Virtual Metering and Allocation using Machine Learning Algorithms

Abstract

In this paper a machine learning algorithm based on neural networks is proposed. The aim is to test a model for multiphase flow rates estimation, taking into consideration the process parameters and the performance of the installed equipment. This model can work in parallel with the Multiphase Flow Meter (MPFM) measurements or acts as a back-up when it fails or it is not present. The model mainly consists in a multilayer feed-forward neural network, suitable for non-linear regression problems such as the one here considered. The model solution takes a set of reservoir and process parameters as inputs and returns a set of fluid flow information as outputs. The model is trained using actual well test data from producing wells. Generalization and network complexity are regulated using cross-validation. In addition, the mean square error is used as the performance function for training the feed-forward neural network. The predictive capabilities of the neural network for estimating oil and gas flow rates in multiphase production wells have been assessed against field measurements collected from several production wells operating in west Africa. The developed machine learning model shows promising results and a high level of accuracy, with predicted output very close to the actual MPFM measurements for oil and gas flow rates for specific data considered in the present analyses. Thus, the model can be applied to predict multiphase flow rates in different wells and acts as a back-up when MPFM fails or is not present. The great advantage provided by this model is its inherent simplicity and the small computational time taken to provide an output. Therefore, it can also be thought as a valid alternative to virtual flow metering techniques that are based on physical models and are highly dependent on variables with a high degree of uncertainty (e.g. fluid properties). This work presents a promisingapproach for virtual metering technique. The methodology has possible application in real time monitoring but also in production optimization, production allocation and reservoir management and it is a cheaper alternative with respect to MPFM, reducing in this way operational and maintenance costs.