A Predicting Model for Near-Horizontal Directional Drilling Path Based on BP Neural Network in Underground Coal Mine

Abstract

This study establishes a prediction model based on the back propagation (BP) neural network for controlling the underground directional drilling path in a coal mine. The four-layer BP neural network model chooses 11 trajectory parameters (dip angles and azimuths from 12 m after the measurement while drilling (MWD)) and two control parameters as inputs. Two parameters, the dip angle and azimuth at bit, are the outputs. Trained with data from 502 groups, the model was used to forecast 12 test data groups. The results were then compared with the prediction results of artificial experience from 24 technicians. The study shows that the mean absolute error of the dip angle and azimuth at bit are only 0.51° and 0.68°, respectively, as predicted by the prediction model, which uses the logsig activation function and has a double-hidden-layer with a point structure of 9×6. The prediction error also follows a normal distribution. Compared with technicians who have worked for more than five years, the accuracy of prediction results from the BP neural network model is reduced by 33.9% and meets the needs of drilling path control.