Well data preprocessing using machine learning algorithms

Abstract

The task of interpolating missing data is important for integrated asset modeling, since the accuracy and reliability of modeling directly depend on the quality of the input data. If the model is missing some data, then this may lead to the impossibility of modeling for this time step. Thus, the problem of interpolation of missing data is important and its solution improves the accuracy and reliability of forecasts.

Aim. The aim of the work is to improve the quality of data for integrated asset modeling (IMA).

Materials and methods. The data from the technological regime of wells, necessary for well modeling within the framework of IMA, are considered as initial data. The table view data contains gaps (missed values), which reduces the amount of data available for modeling.

Results. The paper shows a method for filling gaps with machine learning algorithms on the example of a real field. Missing GOR readings for many dates were predicted by machine learning models.

Conclusions. As a result of the work, an algorithm for interpolating data was proposed to identify one case of infection. This step can increase the amount of data without gaps and significantly improve the quality of the models. In this problem, we were able to increase the amount of data for training models by 98 % in this way, which meets the requirements for the maximum work model by an average of 41 %.