Research On Fault Diagnosis Method For Rod Pumping Wells Based On Stochastic Configuration Networks

China’s oil extraction work is proceeding vigorously,and the pumping unit is an important equipment for oil extraction.Once a malfunction occurs,not only will oil extraction be unable to proceed in an orderly manner,causing economic losses,but it may also lead to safety accidents.Therefore,achieving fast and accurate diagnosis of pumping units is of great significance for reducing losses,avoiding safety accidents,and ensuring oilfield production.Currently,the fault diagnosis method for pumping units relies on identifying the shape of the dynamometer diagram to judge the working condition of the pumping unit.Traditional methods usually rely on manually extracting geometric features of the dynamometer diagram for classification and identification,but there are problems with feature errors and low classification accuracy.Intelligent algorithms can extract various features of the dynamometer diagram well,but different datasets have different effects on different training networks in classification models.Research on mixed faults also faces challenges such as insufficient data and difficulty in classification.In response to these problems,this thesis studies the following:(1)Research on a pumping unit fault diagnosis method based on feature fusion SCN.Firstly,in response to the problem of errors in manually extracting features,this thesis fuses texture features with contour features and proposes the Stochastic Configuration Networks(SCN)model for feature fusion.Comparative experiments using this model show that the fused feature vector can better represent the feature information of the dynamometer diagram.Secondly,in order to classify and identify the dynamometer diagram more efficiently and accurately,this thesis proposes to integrate the SCN with ensemble learning to form the SCN-ensemble learning model,which makes the network more compact and improves accuracy through voting.The experimental results show that this model is effective and feasible for fault diagnosis of pumping units with a sucker rod pump.(2)Research on a fast pumping unit fault diagnosis method based on BSCN-ensemble learning.Firstly,in response to the long training time of the SCN-ensemble learning model,this thesis proposes the BSCN-ensemble learning model based on the Block-Incremental Stochastic Configuration Networks(BSCN),and comparative experiments show that the BSCN-ensemble learning model can speed up training time and improve accuracy.Secondly,in order to enhance the texture features of the dynamometer diagram,this thesis proposes to fill the data form of the dynamometer diagram,and experimental results show that filling the data of the dynamometer diagram can achieve higher accuracy in the model.(3)Research on a fault diagnosis method for mixed faults in pumping units.Due to the difficulties in obtaining data,limited sample size,and complexity of classification tasks in research on mixed faults in pumping units,this thesis uses a combination of virtual and real data as the dataset;processes the data using the form of filled dynamometer diagrams and feature fusion;and uses the BSCN-ensemble learning model for classification.Experimental results demonstrate the feasibility of data combination and the accuracy of the model.