Drilling performance is greatly driven by the rate of penetration (ROP) and its optimization. Therefore, accurate prediction of ROP before drilling can help take measures to achieve drilling objectives smoothly. Numerous inter-twined formation and drilling parameters affect the ROP. Consequently, optimizing and predicting ROP is a complex challenge faced by the industry. This study proposes a novel computational intelligence-based (CI) model to predict ROP as a function of multiple readily available mud- and petrophysical- logs. Furthermore, multiple algorithms are utilized, and a comparative analysis is presented.
Adaptive Neuro Fuzzy Inference System (ANFIS), along with Artificial Neural Network (ANN) based algorithms are utilized to analyze the data and develop models. A process-based approach is followed starting with systematic data analysis, which includes a selection of the most relevant input parameters, data cleaning, filtering, and data-dressing to ensure optimized inputs into the computational models. This step allows for developing ROP predictor as a function of pivotal formation-related logs such as neutron, density, wave velocity, and gamma ray. Moreover, historical drilling-related parameters such as weight on bit, mud pump parameters, and bit flow rate are also used as model inputs. Next, CI model parameters are tuned for both ANN and ANFIS through a sensitivity analysis on the number of neurons and the cluster radius, respectively. Finally, a comparison of ANN and ANFIS models is performed using graphical and statistical analysis.
The developed models are tested on an unseen dataset to verify their efficiency. An error metrics analysis, comprising of average percentage error, root mean squared error, and the correlation coefficientis performed. This analysis ranks the respective CI models based on the highest performance efficiency and lowest prediction error. Resultantly, it is observed that the ANN-based model can perform better than the ANFIS model by accurately predicting ROP. The superior performance of ANN is due to its ability to map the complex non-linear interactions of formation and drilling parameters to the resultant ROP.