A novel online learning RECOC-KFDA method for lithologic identification in drilling process

Aiming to achieve a safe and efficient drilling, this paper is concerned with identification of formation lithology, which provides critical information for drilling control. Notice that it is hard to make accurate geological prediction using conventional identification approaches, due to the data characteristics of imbalanced, multi-classification and low value density, a novel reduction error correcting output code kernel fisher discriminant analysis algorithm(RECOC-KFDA) method is developed in an online manner. It consider design optimal error correcting output code(ECOC) matrix based on a reduction algorithm, and it proposed an online method to reduce the computation complexity required for updating the kernel fisher discriminant analysis(KFDA) classifiers rather than recalibrating them. Proposed method has been applied to lithologic identification in drilling site. Simulations and comparisons demonstrate that our method is superior to the existing ones for both offline training and online prediction model.