Improving the classification of facies quality in tight sands by petrophysical logs

As conventional hydrocarbon reserves are running out, attention is now being paid to unconventional hydrocarbon resources and reserves such as tight sands and hydrocarbon shales for future energy supplies. To achieve this, the identification of tight sand facies is based on zones containing mature hydrocarbons in priority. Organic geochemical methods are the commonest methods to evaluate the quality of these reservoirs. In this study, using a deep learning approach and using petrophysical logs, a suitable classification model for facies quality is presented. Moreover, the proposed method has been compared with two common methods: multilinear regression and multilayer perceptron neural network. The results indicated that the accuracy of facies classification using these three methods is about 63%, 71%, and 84% for linear multilinear regression, perceptron multilayer neural network, and deep learning, respectively. Finally, the accuracy of the deep learning networks was optimized using two gravitational search and whale optimization algorithms. It has been shown that the accuracy of deep learning was increased from 84% to 87% and 90.5% using the gravitational search algorithms and whale algorithms, respectively.