A novel evaluation method of mining goaf ground activation under high-speed railway load

Abstract

With the continuous improvement of infrastructure, some high-speed railway lines will inevitably cross the goaf ground, and there is less research on the safety of high-speed rail construction in goaf ground. To make a reasonable and accurate safety evaluation of the high-speed railway construction in the mine goaf ground, machine learning combined with numerical simulation is used to evaluate the safety depth of goaf under the impact of high-speed railway load. An optimal algorithm is selected among BP, RBF, and PSO-RBF neural networks based on the accuracy of the predicted height of a caving fracture zone. Numerical models for simulating high-speed railway founded on goaf are set up using the commercial software package FLAC3D, where factors such as subgrade height, train speed, and axle load are investigated in terms of train load disturbance depth and the extent of the caving fracture zone. The results indicate that the PSO-RBF neural network has an error of 2.76% in predicting the height of the caving fracture zone; the depth of train load disturbance is linearly related to the axle weight and roadbed height but is a sinusoidal function of the train speed. Based on the numerical simulation results, a formula for calculating the depth of train load disturbance is proposed, which provides a certain reference value for the construction of high-speed railways in the goaf ground.

Stability analysis and numerical simulation of foundation in old goaf under building load

“…The stability of a foundation can be quantitatively evaluated before a new building is built on the ground of an old goaf according to the spatial relationship between the depth of the load influence and the height of the caving fracture zone of coal seam (Xu et al, 2014;Liu et al, 2015;Zhang et al, 2019). Ren et al (2022) chose an optimal algorithm to calculate the influence depth of the train load and the development height of the caving fracture zone. The results showed that a 'particle swarm optimization' -'radial basis function' neural network model obtained the smallest error, and the prediction was the most accurate.…”