Optimization of drilling parameters using combined multi-objective method and presenting a practical factor

Abstract

Several optimization approaches have been developed to enhance the efficiency of drilling performance. The characteristics of the problem require a procedure capable of a precise definition of the model while optimizing the factors. Therefore, a hybrid machine learning approach was proposed using random forest (RF) and multi-objective particle swarm optimization (MOPSO) to optimize the drilling factors. The objectives were to achieve a higher rate of penetration (ROP) and lower torque on bit (TOB). The geomechanical and energy-based features were involved in the settings of the optimization process. At first, a dataset was made consisting of four wells and 7231 points with carbonated reservoirs. Next, the inputs including vital controllable parameters were preprocessed based on requirements to reveal their relationships with the targets. The flow rate of tools was the most influential factor as suggested by RF's permutation analysis. The process followed with proposing the models of ROP and TOB through the training section using RF learning. The models of training and testing parts showed a correlation coefficient of about 0.97 and 0.84 for the TOB and ROP of wells, respectively. The expression of the constraints and cost functions is of crucial significance in every optimization scheme. Here, the cost functions were established via RF models and constraints defined as the ratio of drilling efficiency (DE) to weight on bit (WOB). DE is related to confined compressive strength (CCS) and mechanical specific energy (MSE) thus, the combined RF-MOPSO is implemented by specifications of the formations and drilling aspects. The introduced practical ratio revealed the zones with the least drilling efficiency, devoted to the testing section and optimization target. As outcomes of RF-MOPSO, the ROP was considerably improved by 43.8% while TOB showed a reduction to 9% on average of the wells. Another advantage of the multi-objective optimization corresponded to factors of means MSE and DE, showing optimal values. Consequently, the concept of hybrid machine learning with multi-objective optimization can (1) enhance the speed and accuracy of providing drilling modeling and 2) optimize drilling factors using a hybrid method simultaneously with proper real-time tools. Moreover, the proposed ratio of DE to WOB connects the rock mechanical side to the drilling factor, making it more possible to predict the efficiency and hazards of drilling.