Modeling of drilling rate of penetration using adaptive neuro-fuzzy inference system

Drilling rate of penetration (ROP) is a crucial factor in optimizing drilling cost. This is mainly due to the excessive cost of the drilling equipment and rig rental, where the longer the drilling activity would reflect a higher expenditure. If the drilling rate of penetration can be predicted accurately, we would be able to avoid unnecessary spending. Hence, this can lead to minimizing the drilling cost significantly. In this paper, an Adaptive Neuro-Fuzzy Inference System (ANFIS) model is generated using MATLAB environment. A total number of 504 data sets from a Sudanese oilfield is used to develop a well-trained and tested ANFIS model for ROP prediction. The parameters included in the model generation are: depth, bit size, mud weight, rotary speed and weight on bit. Training options were set to give the best predicted ROP against the real data. This model is proven to give a high performance with an error as low as 1.47% and correlation coefficient of 98%. With this model, the estimation of the duration of drilling activities in the nearby wells can be done accurately if relevant data from the same reservoir is available. Caution must be taken to avoid using the results from this model beyond the range of training data.