RIASSUNTO
Abstract
Along the enhancement of data processing and storing capabilities, introduction of cloud computing and a broader connectivity between different systems, data mining techniques and machine learning consolidate themselves between the main exponents for business improvement. Even areas of industry considerably mature, as the oil & gas, shall handle these tools to modernize its processes and enhance their efficiency. The applicable fields are diverse, from the operational realm to management areas. It is remarkable to consider the benefits that the adoption of richer prediction models would provide, in substitution of tasks so far performed only from empiricism, or under minimal premises. Towards planning, data mining associated with machine learning turns into an important tool for some services demand prediction. Especially those at which the occurrence is essentially probabilistic. Such analysis may be implemented crossing multiple input data, allowing the model to be a fair representation of reality. Fishing occurrence is an unmistakable example of well service with probabilistic incidence. Even if, at a first glance, their manifestation seems chaotic, fishing incidence varies according the activities performed or well specifications. This means the event probability depends not only if the rig is drilling or completing, but also on well specification. In a large oil company, with a large amount of wells, the possibility of a multi-variable prediction for this kind of occurrence is very valuable for a proper mapping and dimensioning of the service amount.
This present paper shows the steps of quantifying the demand for fishing services using previous experience. These steps are explained, from input data classification and pre-processing through the choice of the fittest machine learning model, and finally, the process and analysis of the obtained results.
Once the model is defined and implemented, each new analysis can be performed quickly. This represents a massive time saving, especially when schedule changes happen very often. However, the advantages obtained are not only restricted to the boost in performance, but also the possibility to consider a larger assortment of input variables, and therefore allow the user to obtain a model closer to reality, and still capable of be continuously improved and adapted to new scenarios.
Regardless being the purpose of this work the amount of services to hire, the obtained data are also a great source for fishing prevention, aiming to reduce nonproductive time (NPT). It can provide an intensity map, indicating the activities at which shall the efforts be prioritized. They are still useful in rig schedule forecasting, to permit predicting the amount of time for each activity regarding fishing events. Finally, regardless of referring to fishing activity, the methods and process used in this work may be, in general, used for other purposes, within or outside the oil industry.