人工神经网络对矿山安全状态的评判能力分析

通过改变神经网络训练样本等方法,对比分析了神经网络对不同训练样本的反映能力,讨论了人工神经网络对矿山安全程度进行评价的适应性.为了研究人工神经网络用于矿山安全评价时的优化设计,通过改变神经网络的神经元数目及初值赋值方式等方法,测试了不同结构、不同参数的神经网络对相同训练样本的评价结论.本文的研究为人工神经网络用于矿山安全评价时的进一步改进及其优化设计提出了合理的建议.

Analysis on ability of artificial neural network to assess mine safety

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.