Lifespan Prediction and Multiobjective Optimization of Electrical Submersible Pump Wells Using Machine Learning

Summary To enhance the lifespan of electrical submersible pumps (ESPs) and identify key influencing factors, we develop a survival-informed deep learning framework that combines Cox proportional hazards (CPH) analysis with a long short-term memory (LSTM) network for remaining life prediction. A multiobjective optimization model based on the version 2 nondominated sorting genetic algorithm (NSGA-II) is further introduced to balance lifespan and production. A distinctive feature of the proposed framework is its integration of survival analysis, deep learning, and multiobjective optimization into a single methodology, enabling both accurate lifespan prediction and practical operational optimization. Field applications show that it outperforms contemporary deep learning baselines, achieving a magnitude of relative error (MRE) of 0.23, mean absolute error (MAE) of 0.05, and root mean squared error (RMSE) of 0.065 on the test set. In a field validation involving 30 ESP wells over a 1-year evaluation period, the proposed multiobjective optimization achieved up to 20.3% higher cumulative production compared with single-objective optimization, while maintaining a comparable average lifespan. These results demonstrate both predictive accuracy and significant operational value.