Hybrid Waveform Classification Applied to Delineate Compartments in a Complex Reservoir in the Malay Basin

Abstract

The delineation of heterogeneity and compartments within a reservoir is essential for field development and reserve estimation. In a complex reservoir in the Malay Basin, conventional seismic attributes are ineffective at characterizing the reservoir facies. We have developed a hybrid seismic waveform classification that combines standard unsupervised classification with a highly-interactive supervised classification. This combines the simplicity of unsupervised classification with the flexibility of supervised classification. The new tool successfully delineates previously unknown reservoir compartments, allowing us to plan for the field development and resource management. Introduction Characterizing reservoir heterogeneity is essential for optimum resource estimation and field development. Reservoir heterogeneity should be understood as early as possible in field development to design an appropriate drainage strategy that increases the efficiency of producing the reserves. As more complex reservoirs are targeted, it becomes more difficult to predict facies types and extents. The field involved in this study, which is in the Malay Basin, is especially complex due to compartmentalization, making it difficult to ascertain the reservoir boundaries and heterogeneity. The main reservoirs in the field are of excellent quality with porosity up to 35%. They are characterized by 15 to 20 m of sand fining upward that was deposited in prograding mouth bars in a shallow deltaic plain. In contrast, the reservoir of interest in this study is very thin, from 2 to 10 m thick, with a porosity of 15%. Conventional seismic attributes are unable to distinguish compartments in this reservoir, making it difficult to monitor the water injection and understand the cross-well injector-producer pressure response. Seismic data contains significant stratigraphic information that helps identify regions with similar reservoir facies. This information is contained within waveforms, which are small segments of seismic traces that could represent single reflections or patterns of interfering reflections. A waveform is a function of amplitude, phase, and frequency. Although individual waveforms lack inherent geologic meaning, modeling and clustering can help relate them to known geology (Poupon et al., 1999; Coleou, et al., 2003). However, realistic models require so many variables that modeled waveforms are highly non-unique and have little diagnostic value. As a result, waveform maps are always interpreted qualitatively. An effective and popular method for automatic pattern recognition that uses neural networks to analyze regions of similar waveform is the Kohonen Self-Organizing Feature Map, or KSOFM (Addy, 1997; Barnes and Laughlin, 2002). As with any method of unsupervised classification, KSOFM is driven entirely by the seismic data, and the number of representative waveforms (classes) must be pre-determined before starting the neural networks training. The resulting waveform facies map is therefore limited to a pre-defined number of classes, which usually do not correspond to the facies of greatest interest. In order to address this shortcoming, we develop a hybrid seismic waveform classification that combines the advantages of unsupervised and supervised classification, and which is better at delineating reservoir heterogeneity.

Spectral Decomposition and a Waveform Cluster to Characterize Strongly Heterogeneous Paleokarst Reservoirs in the Tarim Basin, China

“…The Waveform cluster is a pattern recognition technique often used to analyze seismic waveforms [76,77]. Within a certain interval, the variation in seismic facies gives rise to changes in the amplitude, phase, and frequency of the seismic reflection.…”

Application of Geologically Constrained Machine Learning Method in Characterizing Paleokarst Reservoirs of Tarim Basin, China

“…To build the final machine learning model, automatic waveform clustering, such as fuzzy C-means cluster [26], can be used to take the key seismic time-frequency signal as input and extract the natural clustering of the underlying seismic facies. Waveform cluster is a pattern recognition technique, which is often used to analyze seismic waveform [27]. Within a certain interval, the seismic facies variation gives rise to changes in the amplitude, phase, and frequency of the seismic reflection.…”