Extended multi-kernel relevance vector machine optimized Kriging interpolation for coal seam thickness prediction in coal-bearing strata

During the drilling process of a coal mine roadway drilling rig, coal seam thickness variation affects the efficiency of coal seam mining. However, the cost of geological drilling required for coal seam exploration in practical engineering is high, the sample size of the data is small and the distribution is discrete, and spatial interpolation is required for coal seam thickness prediction in unexplored coal seams. Therefore, this paper proposes an improved method of kriging spatial interpolation for small sample, acquired from geological drilling. Firstly, for the small sample problem, we use a Relevance Vector Machine (RVM) to reconstruct the variogram in kriging interpolation. Secondly, multi-kernel RVM (MKRVM) is used to improve the fitting effect in global and local, respectively. Finally, Particle Swarm Optimization (PSO) is used as an extension of MKRVM to optimize the hyperparameters in the multiple kernel functions and the weights among different kernel functions to improve the fitting effect of the overall model. Through a series of comparative experiments, the superiority of the extended multi-kernel RVM (EMKRVM) method proposed in this paper is verified. At the same time, the method is applied to a practical project, and the results illustrate that our method has lower error in the prediction of coal seam thickness variation in coal-bearing strata, which can provide a better reference basis for the subsequent adjustment of drilling speed, rotation speed, and drilling pressure.