A Novel Time Series Approach to ESP Gas Lock Predictive Classification Based on Dynamic Time Warping & K-Nearest Neighbor

Abstract

Abstract Electrical Submersible Pump (ESP) systems are a widely used Artificial lift method in the Oil and Gas industry- however, they do present failures with consistent use, as does any other tool. It is important to reliably predict the onset of damaging operating conditions and take proactive action to prevent early failure of ESPs. Data processed from ESPs are considered time series due to their time-sensitive occurrences. Deep learning methodology needs to be adapted to the time series conditions of this data. Current deep learning methods deployed in the industry have demonstrated superior performance under the general requirement of the availability of a large set of training instances. However, many lack the efficient adaptability of the time component in the prediction of tool failure. Existing solutions do not further consider detecting signal similarity and reliable time alignment between pattern-matching measures. With this, we propose the use of K-Nearest Neighbor (KNN) with Dynamic Time Warping (DTW) as a distance-based classification method to classify the deficiency of a tool accurately and efficiently while considering pattern matching in specified time intervals. The amalgamation of DTW and KNN is considered an effective method for time series classification of ESP gas lock prediction. Although the time complexity of this approach is considered as lower efficient than that of methods such as Time Series Forest Classifiers and Support Vector Machines, our approach optimises the search space and speeds up the execution at a more effective rate, thus providing a novel way for ESP predictive maintenance.