Sand Production Prediction Based on Deep Learning and BP Neural Network

Abstract

ABSTRACT Deep learning can better simulate highly nonlinear engineering problems through training adjustment, and has good practicability for predicting invisible data. BP Neural Network (BPNN) is a widely used deep learning algorithm, and strengthens the accuracy of developing training models through feedback adjustment to increase the accuracy of target prediction. This work considers the similar lithology characteristics in the same block, the sand production profile of a single production well is first predicted by using empirical formulas and logging data, and further the logging data and sand production profile are trained by back-propagating to continuously adjust the weights and values of BPNN. Finally, one develops a sand production prediction model based on deep learning and the BP neural network algorithm, and uses the model to predict the potential of sand production of other production wells in the same block. The prediction results (training set) of the BP neural network algorithm are in good agreement with the traditional empirical formula (testing set), and the goodness of fit reaches 0.9 or more. Therefore, it increases the confidence in the proposed sand production prediction model based on deep learning and the BPNN algorithm to improve the early warning of the risks of sand production. INTRODUCTION Both the poorly consolidated sandstone reservoirs with lower strength and high-pressure gas wells with a higher production speed are potential candidates for severe sand production. Therefore, accurately predicting sand production plays a role in the choice of sand control methods. The core samples are often restricted to enough obtained, and thus it fails to get a sufficient amount of test data related to the rock mechanics and strength parameters in a laboratory, such that a poor prediction of the sand production is made. However, a good choice that the logging data combines with the classical statistical empirical formulas, such as combined elastic modulus, sand production index (Chen et al.,2008), and Schlumberger sand production index (Tixier et al.,1975), can quickly provide the sand production profile of a single well, and thus sand production profile is convenient for engineers to capture the potential formation section of sand production in oil and gas wells, and helps engineers to timely adjust production allocation and optimize sand control methods.