Recognition method of drilling conditions based on bi-directional long-short term memory recurrent neural network and conditional random field

In conventional drilling operations, the operation working conditions of drill rigs is mostly identified using the mechanism model and manual determination, which fails to deliver desirable timeliness and accuracy for capturing the working conditions of rigs. Given this, the artificial intelligence algorithms emerging recently were applied. Nine drilling parameters were used as the input features, namely the difference between the well depth and bit position, bit position, well depth, hook height, hook load, rotation speed, weight on bit, torque and pump rate, and the drill rig operation working conditions intelligent recognition model was built, trained and optimized. Based on the bi-directional long-short term memory recurrent neural network and conditional random field, this model can identify totally 20 types of drill rig working conditions, such as composite drilling, slide drilling, lifting-circulating-reaming, lowering-circulating-reaming, idle, setting and circulating without penetration, with the accuracy of the training set is 96.49%, and the testing set is 97.23%. The development of this model demonstrates the superiority of the artificial intelligence algorithm and provides abundant experience for future applications of such algorithms in drilling engineering.