A comparison of methods for denoising of well test pressure data

Abstract

Abstract Pressure transient data from downhole gauges are one of the key parameters in characterizing reservoir properties and forecasting future reservoir performance. Reservoir pressure is usually measured under dynamic changes. The collected data usually contain different levels of noise, particularly due to imperfections in measuring instruments and imperfect calibrations. The latter is due to changes between the laboratory environment and reservoir conditions. To have accurate descriptions of reservoir, it is essential to smooth the pressure data. Most related studies have employed the wavelet transform to reduce noise. However, there appears to be little research addressing the use of other smoothing techniques for pressure transient data. This paper, therefore, evaluates and compares the performance of three types of smoothing and noise removal methods, namely wavelet transform as a widely used filtering technique, regression-based smoothers, and autoregressive smoothing methods to reduce artificial noise added to simulated dual-porosity pressure data. Particularly, noise is more pronounced in pressure derivative, and so denoising of pressure derivative requires more effective tools. The effectiveness of the noise removing methods was compared using mean square error. The results show that the regression-based methods lead to the same or even better reduction in the noise level as compared to the wavelet domain filter, while the employed autoregressive method results in a moderate performance. We also test the performance of various combinations of the different smoothing methods to filter the same noisy data. It is shown that the combined locally weighted scatterplot smooth (LOESS) and autoregressive moving average (ARMA) gives the best smoothing performance for pressure derivative data. Application of the combined LOESS–ARMA to real field data shows promising results.

Keywords

• Well test data
• Noise
• Wavelet transform
• Regression-based smoothers
• Autoregressive moving average