Asset optimization has recently become a crucial issue in Oil&Gas industry, considering oil price conjuncture and an increased awareness on environmental aspects. In this paper, an Artificial Intelligence (AI) technique is presented, which is able to manage big dataset to automatically match the entire production model against measured field data. The tool is based on a hybrid in-house developed AI technique, integrating deep neural networks, biogenetical algorithms, commercial simulators and real-time data. The workflow starts with the modeling of the production system through physics-based commercial simulators. A sensitivity analysis identifies the critical variables, which are then randomly varied with a Sobol distribution, exploring the entire solution domain. With these data, a proxy model to the commercial software is generated using an artificial neural network. Finally, the AI tool fed by real-time data is used to match the field behavior: uncertain parameters are modified through a differential evolution algorithm that minimizes the error between calculated and measured variables. The matching parameters are, then, passed to the simulators achieving a field representative model. The tool has been developed considering an operating field in offshore western Africa. The typical uncertain parameters in this kind of field are related to the fluid characteristics, in particular densities and compositions, but also to the physical characterization of the pipelines such as roughness and heat transfer characteristics. The matching process has been performed coupling the proxy model, which is a neural network able to replicate the field behavior, and a differential evolution algorithm as the optimization algorithm. The fitness function to be minimized is a Mean Absolute Percentage Error (MAPE) that represents the distance between the actual field production parameters and the modelled ones. The best configuration of both the neural network and the differential evolution algorithm required a computational time of 6 seconds with a MAPE equal to 2.6%. These results are compared to the one obtained coupling the same differential evolution algorithm with the commercial simulator to perform the matching. The required computational time is equal to about 20 hours (70400s) and a MAPE equal to 2.2%. The big gain with the novel approach is clearly the knocking down of computational time with a comparable error. In this paper, it has been shown how substituting the physical model with a proxy one can give substantial advantages in terms of computational time. In principle, with the velocity of the tool implemented, the matching procedure could be done on a daily basis. This is a breakthrough because it allows having the simulator model always tuned and ready to be utilized. © Copyright 2018, Society of Petroleum Engineers.

Hybrid Artificial Intelligence Techniques for Automatic Simulation Models Matching with Field Data

Alessandro Niccolai;Marco Mussetta;Francesco Grimaccia
2019

Abstract

Asset optimization has recently become a crucial issue in Oil&Gas industry, considering oil price conjuncture and an increased awareness on environmental aspects. In this paper, an Artificial Intelligence (AI) technique is presented, which is able to manage big dataset to automatically match the entire production model against measured field data. The tool is based on a hybrid in-house developed AI technique, integrating deep neural networks, biogenetical algorithms, commercial simulators and real-time data. The workflow starts with the modeling of the production system through physics-based commercial simulators. A sensitivity analysis identifies the critical variables, which are then randomly varied with a Sobol distribution, exploring the entire solution domain. With these data, a proxy model to the commercial software is generated using an artificial neural network. Finally, the AI tool fed by real-time data is used to match the field behavior: uncertain parameters are modified through a differential evolution algorithm that minimizes the error between calculated and measured variables. The matching parameters are, then, passed to the simulators achieving a field representative model. The tool has been developed considering an operating field in offshore western Africa. The typical uncertain parameters in this kind of field are related to the fluid characteristics, in particular densities and compositions, but also to the physical characterization of the pipelines such as roughness and heat transfer characteristics. The matching process has been performed coupling the proxy model, which is a neural network able to replicate the field behavior, and a differential evolution algorithm as the optimization algorithm. The fitness function to be minimized is a Mean Absolute Percentage Error (MAPE) that represents the distance between the actual field production parameters and the modelled ones. The best configuration of both the neural network and the differential evolution algorithm required a computational time of 6 seconds with a MAPE equal to 2.6%. These results are compared to the one obtained coupling the same differential evolution algorithm with the commercial simulator to perform the matching. The required computational time is equal to about 20 hours (70400s) and a MAPE equal to 2.2%. The big gain with the novel approach is clearly the knocking down of computational time with a comparable error. In this paper, it has been shown how substituting the physical model with a proxy one can give substantial advantages in terms of computational time. In principle, with the velocity of the tool implemented, the matching procedure could be done on a daily basis. This is a breakthrough because it allows having the simulator model always tuned and ready to be utilized. © Copyright 2018, Society of Petroleum Engineers.
Abu Dhabi International Petroleum Exhibition & Conference
978-161399632-4
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11311/1119357
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 4
  • ???jsp.display-item.citation.isi??? ND
social impact