Rapid Characterisation of Fractures and Reservoir Properties using Automatic History Matching: An Investigation of Different Production Performance in Hydraulically Fractured Wells in Sirikit Oil Field

Recently, many hydraulic fracturing has been executed in Sirikit oil field (S1), an onshore oil field in Thailand, to unlock the production from tight sands. However, production performances of each stimulated well were varied despite a similar fracturing technique. The variation may be due to different fracture geometry, fracture properties, and reservoir properties. Although these parameters are critical in optimizing fracturing design, they are unfortunately difficult to be quantified by analytical method, especially the diagnostic of hydraulic fracture after having actual production data. To answer this question, we leveraged the automatic history match (AHM) scheme based on Neural Network-Markov Chain Monte Carlo (NN-MCMC). We utilized the production data to characterize fractures and reservoir properties and stochastically quantify their uncertainty.The framework is based on a practical and efficient iterative workflow that integrates four main stages: (1) Embedded Discrete Fracture Model (EDFM) preprocessing for the best fracture characterization over Local Grid Refinement (LGR), (2) multiphase fluid reservoir simulation, (3) neural network application for generating proxy models, and (4) proxy-based Markov Chain Monte Carlo (MCMC) algorithm for screening the best stochastic solutions. Three wells from the same wellsite and hydraulic fracturing campaign were selected for a study. Uncertain parameters including hydraulic fractures geometry and properties, reservoir permeability, water saturation and relative permeability curves were included for automatic history matching. Rapid uncertainty quantification was completed by screening through 1 million realizations and proposed only 325 realizations to be validated with reservoir simulation. The automatic history matching was executed and required running time less than a day for each well. The posterior distributions of uncertain parameters emphasizing most likely values and their uncertainty were obtained. The difference in fractures and reservoir properties were obtained. Also, the production forecast for each well can be performed probabilistically based on multiple history matching solutions. The automatic history matching workflow could extract the valuable information of fractures and reservoir geometry from production data, which does not require any additional cost. This characterization of fracture geometry and properties, integrating with other methods, can help optimizing fracturing and improving completion design in hydraulically fractured wells in Sirikit oil field in the future.