融合电流-电磁应力特征的XGBoost-SHAP潜油电泵故障诊断OA

Aiming at the problems of complex working conditions,fault types and poor interpretability of traditional diagnosis of submersible electric pumps in offshore oilfields,this paper takes an offshore oil-field submersible electric pump as the object,collects stator current and electromagnetic stress signals un-der normal working conditions and five fault conditions,and uses sliding window and standardization pro-cessing to construct 47-dimensional fusion features including time domain,frequency domain and electro-magnetic interaction.Based on this feature set,classification models such as XGBoost are established and their performance is compared.The results show that the tree ensemble model,and the accuracy of XG-Boost and random forest test is about 97.9％and 95.7％,respectively,which is significantly better than that of BP neural network.he global and local interpretation of XGBoost is further combined with SHAP to identify key features such as current spectrum standard deviation,waveform factor and electromagnetic stress spectrum energy,reveal the difference of electro-magnetic response under different working condi-tions.The research shows that the fusion feature diagnosis method based on XGBoost-SHAP has both high accuracy and interpretability,provid theoretical and technical support for the intelligent operation and maintenance of submersible electric pump.