Application research on the prediction of tar yield of deep coal seam mining areas based on PSO-BPNN machine learning algorithm

Abstract

There are abundant deep coal resources in northern Shaanxi, but the fragile natural environment in this area hinders the large-scale exploitation of oil-rich coal. In-situ thermal conversion of deep coal to oil and gas will become an environmentally friendly technology for oil-rich coal mining. Accurate prediction of oil-rich coal tar yield in various regions is a prerequisite. Based on a particle swarm optimization algorithm and two machine learning algorithms, BP neural network and random forest, a prediction model of tar yield from oil-rich coal is constructed in this paper. Combined with the particle swarm optimization method, the problem of slow convergence speed and possibly falling into local minimum value of BP neural network is solved and optimized. The results showed that the PSO-BP had a convergence speed about five times faster than that of the BP neural network. Furthermore, the predicted value of the PSO-BP was consistent with the measured value, and the average relative error was 4.56% lower than that of the random forest model. The advantages of fast convergence and high accuracy of the prediction model are obviously apparent. Accurate prediction of tar yield would facilitate the research process of in-situ fluidized mining of deep coal seams.