| Popis: |
Abstract A classical objective of reservoir characterization using seismic attributes is to provide a lithology-related attributes that may be used as an aid for interpretation and ultimately as an input for the geological model building. For such an objective, supervised classification techniques take advantage of well data in order to constrain the learning phase of the classification methodology. The output can be pseudo-petrophysical cubes or facies probability cubes. Total has developed an internal tool which enables conversion of inverted seismic attributes (such as acoustic impedance and Poisson's ratio) into ‘lithocubes’ describing the probability to find some selected lithologies. These lithocubes are seismic attributes and hence have the same resolution as the seismic. In order to achieve proper on supervised classification it is essential to have the facies description compatible in terms of scale with the seismic attributes cubes. This up-scaling can be performed qualitatively without implicit use of the petro-elastic quantitative information such as Vp, Vs and density. Moreover, there could be significant variability in the up-scaling results according to the choices made by the operator performing a manual up-scaling (facies grouping and minimum thickness of each layer corresponding to a given facies for example). This paper describes a semi-automatic workflow capable of limiting the subjective influence of the operator on the final up-scaled well facies. By using quantitative criteria based on the petroelastic behavior of each facies it is possible to help the operator to achieve objective up-scaling choices. Ultimately, this workflow allows to improve the reliability of classified lithocubes. Introduction Within the framework of supervised classification, it is essential to have well data and seismic attributes cubes (often coming from the inversion) compatible in terms of vertical resolution. When considering petrophysical log corresponding to a continuous property it is relatively easy to adapt the scale at which it is defined. It is usually achieved by time converting the log property and filtering it in the seismic bandwidth. However, in the case of up-scaling of the facies log which is a discrete property, it cannot be performed by filtering. Until now, this up-scaling was done manually, where often Vclay and porosity properties were taken into account, but rather rarely the petro-elastic data such as Vp, Vs and density. Indeed we prefer to propagate the well information by using cubes of seismic attributes, such as the acoustic impedance and Poisson's Ratio. Moreover, the variability of the up-scaling results could be important according to the operator in charge. The choices made during this process are sometimes difficult to justify rationally, and even more difficult to quantify. This paper describes a fast semi-automatic workflow capable of limiting the influence of the person on the final up-scaled results, and being able to quantify the up-scaling process. |
