Research Of Sequential Stochastic Pattern Recognition Methods And Applications In Facies Prediction

With the reduction of current petroleum resources and the increased difficultiesof oil and gas exploration, while adding the existing considerable ambiguity anduncertainty of underground oil and gas reservoirs, researchers need to understand thereservoir more meticulous in order to avoid the huge loss of funds caused by theabove-mentioned problem. So, it is becoming more and more crucial to do thedivision of sedimentary in geological work.In view of the problem of uncertainties in the process of predicting thesedimentary facies using traditional pattern recognition, the essay presents a newmethod-sequential stochastic pattern recognition, which is based on Su10area inSulige gas field. The method is based on pattern recognition and systematicallyintegrates the principles of sequential and the thoughts of random, which realizes theprediction of sedimentary facies quantitatively, and thus determines the distribution ofsedimentary.Compared to the traditional pattern recognition, this method mainly solves theproblems of prediction uncertainty, variability of spatial data and prediction usingpurely well data.Firstly, if only just considering the well data in the process of predictingsedimentary facies, the results obtained are not precise enough. So the study adds thehigh-resolution seismic attribute data athwartships. By analyzing the relationshipbetween the sedimentary facies and seismic attribute, it is essential to add the seismicdata for improving the prediction. Seismic attribute analysis shows that seismicattributes which response is more sensitive include bandwidth, reflection strength,RMS amplitude, instantaneous frequency, the average arc length and so on.Secondly, the traditional pattern recognition methods are mainly based onclassical statistics, which ignore the features of spatial structure. However, both welldata and seismic data we used have a spatial structure, which can be seen as theregionalized variable. In the research of algorithm, the well data searched use wellpoint variogram and seismic data searched use seismic variogram. Finally, the results of traditional methods have a feature of a certain overlap,which demonstrates the uncertainty of prediction. In order to reduce the uncertainty,the study introduces the thoughts of sequential random. When traversing the grid withthe different paths, it will get a number of different simulations and the differences ofthe simulations reflect the implied uncertainty in probabilistic model. At the sametime, both well data and simulated data will be considered, which will undoubtedlyimprove the prediction accuracy of the results.The article uses the C plus plus language as a tool to implement the algorithmwhich overcomes the disadvantages of traditional pattern recognition. The method hasimportant reference value for the exploration, development and research of petroleum.