Research On Fault Diagnosis Algorithm Of Electric Submersible Pump Based On Neural Network

Electric submersible pump is a mechanical device with strong lifting capacity and plays an important role in the whole petroleum industry.Once the electric submersible pump fails,it will cause serious economic loss.Traditional submersible pump fault diagnosis methods mainly rely on manual calculation and expert experience for fault analysis,which has certain limitations.Based on the above problems and relevant literatures in the field of electric submersible pump at home and abroad,this paper systematically analyzed the parameter changes of electric submersible pump fault mode,and put forward a one-dimensional convolution neural network algorithm based on derived parameters according to the characteristics of electric submersible pump parameters.Due to the large number of abnormal and missing data in submersible pump data and the large gap between features of each fault of submersible pump,it is difficult to characterize and detect faults with uniform features.Principal Component Analysis(PCA)is used to study and analyze the real-time data of pump production.A diagnostic model is established by using Hotelling Square Statistics and Square Prediction Error.This model can effectively predict the time of pump failure and the main factors causing pump failure.In order to further explore the fault types of electric submersible pump,a one-dimensional convolution neural network model based on derived parameters is constructed in this paper.The model uses the structure of large convolution core and multi-layer small convolution core to classify the submerged pump under the condition of high receptive field,which effectively shortens the calculation time of the algorithm.Fault classification of submerged pump has achieved good results under high number of iterations.There are a lot of noises and large load changes in the working environment of electric submersible pump.In this paper,adaptive batch normalization algorithm is introduced to improve the convolution neural network model layer by layer,and t-SNE data visualization technology is used to analyze the training situation of each layer in the algorithm.The model is used to analyze the field data of submersible pump under various operating conditions.The results show that the improved algorithm greatly improves the accuracy of convolution neural network diagnosis in noisy environment and variable load conditions.More in line with the actual needs of industrial sites.