New Tools for Fracture Design Optimization

Abstract

Abstract Designing an effective and optimum fracture treatment is a complex and time-consuming task, especially in the case of new fields where the number of unknown variables is much larger than the number of known variables. Even with growing field maturity, changes in the fracture design often tend to be marginal and do not lead to a drastic improvement in the fracture design manifested by the post fracture production. This is not say that even marginal treatments are economically unattractive. Fracturing even when done badly is still one of the most economically rewarding petroleum operations. What we are addressing is a much more advanced and refined approach to design where economic benefits can be doubled or tripled. One explanation for the absence of continuous improvement of fracture treatments is lack of the appropriate tools to perform a thorough data-mining task, which would lead to a definitive improvement in post fracture production benefits. The appropriate tools for this task are presented in this paper. Although these tools are commonly used in other industries, they are still relatively new to most petroleum engineers. This paper provides an introduction to the technology and procedures to mine large databases for hidden, but valuable underlying relationships. An included comprehensive case study presents the workflow of the data mining approach to fracture optimization. A combination of Neural Networks and Genetic Algorithms, both regarded as artificial intelligence tools, are first applied to the data. The Neural Network is trained on historic data and captures the relationship between a certain data set of input parameters and the corresponding output parameter. This relationship is stored in the architecture of the Neural Network. In a following step, the Neural Network is used by the Genetic Algorithm optimization tool to calculate the optimum values for the various input parameters: in this case to maximize the output – the production after fracture treatment. Genetic algorithms are inspired by Darwin's law for the survival of the fittest. For the particular case the fittest values for the different parameter are those, which generate a high post fracture production. The results are than verified with the Unified Fracture Design approach. Neural Networks use historic data to train. It is required that the data show significant variation to capture the relationships between input and output. Data sets, which show too low variation in production (mainly on the poor side), cannot be optimized based on the artificial intelligence approach. For this case data mining results are integrated into the Unified Fracture Design approach, which will point to the optimum design. The combination of data mining with an appropriate fracture optimization design approach is a powerful combination in production engineering and one assured to provide large incremental economic benefits to the industry.