Signal-noise analysis of PTA data from modern well surveillance system

Abstract

In industries such as oil and gas, geothermal and carbon storage most wells are equipped with permanent downhole gauges(PDGs). These gauges are used to monitor the well pressure in real time. The pressure read- ings are used to interpret and calculate information about the wells and underground flows governing injection, production and overall process effi- ciencies.

The process of interpreting the data is labour intensive and requires exten- sive experience in the field. This thesis focuses on automating parts of the process to speed up the interpretation, and minimise the labour-intensive work.

The automation process includes removing outliers, filtering and prepar- ing the data for a pressure transient analyis (PTA). Real-world raw data presents some problems, such as the well operations and rate fluctuations, which can complicate the interpretation.

The paper examines different filtering methods, such as mean filter, low pass filter, various types of regression, Fourier transformations and wavelets. The paper also looks at combining different filters to enhance the best parts from both filters.

Our thesis showed the importance of resampling a signal before processing it. The best results came from combining a moving average filter with LOWESS. This combination eliminated almost all noise, while still giving a fairly good representation of the transients.

The paper tested with a synthetic dataset to calculate a score for how good each filter performed. This dataset is not a perfect representation of real world conditions, but the result still concluded that a LOWESS filter is the best choice.

In industries such as oil and gas, geothermal and carbon storage most wells are equipped with permanent downhole gauges(PDGs). These gauges are used to monitor the well pressure in real time. The pressure read- ings are used to interpret and calculate information about the wells and underground flows governing injection, production and overall process effi- ciencies.

The process of interpreting the data is labour intensive and requires exten- sive experience in the field. This thesis focuses on automating parts of the process to speed up the interpretation, and minimise the labour-intensive work.

The automation process includes removing outliers, filtering and prepar- ing the data for a pressure transient analyis (PTA). Real-world raw data presents some problems, such as the well operations and rate fluctuations, which can complicate the interpretation.

The paper examines different filtering methods, such as mean filter, low pass filter, various types of regression, Fourier transformations and wavelets. The paper also looks at combining different filters to enhance the best parts from both filters.

Our thesis showed the importance of resampling a signal before processing it. The best results came from combining a moving average filter with LOWESS. This combination eliminated almost all noise, while still giving a fairly good representation of the transients.

The paper tested with a synthetic dataset to calculate a score for how good each filter performed. This dataset is not a perfect representation of real world conditions, but the result still concluded that a LOWESS filter is the best choice.