Shor, RomanAguilera, RobertoMatoorian, Raya2024-09-182024-09-182024-09-13Matoorian, R. (2024). Intelligent production optimization in real-time by implementing hybrid data-physics simulation (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.https://hdl.handle.net/1880/119769This research introduces a novel approach to overcoming key challenges in applying machine learning (ML) for production forecasting and performance evaluation in conventional and unconventional reservoirs. By leveraging a hybrid data-physics architecture (HDP), this approach addresses limitations such as poor generalizability, the need for extensive training datasets, and discrepancies between model outputs and physical principles. The HDP integrates physical equations such as decline curves into a deep neural network, enabling the two to function together during forward and backward propagation. The training data includes various factors influencing production rates, encompassing information that may not readily conform to conventional physical equations. This hybridization enables the inclusion of supplementary data that influence production. These data points help derive the model's physical parameters, leading to more accurate production rate calculations, and improving production forecasts. An extensive assessment was conducted using publicly available data, including Duvernay formation, SPE-RTA well data, and Volve oil field. Three different methodologies were used to compute future production rates: traditional decline curves, ML, and HDP modeling. The results compared with different statistical metrics, demonstrated that HDP model consistently exhibited superior precision in production forecasting. A key advantage of HDP is its ability to generate accurate predictions without extensive training samples (physics acts as a constraint and helps the network to train faster), beneficial for newly established wells with limited production histories. The predictive outcomes align well with fundamental physical models, validating their applicability for short-term and long-term production forecasting. For a real-world case, horizontal wells in Duvernay were used (featuring different well designs, drilling, and completion parameters) to build different predictive models and then generate thousands of scenarios. These scenarios are used to find an optimized solution for drilling and hydraulic fracturing operations. Afterward, the best scenario was selected using multi-criteria decision-making and decision tree methodologies. The applicability of HDP was extended to generate reliable sweet spot maps for new drilling targets. Finally, this research advances production performance evaluation by bridging the gap between data-driven and physics-based approaches and enhances the accuracy and reliability of production forecasts and optimization, offering a robust tool for both operators and researchers.enUniversity of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission.Production OptimizationProduction ForecastingHybrid Data-PhysicsMachine LearningReal-TimeEngineering--PetroleumArtificial Intelligencedoctoral thesis