An Optimized Hybrid Intelligent System for High-Accuracy Dewpoint Pressure Estimation

Abstract

Dewpoint pressure (DPP) is a critical property of gas condensate reservoir development. Accurately estimating this property remains a significant challenge. Existing empirical correlations and iterative methods lack sufficient accuracy due to complexity and computational intensity. However, despite their utilization involving complex computations, they have not achieved sufficient accuracy. Several individual intelligent systems have been utilized to predict this property with good accuracy, but the application of hybrid models is less common. Therefore, this study proposes two hybrid intelligent models—Particle Swarm Optimization combined with Neural Networks (PSONN) and Neuro-Fuzzy (NFuzzy)—to enhance prediction accuracy of dewpoint pressure. Approximately 860 collected data points were used to develop these hybrid models. Inputs such as temperature (T), hydrocarbon composition, specific gravity, and molecular weight of heptane plus were utilized to predict the dewpoint pressure. In this study, the performance of both intelligent hybrid systems is compared to the most widely published Artificial Intelligence (AI) models. Based on statistical error analysis results, the new hybrid models outperform the published models. The results confirm that the PSONN hybrid model achieved the best performance with an absolute percent relative error (APRE) of 2.47%.