Artificial neural network models to predict lost circulation in natural and induced fractures

Abstract

Mud loss is a challenging obstacle in the oil and gas industry. Predicting mud loss can be very useful to stop or prevent this problem. In this study, data of more than 3500 wells collected worldwide were used to create two neural network models to predict mud loss in natural and induced fractures. For both networks, data were separated into three sets: 60% for training, 20% for validation, and 20% for testing. The number of hidden layers and the number of neurons in each hidden layer were optimized after multiple trials. The findings proved that the created models can estimate mud loss for natural and induced fractures within a small error. The overall R2 for the natural fractures model was 0.956 while the overall R2 for the induced fractures was 0.925. To further investigate and verify the created networks, both models were tested on 24 new wells (wells not used in the process of constructing the networks). The results indicated the models’ predictions closely tract the actual mud loss data with a maximum error of 6.34%. The models have proved their robustness in predicting mud loss and can be used worldwide for mud loss prediction as well as mitigating mud loss by altering the key drilling parameters to prevent or minimize mud loss.