical foundation for further application research;the characteristics of dynamometer diagram of pumping units are analyzed,and the characteristics of dynamometer diagram of common faults of pumping units are summarized.Secondly,the goal of fault detection is the dynamometer diagram.The feature extracted from dynamometer diagram will affect the final detection results.Based on this,a holistic method for extracting features of the dynamometer diagram using BP neural network dimension reduction is proposed.This method regards the dynamometer diagram as a graph composed of discrete points,and then extract the features of the dynamometer diagram from the perspective of the whole dynamometer diagram.Using the working principle of BP neural network,abstract layer by layer,abstract high-dimensional features into low-dimensional features,thereby extracting the features of the dynamometer diagram.The advantage of this method is that the extracted dynamometer diagram features are very representative.Thirdly,the application of the LSTM neural network model to the prediction problem is analyzed,and a real-time fault detection method for pumping units based on the LSTM neural network model is proposed.Based on the feature extraction algorithm,this method uses the LSTM neural network model to make a one-step prediction of the characteristics of the dynamometer diagram,and then uses the box plot method to formulate the interval value of the failure-free occurrence of the features of the dynamometer diagram without failure.The result is compared with the interval value to complete the fault detection.Finally,the prediction results of the proposed method and the existing regression model are compared to verify the effectiveness and feasibility of the feature extraction method based on BP neural network and the fault detection method of the pumping units based on the LSTM neural network model.