Abstract: Electric Submersible Pump (ESP) is currently widely employed to help enhance production for non-flowing well with high production and high water cut well. However, ESP failures are common in the oil industry, and these failures lead to production disruptions, resulting in significant economic loses. The purpose of this paper is to evaluate Principal Component Analysis (PCA) as an unsupervised machine learning technique to detect the cause of ESP failures and estimate the remaining life of ESP. The data provided by the ESP system are usually closely correlated; principal component analysis utilizes these data to extract eigenvalues and create new space, then reevaluate ESP system with multiple new principal components. Hortlin square statistical algorithm and the square error analysis algorithm are used to establish a PCA diagnostic model. This model was successfully applied in the Bohai oilfield to diagnose the ESP performance real-time. The most responsible decision variable for the potential ESP failures are determined according to the order of contribution. Also, the PCA diagnostic model was able to determine the time at which the ESP failures occurs. This paper demonstrates that the application of PCA method preforms well in monitoring ESP operations and predicts the impending ESP failures.

Key words: electric submersible pump, principal component analysis, pattern recognition, feature extraction, fault diagnosis