Inferring the Effect of Chemical Injections for Asphaltene Producing Deepwater Wells Using Machine Learning

Abstract

Abstract Asphaltene production in Deepwater wells is an important operational issue which may result in planned and unplanned shut-ins. Excessive asphaltene precipitation and deposition can cause production curtailment that adds up to significant costs yearly tooperators in Deepwater production and transportation of Deepwater asphalt base crude oil. Costly chemical injections such as xylene soaks are used to dissolve the asphaltenes in the tubing. This study proposes a new machine learning technique to increase the effectiveness of such soaks in a Deepwater well. A predictive solution is developed where a scalable Machine Learning (ML) model predicts un-commanded shut-ins by analyzing historical and real-timefeed of sensor and simulation data. Deployed workflow can inform Control Room Operators hours or days before a potential un-commanded shut-inoccurs. A common unsupervised learning framework for predictive maintenance called anomaly detection algorithm is built. Multiple anomaly detection models are investigated within the scope of dimensionality reduction. Principle Component Analysis (PCA) and Artificial Neural Net(ANN) base LSTM Autoencoders are deployed to tackle the problem through reconstruction of the original input. Anomaly score and threshold as ML outputs are streamlined in near real-time back to the database to serve the operators. Following this, further analytics is conducted toassess the impact of chemical soaks on anomalies. In this work, ML model output is benchmarked against a Gulf of Mexico Deepwater well where asphaltene precipitation and deposition is known to occur. The ML architecture can ingest real-time data in batch, maintained by OSI PI historian. The architecture is proved to detect anomalies hours or days before a shut-in event happens, so that operators can take early actions before severe damage to wellhead or downhole equipment occurs instead of reacting to a possible asphaltene event offshore. This study shows that, if used properly, data science can be an effective and reliable tool for Petroleum Engineers and Offshore Operators to not only detect anomalous events but also assess the impact of well interventions during drilling, completions and operations in Oil and Gas Industry.