Development of a gas flow rate model for multi-stage choke system in HPHT gas wells using artificial intelligence

Objective/Scope: Accurate well production rate measurement is critical for reservoir management. The production rate measurement is carried out using surface devices, such as orifice flow meter and venturi flow meter. For large offshore fields development with a high number of wells, the installation and maintenance costs of these flowmeters can be significant. Therefore, an alternative solution needs to be developed. This paper described the successful implementation of Artificial Intelligence in predicting the production rate of big-bore gas wells in an offshore field. Methods, Procedures, Process: Successful application of AI depends on capitalizing on a large set of data. Therefore, flowing parameters data were collected for more than 30 gas wells and totaling over 100,000 data points. These wells are producing gas with slight solid production from a high-pressure hightemperature field. In addition, these wells are equipped with a multistage choke that reduces the noise and vibration levels. An Artificial Neural Network is trained on the data using Gradient Descent method as the optimization algorithm. The network takes as an input the upstream and downstream pressure and temperature, and the choke size. The output is the gas rate measured in MMscf/day. Results, Observations, Conclusions: The data set was divided into 70% for training the neural network and 30% for validation. Artificial Neural Network (ANN) was used and the developed model compared exceptionally well with the gas rates measured from the calibrated venturi meters. The gas rate estimation was within a 5% error. The model was developed for two types of completions: 7″ and 9-5/8″ production tubing. One of the challenges was how to estimate the choke wear which plays a major role in the quality of the choke size data. A linear choke wear deterioration is applied in this case, while work in progress is taking place for acquiring acoustic data that can significantly improve the choke wear modeling. Novel/Additive Information: The novel approach presented in this paper capitalizes on Al analytics for estimating accurate gas flow rate values. This approach has improved the reservoir data management by providing accurate production rate values which has drastically improved the reservoir simulation. Moreover, the robustness of the AI model has forced us to rethink the conventional design of installing a flow meter for every well. As shown in this paper, the AI model served as an alternative to conventional venturi meters. We believe that the application of AI models to other aspects of production surveillance will lead to a shift into how operators design production facilities.