New Method for Flow Rate and Bottom-Hole Pressure Prediction Based on Support Vector Regression

Continuous real-time predictions of flow rate and bottom-hole pressure are important components of an intelligent well system for realizing closed-loop control production. The accuracy of the prediction is largely dependent on a reservoir numerical simulation model. Ideally, the reservoir simulation model can be updated constantly and quickly according to real-time measurement data. However, the existing reservoir numerical simulation technology cannot achieve this requirement. Therefore, this study proposes a new prediction method of nonlinear regression modeling based on support vector regression theory and moving-window technology. In the case of an unknown reservoir model and other parameters, this new method can establish a dynamic prediction model using the permanent downhole gauge (PDG) data of an intelligent well as input. The prediction model can be trained and updated continuously and quickly with the latest PDG data from the data acquisition system to achieve a real-time prediction of the flow rate and bottom-hole pressure. The effectiveness of this method is verified by using two examples, namely, simulated data from a reservoir model and real data from an oil field. Evaluation indexes show that the prediction results meet the precision requirements. Therefore, the proposed method based on support vector machine provides a new solution for the realization of the closed-loop control system of intelligent wells.