The Height of Water-conducting fractured zone prediction using principal component analysis and relevance vector machine

Abstract

This study proposes a model to predict the height for water-conducting fractured zone using easily obtained parameters which is Relevance Vector Machine (RVM) based on Principal Component Analysis (PCA). PCA was used to reduce the several factors affecting the water conducted fractured zone height to a few principal components. A nonlinear mapping relationship between the water conducted fractured zone height and principal components was established by using RVM model, and corresponding results of water conducted fractured zone height prediction is inferred. Applying this method to specific examples, comparison and analysis the predicted results of Back Propagation (BP), Partial Least Squares (PLS), Partial Least Squares-Back Propagation (PLS-BP) and Particle swarm optimization-Radial Basis Function (PSO-RBF) methods under the same learning sample conditions, the results reflect that the PCA-RVM prediction method reduces several influencing factors into a few principal components reasonably by analyzing the correlation and contribution rate of various factors, which is significantly better than the remaining 4 methods in information screening. The prediction results of each method show that the PCA-RVM method has the advantages of high precision, small discreteness and high reliability, providing a new method for the prediction of the water conducted fractured zone height.