توصيفگر ها :
داده لرزهاي , چاه پيمايي , ليتولوژي , وارون سازي , شات-كنترل , شبكه عصبي , نشانگرهاي لرزهاي
چكيده انگليسي :
Today, with the advancements in science and the complexities of explorations, the integration of seismic data and information obtained from well logging is utilized to acquire the most accurate information from the studied wells while minimizing possible errors. One of the primary steps in the static and dynamic modeling of reservoirs involves the correct calculation of petrophysical parameters, which is of significant importance. Among the essential parameters needed for the accurate estimation of hydrocarbon reservoirs are the rock type and the lithology of the formation. Any errors in the estimation of rock composition could lead to the loss of the reservoir or increase costs incurred during this process. This study focuses on one of the reservoirs in southwestern Iran, namely the Asmari reservoir, utilizing three-dimensional seismic data following processing and the corresponding logs from three wells in this field. Initially, using shot-controlled data, the depth basis of the well logging data was converted to a time basis corresponding to the round-trip travel time of the seismic wave to enable comparison and alignment of well logging data with seismic data. Subsequently, the dominant wavelet was calculated by both statistical methods and wavelets derived from the comparison of synthetic seismograms at the locations of the three wells with seismic data. With the dominant wavelet extracted from the three well locations, seismic data inversion was performed. Among the various seismic data inversion methods, the model-based inversion method, with a correlation coefficient of 0.99 for each of the three wells (10, 151, and 114), was identified as the optimal method, yielding errors of 0.8, 0.12, and 0.10, respectively. Concurrently with the inversion of seismic data, the contents related to anhydrite, calcite, dolomite, and quartz were estimated using mineralogical determination methods. Through multiple regression methods and various seismic indicators, such as mean frequency, inverted sections, integrals, and raw seismic sections, mineralogy was estimated. The best seismic indicators suitable for estimating mineralogy at the well locations, with a training correlation exceeding 80 percent and an estimation error of less than 5 percent, were identified. The use of the top indicators in defining the Probabilistic Neural Network (PNN) and the Multi-Layer Feedforward Network (MLFN) resulted in training errors of 89, 86, and 90 percent, and validation errors of 71, 71, and 70 percent in estimating quartz. For estimating calcite, these networks achieved training errors of 84, 83, and 85 percent, and validation errors of 76, 77, and 77 percent. In the case of dolomite, the training errors were 84 and 83 percent, with validation errors of 68 and 69 percent, while well number 151 was used as a reference well. A comparison of the aforementioned mineralogical sections with the variations of the mentioned minerals showed a qualitative agreement with the geological reports of the Asmari formation horizons. This research aims to predict the mineralogy of rocks using seismic indicators and artificial neural networks with minimal wells available.
