Fusion model of weight on bit in horizontal exploration hole based on wavelet transform and machine learning

Abstract

During the horizontal drilling of the development of energy and the geological exploration, the loss of axial feed pressure is serious due to the frictional resistance in horizontal hole. As the length of the drilling pipe increases, the problem of insufficient drilling pressure becomes increasingly prominent. It is imperative to master weight on bit (WOB) in time and accurately for the adjustment of drilling parameters. Limited by the long drilling pipe, it is difficult and costly to measure WOB directly. Therefore, a novel model of WOB prediction was established. The model realized the coupling of wavelet transform and neural network. Five easily available parameters, feed pressure, rotational speed, torque, pump pressure and drilling-fluid flow rate were selected as input to the model. The WOB was taken as the only output of model. First, Morlet wavelet was used to reform the structure of hidden layer and output layer of neural network. In the reformed model, the self-learning and adaptive ability of neural network and the ability for local information extraction of wavelet transform were fully stimulated. Then, embedded genetic algorithm (GA) was used to optimize the weights and wavelet parameters of the model, which improves the global optimization ability of the model. The established model was named GA-WNN, and the experimental results demonstrate that the R2 of prediction model reaches 0.9982. The mean absolute percentage error (MAPE) of GA-WNN prediction can yield as low as 6.22%, while the MAPE of standard neural network prediction can yield as high as 47.04%. Furthermore, the accuracy of the model was further verified by response surface methodology (RSM), and the results show that the predicted data of the model can meet the requirements of RSM for data quality. The RSM analysis of WOB based on predicted data was successfully carried out, and the interaction influence between drilling parameters on WOB were obtained, which provides reference for parameter adjustment. The high-quality algorithm of WOB prediction provides support for long-distance horizontal drilling.