Machine learning application for oil rate prediction in artificial gas lift wells

Well production rate is one of the most critical parameters for reservoir/production engineers to evaluate performance of the system. Given this importance, however, monitoring of production rates is not usually carried out in real time. Some cases flowmeters are used which are known to carry their own inherent uncertainties. The industry, thus, relies on the use of correlations to allocate production to wells. Over time, it has been realized that the generally used correlations are not effective enough due to multiple technical and economic issues. The focus of this work is to utilize machine learning (ML) algorithms to develop a correlation that can accurately predict oil rate in artificial gas lift wells. The reason for using these algorithms is to provide a solution that is simple, easy to use and universally applicable. Various intelligent algorithms are employed, namely; Artificial Neuro Fuzzy Inference Systems (ANFIS), and Support Vector Machines (SVM), along with the development of Artificial Neural Network providing a usable equation to be applied on any field, hence demystifying the black-box reputation of artificial intelligence. In addition, non-linear regression is also performed to compare the results with ML methods. Data cleansing and data-reduction were carried out on the dataset comprising of 1500 separator test points. This practice yielded in only the common wellhead parameters to be used as input for the model. All ML models were compared with the non-linear regression model and with previously derived empirical models to gauge the effectiveness of the work. The newly developed model using ANN shows that it can predict the flow-rate with 99% accuracy. This is an interesting outcome, as such accuracy has not been reported in literature usually. The results of this study show that the correlation developed using ANN outperforms all the current empirical correlations, moreover, it also performs multiple times better in comparison to previously developed AI models. In addition, this work provides a functional equation that can be used by anyone on their field data, thereby removing any ambiguities or confusion related to the concept of artificial intelligence expertise and software. This effort puts forth an industrial insight into the role of data-driven computational models for the production reconnaissance scheme, not only to validate the well tests but also as an effective tool to reduce qualms in production provisions.