Principal component analysis-assisted selection of optimal denoising method for oil well transient data

Abstract

Oil and gas production wells are often equipped with modern, permanent or temporary in-well monitoring systems, either electronic or fiber-optic, typically for measurement of downhole pressure and temperature. Consequently, novel methods of pressure and temperature transient analysis (PTTA) have emerged in the past two decades, able to interpret subtle thermodynamic effects. Such analysis demands high-quality data. High-level reduction in data noise is often needed in order to ensure sufficient reliability of the PTTA. This paper considers the case of a state-of-the-art intelligent well equipped with fiber-optic, high-precision, permanent downhole gauges. This is followed by screening, development, verification and application of data denoising methods that can overcome the limitation of the existing noise reduction methods. Firstly, the specific types of noise contained in the original data are analyzed by wavelet transform, and the corresponding denoising methods are selected on the basis of the wavelet analysis. Then, the wavelet threshold denoising method is used for the data with white noise and white Gaussian noise, while a data smoothing method is used for the data with impulse noise. The paper further proposes a comprehensive evaluation index as a useful denoising success metrics for optimal selection of the optimal combination of the noise reduction methods. This metrics comprises a weighted combination of the signal-to-noise ratio and smoothness value where the principal component analysis was used to determine the weights. Thus the workflow proposed here can be comprehensively defined solely by the data via its processing and analysis. Finally, the effectiveness of the optimal selection methods is confirmed by the robustness of the PTTA results derived from the de-noised measurements from the above-mentioned oil wells.