Build-up rate prediction using data augmentation with VAE-based feature extraction

The build-up rate (BUR) prediction is crucial for reliable trajectory control in the field of drilling. Machine learning-based prediction methods have superior performance compared to conventional methods. However, acquiring sufficient BUR data is challenging, which not only renders prediction models prone to overfitting but also compromises the quality of generated data due to the difficulty in capturing the real distribution of the original data for generative models. To address the two issues, a new data augmentation method with VAE-based feature extraction is proposed, which integrates the variational autoencoder (VAE), generative adversarial network (GAN), and support vector regression (SVR). Specifically, the VAE is employed as a feature extractor rather than a generator due to its strong feature extraction ability and learning ability of data distribution, thereby enhancing the generation ability of the augmentation model. Then, the GAN and SVR are constructed to generate synthetic data. Finally, based on the generated data and the real data, the SVR model optimized by grey wolf optimizer (GWO) is built to achieve the accurate prediction of the BUR. To validate the performance of the proposed method, the drilling data from the Z48 well is applied, and the results indicate the proposed augmentation method can improve the quality of the generated data and the prediction accuracy of the BUR. The mean absolute error (MAE) and mean square error (MSE) are reduced by 25.1 % and 53.8 %, respectively, compared to existing methods.