Background. Facial zoning based on seismic data is an important task of dynamic analysis. There are numerous approaches of solving this problem using various algorithms. The most common method is clustering by reflection shape. This approach belongs to unsupervised learning algorithms, due to the mapping of seismic facies is based on the internal data structure and the key feature is the change in the wave packet within the target interval. The disadvantage of this method is requirement of further tying clustering results and geological information. Another way of directed solution of this problem is the use of supervised learning algorithms. This category includes various classification methods that relate to the category of machine learning. In comparison to traditional approaches of seismic facial analysis, this method accounts geological information at the computation stage. Aim. This paper considers the results of a research carried out with the study of the facies structure of the Tyumen formation at a group of fields in the Khanty-Mansiysk Autonomous Region. The Tyumen formation is characterized by the predominance of channel facies associated with the development of complex river systems, which are clearly observed in the dynamic characteristics of the wave field. A complicating factor in the study of these deposits is the rather low coverage of well data, which makes difficult the geological interpretation of the results obtained. Materials and methods. The authors used the Random Forest classification method to deal with the assigned task. The application of the method is considered on the cluster consisting of three seismic surveys obtained at different times. For training, expert marking by area was used based on the distribution of amplitudes along the reflecting horizon. Results. As a result of the research, a probabilistic assessment of the distribution of channel facies was obtained, that is related to the perspective of this type of deposits in the study area. Thus, the authors have developed a methodology that gives an opportunity to obtain an estimate of the probability of the presence of a certain facies using seismic data. Conclusions. The performed study shows the possibility of using the Random Forest classification method to solve the problem of facial zoning.
Facial Recognition Using Aggregation and Random Forest Classification Method
