采用人工神经网络对矿山进行安全评价时,总是希望它对相同的或稍有改变的原始仿真数据产生比较稳定的仿真输出,否则,安全评价就会容易变成胡乱评价。为了得到比较稳定的安全评价结果,必须找到足够多的原始样本数据来训练神经网络,而找到多达数百组的经验数据是一件非常困难的工作。 When using artificial neural network to evaluate the safety of a mine, it is always hoped that it will produce a more stable simulation output for the same or slightly changed original simulation data. Otherwise, the safety assessment will easily become a random evaluation. In order to obtain a stable safety evaluation result, enough raw sample data must be found to train the neural network. Finding up to several hundred groups of empirical data is a very difficult task.