Working Condition Recognition Method Of Pumping Unit Based On Improved Kernel Fisher Discriminant Analysis

At present,oil is an important chemical raw material and power energy,which plays an irreplaceable role.It is of great significance to identify the abnormal working conditions of the pumping unit in time to ensure the oil production,reduce the loss of equipment and improve the economic efficiency.With the development of Supervisory Control and Data Acquisition(SCADA)system,oil production process has saved a lot of production data,which provides favorable conditions for data driven fault diagnosis technology.Based on the real-time acquisition of pumping unit production data and considering the nonlinear characteristics of the process,this paper studies the identification method of pumping unit working condition based on improved kernel Fisher discriminant analysis(KFDA).And the working condition monitoring software of the pumping unit is developed.The main work of this paper is as follows:First,in view of the limited amount of label data and the large number of unlabeled samples that have not been fully utilized,the condition recognition method based on semi supervised local kernel Fisher discriminant analysis(SLKFDA)is studied in this paper.This method can utilize the supervised information of labeled samples and the global structure information of a large number of unlabeled samples at the same time,so as to prevent the KFDA model from overfitting and the singular problem of the class scatter matrix.The simulation results on typical industrial process data sets show that the proposed method has a better performance recognition performance than the KFDA.Secondly,in view of the problem that traditional algorithm can't dig out the high-order statistics of the variables and cause the waste of useful information,this paper studies the KFDA working condition recognition method based on statistical pattern analysis(SKFDA).Furthermore,combined with semi supervised local methods,this paper studies the SLKFDA working condition recognition method based on statistical mode analysis(SSLKFDA).This method can recognize the high order statistical information of the variable as working condition recognition feature,and use more abundant classification information to improve the ability of working condition recognition.The simulation results on typical industrial process data sets show that the proposed method has better performance on working condition recognition.Thirdly,this paper studies the feature extraction of electrical power data of pumping and the application of working condition recognition.In view of the different characteristics of the spectrum power of different power curves,the wavelet packet transform is used to decompose the electric power data,and the energy of multiple sub bands is taken as the feature set of working conditions.The research method is applied to the pumping power measured data,and achieve good recognition effect,and online monitoring process of pumping unit are discussed.Finally,the working condition monitoring software of the pumping unit is developed.In order to realize the running condition identification of the pumping unit,a monitoring software for the working condition of the pumping unit is developed by using the hybrid programming technology of Java and Matlab.The software not only can identify the working conditions of various algorithms,but also can identify the production data of pumping unit in real time.It can also draw the curve of its indicator diagram and electric power map,which is convenient for the operator to analyze.