Intelligent real-time predicting method for rock characterization based on multi-source information integration while drilling

Real-time digital description of rock boring while drilling process is of great significance for the development of geological exploration technology. With the natural change of geological environment, drilling bit interaction with rock mass will produce different characteristics of the rock-related information obtained while drilling, so the information can be used to indicate our rock characteristics in real time. However, due to the complexity of geological environment, it is difficult to directly establish the correlation function model between observed signals and rock characteristics. In this paper, we put forward the concept of multi-source drilling process information integration. For effective input signal sources, KNN (K-nearest neighbors) algorithm was used to predict rock identification quality based on each signal source. Based on the independent prediction of rock identification quality by using several kinds of signals, the integration classification and prediction of several kinds of rocks were realized by using conditional probability and minimum risk decision theory. The prediction results show that the accuracy of multi-source decision level information integration based on conditional probability and minimum risk is significantly better than that based on single information source, and the method has stronger learning performance and generalization ability for small samples and unbalanced samples.