Intelligent Porosity Prediction of Old Oilfield Based on Residual Multilayer Perceptron

Abstract The objective of this paper is to summarize the porosity interpretation and prediction methods of old oilfields. Due to the existence of a large number of oilfields with a long production time in China, the logging data of this type of oilfield have problems such as poor data quality, relatively long data, and different data standards, and it is very difficult to interpret and predict the physical properties such as porosity in production and development. In order to solve this problem, this paper proposes a new technical route, in which the data obtained from core analysis are first preprocessed by homing, baseline, normalization, etc., and then the residual multilayer perceptron model is established, and finally the trained model is applied to the practice to explain and predict porosity. The correlation of the prediction results of the test set reached about 0.91, which had a good generalization ability, and then the reservoir parameter model was established by using the prediction results to provide a basis for the production practice in this area. The practical significance and suggestions of the prediction results, as well as the possible research limitations and future research directions are also discussed. In general, this study provides a technical route for solving the problem of porosity interpretation and prediction in old oilfields, and provides a reference for practice and research in this field.