Estimation of Water Cut, Gas Oil Ratio and Inflow Parameters for Deep Water Wells with Infrequent Well Tests Using Regression-Based Method

Abstract

Abstract Real-time production surveillance and optimization requires availability of predictive well models with reasonable accuracy that can estimate production rates in response to change in operating conditions. These models that are mostly physics based are parameterized in terms of gas oil ratio (GOR), water cut (WC), productivity index (PI), and reservoir pressure. Further, the models are tuned using measured data namely production rates of oil, water, gas, and pressure measurements such as downhole gauge pressure, wellhead and flowline pressures that are captured during a well test. For mature oil fields where, well tests are infrequent or sensors start malfunctioning, relevant data required for model tuning is no longer available. In absence of updated models, real-time predictions tend to deviate with time compared to the observed data resulting in less reliable well rate allocation and production optimization recommendations, if any. This work describes a regression-based technique and its application for predicting quantitative as well as directional changes in well parameters between well tests that can be used to improve well allocations. A regression method that estimates well parameters while minimizing a non-linear least square loss function derived from deviation between measured and model-based estimates of rate and pressure data is implemented. The method can estimate well parameters for both an individual and multiple wells simultaneously while solving for a network of wells connected to production separator using IPM GAP. The application of the method to a subsea asset is demonstrated while evaluating its performance for different scenarios comprising variation in number of wells and well parameters. Additionally, the capability of the method to predict directional changes in well parameters is demonstrated by validating it against historical data. The estimates from regression method were found within 5% difference compared to well parameters obtained from multiphase flow meters for the scenarios where number of wells and the number of regression parameters per well were limited. This difference increased with increase in number of wells and number of regression parameters per well owing to the fact that solution space expands with increased degrees of freedom. However, the directional changes in parameters were predicted accurately when looked at larger time scales. It was inferred that the application of regression method is best suited for the scenarios where well parameter estimations are needed for a limited number of wells and the parameters for the remaining wells in a network are representative. Additionally, availability and reliability of sensor data largely impacts the method outcomes.