Intelligent prediction models of downhole weight on bit based on genetic optimization algorithm

Accurate downhole weight on bit (DWOB) is important to improve drilling efficiency. Over recent years, increasing studies have shown that intelligent technology provides an effective way for precisely predicting the target value. Four intelligent prediction models of DWOB (i.e. BP, LSTM, GA-BP and GA-LSTM) were built by combining back propagation (BP) neural network and long short-term memory (LSTM) neural network, and also integrating BP with LSTM and genetic algorithm (GA). Experiments validate that GA does deliver some optimization of the prediction model, with higher prediction accuracy, higher robustness, more consistent prediction trend and faster computation. The GA-LSTM and GA-BP models exhibit the relative error of 40.13% and 47.11%, respectively, less than the single LSTM or BP model, and their computation time is shortened by 12.6 times and 9.3 times, respectively. GA-LSTM is identified as the optimal intelligent prediction model of DWOB, for its overall performance. It can be applied to real-time monitoring of DWOB or combined with a conventional automatic running system to realize accurate control of DWOB, so as to improve drilling efficiency and bit performance, and reduce drilling costs.