Near Real-Time HDD Pullback Force Prediction Model Based on Improved Radial Basis Function Neural Networks

Abstract

Pipeline pullback is a crucial part of horizontal directional drilling (HDD) construction. Accurate pullback force prediction is the prerequisite for ensuring construction safety. However, owing to the influence of factors such as crossing length and formation conditions, it is difficult to predict the pullback force accurately using existing theories. In this paper, a hybrid model based on radial basis function neural networks (RBFNNs) is proposed, which can predict near real-time pullback force based on field monitoring data in the construction process. In this hybrid model, complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is used to denoise the original data, which is conducive to better performance of the RBFNN model in prediction. To test the prediction accuracy of the proposed model, this paper takes two HDD projects in the Sichuan–East China Gas Project as examples. In addition, the stability of the prediction model and the effect of the sliding window length on the prediction results are discussed. The following conclusions can be drawn: (1) the proposed model has higher prediction accuracy than the empirical model, (2) the application of the denoising method can effectively improve prediction accuracy, and (3) the hybrid model has higher prediction stability than the original RBFNN.