Production history is essential for any reservoir engineering study. It used for history matching in
reservoir simulation study, rate transient analysis and decline curve analysis (DCA). The quality
of the production data is important. Better quality of the production data reduces the
uncertainties during modeling the reservoir, characterizing it and forecasting the future
production.
Shale gas reservoirs have been developed heavily in last two decades. They have huge
reserves but there are challenges in evaluating them economically. Transient flow that could last
for long time, liquid loading causing successful shut ins and controlling the bottom hole flowing
pressure cause the production data to fluctuate heavily. The noisy production profile makes it
difficult to detect the different flow regimes precisely and affects analysis such DCA.
In this paper, we used a machine learning algorithm called angular- based outlier detector
(ABOD) to improve the production data of 4 shale gas wells. It was assumed that 20% of the
production data is noise and the algorithm is asked to determine the points with the highest
potential to be detected as noise. After that, the different flow regimes were determined before
and after improving the data quality.
The results show that the ABOD algorithm removed the noise from the production data
efficiently. The production profile was smoothed without any bias and without removing any
significant event. Detecting the different flow regimes was much clear after removing the noise.
Moreover, we determined the masked flow regimes after improving the production data quality
in some cases.
