Maximising Drilling Rates with Real-Time Data-Driven Drilling Parameters Optimisation

Abstract

The hardness and high abrasiveness of shale formations pose great challenges for improving drilling performance with deeper and more complex unconventional gas-reservoir formations explored and developed in the Sichuan Basin of China. The work discussed a workflow of data-driven drilling-parameter optimization based on machine learning. The algorithm was used to construct the correlation among drilling parameters, lithology change, vibration, bit wear, and hole cleaning, dynamically optimize rock-breaking efficiency, and integrate with the rig control system. The data-driven drilling optimization workflow consisted of exploration mode, learning mode, and application mode. The exploration phase trained a linear model at a certain frequency during data collection and updates the increasing trend of rate of penetration (ROP) in real time after starting the workflow. Based on the trend, the direction for further exploration was given. The system entered the learning mode after sufficient exploration to learn the exact functional relationship between ROP/MSE (mechanical specific energy) and operation parameters in the current explored data queue. Current optimal operation parameters were presented based on the function relationship. Then the workflow entered the application mode, maintained the current optimal operating parameters, and kept the efficient rock-breaking state. The workflow constantly monitored the micro-interval ROP and bit energy output in the application mode. When drilling performance was under expectation, the workflow automatically evaluated new conditions (e.g., formation change, Bottom hole assembly (BHA) vibration, and cuttings bed) and switched to the learning mode or exploration mode to adapt to changes in the current drilling state. The algorithm has been integrated with the rig control system, and the field test was carried out in well Ning 209H71-3, a shale-gas horizontal well in the Sichuan Basin. The test showed that the random forest and support vector machine algorithm could fit the nonlinear function relationship among drilling parameters, hole cleaning, and bit working performance, with properly optimized parameters presented. Besides, the workflow could evaluate the trends of ROP, MSE, depth of cutting (DOC), and stick-slips (SS) to capture the limiters for drilling performance, such as bit wear and lithology changes. Two modes integrated with global and local recommendations, and the optimal parameters have been provided to drillers in time. The field performance showed about 20% of ROP improvement with the recommended parameters along the horizontal section, and a 3,100-m horizontal section has been achieved. Machine learning algorithms were applied to drilling parameter recommendations with lower manual intervention. The novel workflow is not limited to bit type, downhole tools, rig equipment, etc. It has shown an outstanding drilling improvement in complex unconventional gas wells, which led the conventional drilling process to the automated era.