Intelligent ESPs Diagnostic Model Based on Big Data and Machine Learning

Abstract There are various types of available data, with different data structures, dimensions, and intervals. Conventional big data algorithms can lead to poor interpretability. In this study, novel methods are proposed based on big data analysis and machine learning. First, the data are categorized into 4 pairs to build parameter analysis model: pump static data, pump dynamic data, production static data, and production dynamic data. Combining static and dynamic data to establish parameter analysis modeling and different data pairs adopt different modeling methods: current and voltage data adopt picture recognition algorithm, vibration data adopt time domain analysis, and production data adopt random forest algorithm. Then, a deep neural network is developed to couple four models to diagnose ESPs production status. All the models have high accuracy through optimization. In the end, a management platform had been established for ESP wells including data visualization, production analysis, fault diagnose and warning. Visual production dynamic tracking management can improve the management efficiency of ESP wells and reduce the cost of future workover and relocation.