UKF-RBF Neural Network Method Based On Noise Adaptive For Fault Diagnosis In Pumping Unit

The fault diagnosis for pumping unit is an important part of petroleum engineering.Due to the complex conditions of downhole conditions of pumping unit,and the entire production is characterized by strong coupling,nonlinearity,and dynamic time-varying parameters,so once a fault occurs,it is difficult to make a timely and accurate diagnosis,and may result in property loss or may even endanger personal safety.Therefore,it has important application value to establish an accurate and effective fault diagnosis system of pumping unit.In this paper,a fault diagnosis method of pumping unit based on adaptive UKF-RBF neural network is proposed,and the key technologies are studied.The main contents are as follows.(1)A method for extracting the indicator diagram features of an approximate polygon curve is proposed.The oilfield indicator diagram has 144 points of displacement and load coordinates,Directly inputting the dynamometer data into the model for fault diagnosis will result in a large amount of computation.In order to solve this problem,the feature extraction for approximate polygonal curve is proposed.Based on fully retaining the shape features of the dynamometer map,a large number of redundant points are removed and the Fourier transform time is reduced,thus providing more realistic and effective feature data for the fault diagnosis model.(2)A fault diagnosis model with adaptive UKF-RBF neural network is proposed.The modeling of RBF neural network is a static modeling,whose effect is limited to applying in the relatively stable industrial systems.When the system parameters are strongly coupled and timevarying,the accuracy of the model cannot be guaranteed.In order to solve this problem,a fault diagnosis model with RBF neural network is first established,and the RBF neural network is optimized with the unscented Kalman filter-UKF algorithm,then a dynamic evolution modeling is implemented.Meanwhile,during the operation,there is a large number of unknown and timevarying noises occurred in the pumping unit,and the fault diagnosis model with adaptive UKFRBF is established.The model is verified to have higher fault diagnosis accuracy with experiment,which meets the actual production requirements of the oilfield.(3)A fault diagnosis software for pumping unit based on adaptive UKF-RBF Model is developed.The proposed fault diagnosis model with adaptive UKF-RBF is written into the software system by using MATLAB and C# mixed programming.The diagnosis results of the data in the database of Tianjin Dagang Oilfield show that,the system can accurately diagnose the fault condition,and achieve the expected results.