عنوان مقاله [English]
The analysis of seismic facies is a technique for mapping geological features and properties using seismic data. To analyze seismic facies, seismic attributes are categorized and classified using machine learning algorithms to identify different seismic facies. Seismic facies analysis due to the nature of seismic data, which always has a degree of uncertainty, can produce different results with even small changes in input parameters of the analyzing method. For this reason, it is necessary to select the different stages of analysis, including the selection of input parameters and algorithm of machine learning, with high accuracy with regard to the objective of the seismic facies analysis. In this study, an interactive method with the supervision of interpreter is proposed for producing seismic facies map, using the optimal selection of the input parameters and the the proper selection of clustering and classification algorithms. In this method, the interpreter in a recursive and rotational process can compare the results of the analysis and generate thr optimal results by changing the input parameters. The method presented in this article is implemented in SeisART software. SeisART has a complete environment for data initialization (importing seismic data and well data). A user-friendly interactive environment allows the user to implement several methods and monitor the corresponding result in 2D and 3D.
SeisART software makes the possibility of the interpreter contribution in the whole stages of seismic facies analysis procedure. The interpreter can select the input attributes and chose the proper methods of pattern recognition to reach the best possible result. In the software, various evaluation utilities have been provided in each stage of seismic facies analysis. These utilities allow the interpreter to monitor the results of each method quantitatively and qualitatively. In the unsupervised system, clustering quality factors are used. The interpreter calculates the validation indices for different methods of clustering and identifies the proper method which has been more successful in discovering the natural grouping of patterns in the data set. Afterward, if there is structural geology information about the horizon of interest, the interpreter can decide on the clustering result with more accuracy. In the supervised system, the most proper method is feasible using minimization of training data and validation data errors. In this case, the interpreter can use geological knowledge and well data information to verify obtained results. In this method, the interpreter can obtain different results by changing the input parameters. Comparing these results, and taking into account the path leading to this result, the interpreter gains more knowledge of existing facies.
This method has been applied to the MSF4 horizons of the 3D seismic data of the North Sea F3 and has been shown which method is more efficient for different purposes.