A New Approach to Predict the Location of Petroleum Reservoirs Using FFNN

In the petroleum industry, data management reis is crucial to the success of petroleum projects. In particular, data acquisition, storage, and classification are major concerns of oil and gas companies. This study therefore focuses on the problem of predicting a petroleum point (a possible oil reservoir) using the data generated by well loggers from the propagation of a set of sensors at different depths within a well. The training of a feedforward neural network (FFNN) model by the LevenbergMarquardt (LM) algorithm involves a random allotment of weight/bias values over multiple epochs to reduce the variance between testing and training data. Random weight assignment degrades the performance of the model, as the variance between testing and training data will remain uncertain. In this paper, a novel method, called modified feedforward neural network (MFFNN) is proposed by freezing the weight/bias coefficients to predict petrol reservoirs with minimal error. The MFFNN outperforms existing conventional models and machine learning algorithms.