LAUSR

Abstract Most of the domestic oil fields are in the middle or late stage of exploration and development, and there are fewer and fewer oil fields that can be easily discovered. Proved reserves are dominated by concealed reservoirs, but concealed reservoirs are difficult to find. The prediction for favorable reservoir is a key in the process of exploration and development, but the traditional prediction methods of favorable reservoir rely on analysis of geological prospectors according to commonly used seismic attributes, which leads to low exploration efficiency. To improve effectiveness of the prediction for favorable reservoir, this paper introduces a multi-classification Twin Support Vector Machine (MTWSVM) method suitable for the case of insufficient reservoir label samples caused by fewer drilling data, which is used to identify favorable reservoirs. Since each sub-classifier of existing MTWSVM uses the same penalty parameters and kernel parameters, ignoring the differences between different classes, it cannot play the best role of each sub-classifier. We propose a multi-classification Twin Support Vector Machine based on hybrid parameters (HP-MTWSVM). This algorithm selects appropriate parameters for different sub-classifiers, maintaining the diversity of classifiers. Twin Support Vector Machine (TWSVM) is facing the problem that its parameters are difficult to be appointed. Additionally, HP-MTWSVM algorithm introduces a large number of parameters. This paper further proposes a hybrid parameter multi-classification Twin Support Vector Machine based on Grey Wolf Optimizer (GWO-HP-MTWSVM). This method uses GWO to optimize the parameters of HP-MTWSVM. Experiments show that the prediction accuracy of GWO-HP-MTWSVM is more than 64%, better than that of manual prediction in complex concealed reservoirs. The model can help geological exploration personnel quickly delineate favorable areas, provide basis for accurately drilling wells and avoid waste of resources caused by empty reservoirs.