Analysis on Ability of Artificial Neural Network to Assess Mine Safety

刘海波,施式亮,刘宝琛

DOI: https://doi.org/10.3969/j.issn.1009-6094.2004.05.017

2004-01-01

Applied Mechanics and Materials

Abstract:In this paper, some characteristic of the network has been carefully studied by analyzing the difference among the simulation results that come from the different training examples or different network constructure when the artificial neural network technology is used to assess the mine safety. In order to explore the optimum design of the artificial neural network that is used to assess the safety of the mine, the same example is used to train the neural network with the different number of the neurons and with different initiation ways of the weights and the biases. In order to analyze the practicality of the neural network to assess the mine safety, the reflection characteristic of the neural network has been studied by training network with different examples that are created and added to by random sampling the data of the original example, or with the example which data have been standardized. By means of these detailed studies, some methods to determine the parameter have been drawn out, i.e. the optimum number of the example data, the optimum constructure of the neurons, the optimum number of the layers, the choice of the transfer functions and the optimum initiation ways of the weights and the biases. By way of analyzing the essential distinction between the mine safety assessment and the artificial neural network technology, some suggestion has been also given to improve the artificial neural network technology when it's used to assess the safety of the mine. These methods and suggestion can greatly simplify the design course of the network constructure that is used to assess the mine safety, and can increase the accuracy and the practicality of the artificial neural network technology in the assessment of the mine safety.