Real-time Optimization of Hydraulic Fracturing Pumping by Machine Learning and Multi-condition Constraints

ABSTRACT: Intelligent fracturing, particularly the real-time decision-making of pumping parameters, is a key development direction in current hydraulic fracturing technology. Presently, decision-making in fracturing pumping relies heavily on limited real-time monitoring data and expert experience, leading to potentially suboptimal fracturing jobs. Leveraging extensive historical data to autonomously determine fracturing parameters is crucial for achieving intelligent fracturing. This study introduces an innovative method for real-time optimization of hydraulic fracturing pumping parameters using feedbacks from two proposed wellhead pressure prediction models. The method accurately predicts the fracturing system response to various adjustment schemes, accounting for the time delay until the ground proppant reaches the well bottom. The optimization objectives, involving pressure stability and multi-condition constraints such as operational restriction pressure, hydraulic horsepower, critical proppant concentration, design compliance rate, and construction time, are designed to ensure safe and efficient operations. A case study of pumping optimization using data from two horizontal wells is presented, demonstrating that our method achieves reasonable optimization results. This paper provides a new framework aimed at guiding safe and efficient fracturing operations, with significant potential for cost savings, risk mitigation, and production enhancement, lays the foundation for hydraulic fracturing 'autonomous driving'.