Intelligent Prediction Model of Downhole Weight on Bit Based on GA-Bi-LSTM and Its Application

In drilling operations, the accuracy of downhole weight on bit (DWOB) constitutes one of the crucial factors in determining drilling efficiency and drilling cost. In this paper, we use genetic optimization algorithm (GA) to optimize the parameters in the bidirectional long and short-term memory (Bi-LSTM) neural network, and establish a GA-Bi-LSTM intelligent prediction model of DWOB based on the comprehensive consideration of each drilling parameter, and compare with the established single BP, LSTM, Bi-LSTM model and the BP, LSTM intelligent prediction model combined with genetic algorithm. The experimental results show that the genetic algorithm plays an optimization role to a certain extent and shows higher prediction accuracy, better robustness and prediction trend. The average relative errors of GA-Bi-LSTM, GA-LSTM, and GA-BP are 32.13%, 34.98%, and 47.77% lower, respectively, than those of the single Bi-LSTM, LSTM, and BP models. GA-Bi-LSTM can be selected as the optimal intelligent model for DWOB considering all aspects of performance, which provides a new idea and method for DWOB and can be seamlessly incorporated into the current automatic drilling system to achieve accurate control of DWOB during automatic drilling delivery, and also has application potential in post-drilling analysis, drilling operation monitoring, drilling optimization and automation.