Wellhead Choke Performance Prediction During Multiphase Flow Using Machine Learning

Multiphase flow metering in oil and gas production wells is essential for monitoring the production performance from oil and gas reservoirs. Accurate measurements of multiphase flow rate that passes through the chock at the wellhead assembly is highly important as it provides a crucial information (data) to estimate and forecast production and ultimately assess and optimize the reservoir performance. The current flow rate metering methods is still in its infancy when it comes to multiphase flow metering. Flow rates are most commonly reported for several wells through a manifold that connects multiple wells together and therefore flow rates a rarely reported for each single well. The flow thought the manifold is directed to two different traps (High pressure test trap and high pressure production trap) in the gas oil separation plant. To accurately measure the flow rate for each single well, production tests operations are conducted for every well separately. The flow from all other connected wells via the manifold to the high pressure test trap (HPTT) in the well system has to be shut down and then the flow rate can be accurately measured by measuring the increase of the fluid volumes at the HPTT in the separation plant. Clearly, this method is not practical and is extremely expensive as it will be time-consuming to frequently conduct it for each well separately. The current practice to conduct such production rate test monthly and for the entire wells that are connected through one manifold. Another way to accurately measure the liquid flow rate at the surface through the wellhead chock is using any of the several well-known correlations that are used to calculate the flow rate using production and wellhead chock data. The high uncertainty in production rate predictions is very well expected and the main source of this uncertainty is the reliance on sporadic well test data and empirical multiphase flow correlations to estimate liquid production rate. In the presence of lack of knowledge about the physical relationship between the different parameters that controls multiphase flow, the usage of correlations is common despite the uncertainty it brings. Utilizing Machine learning (ML) techniques can improve the results of these correlations and give more accurate predictions. ML techniques have been utilized as a tool to overcome the complexity faced in the oil and gas industry for more than two decades. It can be used as data interpreting and data mining tool to predict an uncertain outcome with high certainty and overcome the uncertainty in correlations widely used in the industry. It helps predicting data by training and testing an already existed set of data to build a smart predictive model that lowers the uncertainty of outcomes. Prediction and forecasting of production is always uncertain whether by using regression, empirical correlations or any type of models, unless a very well know physical relationship is fully captured, which is rarely the case in the petroleum industry and especially for the flow behaviour in multiphase flow. Throughout many aspects of the industry, Machine Learning has proven its efficiency and advantages. The objective of this project is to develop ML models to predicts the flow rate of critical and subcritical multiphase flow through the wellhead choke and outperforming the industrys widely used correlations. The results of these models are expected to outperform the currently and widely-used correlations in its accuracy to predict flow rate. The main value added of these models is providing a reliable method to predicted flow rate through the wellhead chock by utilizing available surface data. Having reliable measurements of flow rates for each well is extremely valuable in assessing the well and reservoir performance and to forecast and optimize production. Furthermore, these models are expected to provide cost savings by limiting the need to frequently conduct production rate tests at the HPTT in the separation plant. These production rate tests are very time-consuming to be conducted for connected wells via a manifold let alone to be conducted for every single well at a time for the frequency needed to build a production decline trend for each well to forecast production. In this project, artificial neural networks (ANN) techniques will be used to develop a data-driven flow rate computational prediction models for critical and subcritical flow conditions of the multiphase flow through the wellhead chock. For the data acquisition and preparation, a total of 4,366 data sets (flow rates) from 252 wells were collected from four different fields to model the choke performance. The actual and accurate data of flow rates of these wells are measured at the gas oil separation plants by shutting down flow directed to the HPTT from all connected wells via manifold except the assessed well and measuring the flow rate at any specific time by observing the change in the fluid volume at the HPTT. Also, for each well, the chock performance and well production rate tests data were collected and used in these models. Wellhead chock data such as chock size, upstream and downstream pressure and temperature data are used. Also, from well production rate tests for each well, gas liquid ratio, gas oil ratio and water cut data are collected. These data types collected from the production rate tests are mainly reservoir driven and are not expected to have very high variation for several months, unlike flow rate which is expected to change frequently as the wells are being produced.