Fault Diagnosis Of Pumping Unit Based On Semi-supervised Extreme Learning Machine

Pumping wells will lead to lower production efficiency of the oilfield and other issues under the fault condition,thus affecting the economic benefits of oilfield.Therefore,the realtime monitoring of the operation of pumping well is of great significance for safe and efficient production.In recent years,digital oilfield is constructed and the real-time measurement data acquisition system is developed,which provides the conditions for the data-driven fault diagnosis of pumping unit.Compared with the traditional indicator diagram method using only the displacement and load data,this paper is based on the deep mining of the multi-source realtime measurement data.Aiming at the problems of huge data size,non-linear correlation and low data annotation,this paper is to study the fault diagnosis method based on semi-supervised extreme learning machine.The main works are as follows:Firstly,based on the brief description of the pumping unit working principle and the typical fault conditions,the wavelet packet transform method is used to extract the signal from the measures data of the pumping unit in this paper,and extract the different levels of the wavelet packet energy spectrum as the feature information,which are used to establish the fault diagnosis system of pumping unit.Secondly,aiming at the parameter optimization problem of ELM algorithm nonlinear classifier,a fault classification algorithm based on differential evolution extreme learning machine is studied.With the global optimization ability of the DE,DE-ELM is designed to optimize the initial value of the ELM network,meanwhile takes the classification errors as optimization goal to get the optimal initial values.Aiming at the problem of manual selections of the trail vector generation strategies and their associated control parameters in DE-ELM,the adaptive differential evolution extreme learning machine is studied,which can obtain the optimal results by adaptive selecting trial vector generation strategies and their associated control parameters,and then obtain optimal results.The simulation results show that the DE-ELM and SaDE-ELM algorithms have better classification performance than the basic ELM algorithm on the standard datasets.Then,considering the problem that the ELM algorithm fails to make full use of unlabeled data,a semi-supervised fault classification method Tri-SaDE-ELM based on Tri-Training training mechanism and self-adaptive differential evolution ELM is proposed.The method uses SaDE-ELM algorithm to design three base classifiers,and trains three different base classifiers based on Tri-Training mechanism,which makes full use of unlabeled data to improve the accuracy of fault classification.The simulation results show that the semi supervised ELM algorithm has better pattern classification performance on standard datasets.Finally,researches on the field data collected from the oil pumping wells are carried out.In this paper,the data of indicator diagram and electrical parameters are synthetically considered,and different ELM algorithms are applied to construct the fault diagnosis of the classifier to diagnose the real-time measurement data of the pumping unit.Industrial data application results show that the Tri-SaDE-ELM algorithm proposed in this paper can make full use of unlabeled data to provide better fault recognition results.