Lins I., Moura M., Lopez Droguett E., Zio E., Couto Jacinto C., 2013, Failure Prediction of Oil Wells by Support Vector Regression with Variable Selection, Hyperparameter Tuning and Uncertainty Analysis, Chemical Engineering Transactions, 33, 817-822.
In order to comply with safety requirements and due to the increasing demand on oil production, failure prediction of the related systems (e.g. oil wells) has become an important task. An analytical modelling of the reliability behaviour of these systems is often impractical and this justifies the use of data-driven learning methods like Support Vector Regression (SVR). The paper proposes a comprehensive failure prediction framework based on the combination of Particle Swarm Optimization (PSO) and bootstrap methods with SVR. The PSO portion of the methodology is responsible for the simultaneous selection of SVR hyperparameters’ values and the choice of the most relevant influencing variables. The adjusted SVR model feed bootstrap methods, which provide point and interval estimates of the response variable. The bootstrapped PSO + SVR is applied in the context of the Brazilian oil industry and the obtained results suggest that it is a valuable tool in the support of maintenance-related decisions.
