P-Wave Sonic Log Predictive Modeling with Optimal Artificial Neural Networks Topology

Abstract

Given the financial challenges facing the oil and gas industry, the value of the information is considered relatively high; therefore, data science has been an alternative compensating tool. This study aimed to find an optimal neural network topology that provides an ideal data solution by studying neural network topology. Therefore, we trained different neural networks topologies in terms of the number of hidden neurons and layers. Volve oil field data is used in this study to predict the compressional sonic wave travel time. Optimal Neural Network topology found using five hidden layers and five hidden neurons while using a single layer with different numbers of hidden neurons was ineffective. The highest training and testing accuracy with a single hidden layer found 0.94 and 0.914, respectively. In contrast, it was found 0.947 and 0.934 with 50 hidden neurons and five hidden layers. Yet, increasing the number of hidden layers and hidden neurons is found to cause overfitting; therefore, only an optimal topology is a critical factor.