Uniaxial Compressive Strength Prediction of Rocks based on LSTM-Adaboost from Multi-Sources

Rock strength is a crucial factor in measuring rock stability and safety. Accurate prediction of rock strength can effectively guide mine excavation and blasting design. Uniaxial Compressive Strength (UCS) is a significant parameter in rock mechanics, and its precise prediction can reflect the rock strength. Research on UCS prediction models based on drilling parameters often relates to the nature of the drilling rig, making it difficult to extend the application to other platforms. This paper proposes an LSTM-Adaboost algorithm that collects drilling data from four different sources, training four LSTM weak learners separately, and outputs LSTM strong learning machine that can be applied to different sources of drilling data through the Adaboost fusion algorithm. As drilling data has temporal backward and forward dependencies, LSTM is adopted as weak learners for its capability of maintaining long-term dependencies by using internal memory units. The effectiveness of the algorithm is verified by comparing it with the BP neural network, random forest algorithm, and LSTM without fusion algorithm.