Intelligent optimization of drill bits by combining multi-source data fusion and deep neural networks

Abstract

ABSTRACT Rational selection of the drill-bit type can effectively increase the drilling speed and ensure the stability of the wellbore. This paper considers the shortcomings in the existing research on the extraction and fusion of multi-source data related to drill-bit types. In this in-depth analysis, the internal connection is established between the multi-source data and the type of drill bit used. An intelligent optimization method has been developed for drill bits by combining multi-source data fusion and deep neural networks. In this method, data extraction, data mining, and fusion analysis are implemented for multi-source data that characterize the formation characteristics by using principal component analysis (PCA). Based on this method, the deep belief network (DBN) mathematical model for drill-bit optimization is established, and the drill-bit type is optimized for the whole well. The PCA–DBN drill-bit intelligent selection method was applied to five wells in the K structural zone in the T basin. The comparison results prove that the PCA–DBN method is a powerful drill-bit selection tool that is more effective and applicable than conventional methods. The proposed method is 17.31%, 5.77%, 21.15%, and 1.92% more accurate than the rock mechanics parameter method, back propagation (BP) neural network, PCA–BP neural network, and DBN neural network, respectively. The accuracy of the drill-bit selection increased by approximately 40.38% as compared with the single type data because multi-source data were used. This proves that the intelligent optimization method that combines multi-source data fusion and deep neural networks can be effectively used for drill-bit selection, this method promotes cost reduction and enhances the efficiency of drilling engineering.