Critical hydraulic pressure forecasting of water inrush in coal seam floors based on a genetic algorithm-neural network

M. Zhong, T. Fu, C. Shi, and T. Liu
Abstract -------------------------------------------------------------------------------------------Click the "Download this article" link to view technical paper in PDF format.-------------------------------------------------------------------------------------------
In recent years, the risk of water inrush in coal seam floors has increased due to an increase of the depth that coal is mined in China. Water inrush in coal mines is brought on by an unbalance in the groundwater flow and engineering geomechanics fields. This occurrence is influenced by factors such as rock characteristics of the base plate, destructiveness range of the fracture zone, relative magnitude of ground stress, hydraulic pressure of the confined aquifer, etc. However, it is difficult to quantitatively describe the relationships between these factors and water inrush.  As a result, the issue has not yet been resolved either in theory or in practice.
Experiments show that water inrush in coal seam floors only occurs when the hydraulic pressure is higher than the minimum principal stress of the floor. Therefore, to prevent water inrush, it is important to predict the critical hydraulic pressure of water inrush in coal seam floors. The forecasting process is used to establish the expression of the critical hydraulic pressure in certain conditions, based on the known actual measurement data of some China coal mines, and then to predict the critical hydraulic pressure of water inrush in other conditions or for other coal mines. Due to the non-linear relationship of multiple factors, it is difficult to find the suitable function or equation to describe this forecasting.
This paper will investigate a method for conducting theoretical forecasting of the critical hydraulic pressure of water inrush in a coal seam floor using a combination method of a genetic algorithm-neural network. A neural network is adapted to treat the non-linear variables and may be chosen to analyze the relationship between the critical pressure of water inrush and different conditions in coal seam floors. However, restricted by the local search capacity and the slow convergence rate, general analysis results adopting a neural network are prone to reach the local extremum of a neural network and, therefore, are not ideal. Hence, in this paper, genetic algorithms will be adopted to train neural networks and optimize the topology structure of a neural network. The utilization of genetic algorithms makes the training of neural networks more effective. There is no requirement of continuance or differentiability for the objective function. Its search is always throughout the entire solution space and, therefore, it is easy to gain the global optimum solution. The critical hydraulic pressure of water inrush in coal seam floors will be effectively predicted using the optimum structure of a neural network.
The actual measurement data are chosen from 74 working faces of the Shuangshan, Xiazhuan, Xihe, Longquan, Shigu, Pucun and Hongshan coal mines. Initially, the topology structure of the neural network is optimized using genetic algorithms. The actual measurement data of 60 working faces act as the learning sample of the neural network. The search space of the genetic algorithm has one to ten hidden layers and each one has one to 30 nodes. According to the characteristics of the sample, the number of nodes of the neural network’s input layer is set as eight, and that of the output layer is set as one. The training and learning are conducted for the samples. When the training error is 0.000315, the optimum topology structure of the neural network is achieved through the calculations of the genetic algorithm, that is, the number of hidden layers is one and the number of nodes of the hidden layer is 20. The verification for these samples is conducted using neural networks with the same structure. Comparisons between the predicted results and actual measured results proved that the proposed method in this paper is feasible to forecast the critical hydraulic pressure of water inrush in coal seam floors.
In addition, during the selection of the topology structure of the neural network, there are some factors to consider that influence the performance of the genetic algorithm, including the population size, mutation rate, crossing rate, optimum strategy, etc. Therefore, the system performance based on these factors is also discussed.
Keywords: Water inrush, Coal seam floor, Critical hydraulic pressure, Neural network, Genetic algorithms
Full Access to Technical Paper
PDF version for $20.00
Other papers from
Search