This paper was prepared for presentation at the Young Professional Session of the SPE Russian Oil and Gas Exploration and Production Technical Conference and Exhibition held in Moscow, Russia, 14–16 October 2014.
The technique for selecting candidate wells for hydraulic fracturing and prediction of post-frac well performance using Data Mining tools (neural networks, decision tree algorithm, etc.) is presented. The technique was applied to the BV8 reservoir in the Povkh Oil Field. The reservoir is characterized by the complex clinoform structure.
The objective of the paper is to determine basic criteria for candidate selection and evaluate the effect of different geological and field conditions on the efficiency of hydraulic treatments. To achieve the objective two problems have been solved:
Development of classification models (efficient/inefficient hydraulic fracturing treatment);
Development of regression models (post-frac fluid rate, water cut, oil rate).
Machine learning algorithms were used in petroleum industry in the past, but neural networks is the most common technique. Eleven models have been built based on six algorithms. A comparison has been made between the models. A good consistency between predicted values of post-frac fluid rate and water cut and actual results has been revealed.
Major geological and field criteria for the efficiency of hydraulic fracturing operations have been identified. The models built can be applied to predict the post-frac performance of a single well as well as to do a quick estimation of the whole well stock. The results presented show significant predictive power of the Data Mining tools for geologically-complex reservoirs. It should be assumed that the usage of the technique for less complex reservoirs will allow us to significantly improve prediction of well performance.
Number of Pages
Looking for more?
Some of the OnePetro partner societies have developed subject- specific wikis that may help.