摘要
矿工不安全行为的出现是复杂的非线性动力过程,为预测不安全行为时间序列,选择具有"记忆"功能和解决梯度消失问题的长短期记忆网络。使用TensorFlow下Keras搭建基于长短期记忆网络的不安全行为时间序列预测模型,使用A、B煤矿2年共3405条不安全行为序列数据进行模型训练和测试,根据交叉验证集选择最优参数。实验结果表明:构建的4个时间序列预测模型最小的平均绝对误差为0.0807,最大的平均绝对误差为0.3335,能够很好预测煤矿未来一定时间段内的不安全行为。
The emergence of unsafe behavior of miners is a complex nonlinear dynamic process.In order to predict the time series of unsafe behavior,the long and short term memory with a"memory"function and a solution to the disappearance of gradients is selected.Keras under TensorFlow was used to build a time series prediction model of unsafe behavior based on long and short term memory.A total of 3405 time series data in coal mine A and B were used for model training and testing,and the optimal parameters were selected according to the cross validation set.The results showed that the minimum average absolute error of the four time series prediction models is 0.0807 and the maximum average absolute error is 0.3335 and those models can well predict unsafe behavior in a certain period of time in the coal mine.
作者
鞠春雷
聂方超
刘文岗
郭金山
张江石
JU Chunlei;NIE Fangchao;LIU Wengang;GUO Jinshan;ZHANG Jiangshi(Beijing Tiandi Huatai Mining Management Co.,Ltd.,Beijing 100013,China;School of Emergency Management and Safety Engineering,China University of Mining and Technology(Beijing),Beijing 100083,China)
出处
《煤矿安全》
CAS
北大核心
2020年第9期260-264,共5页
Safety in Coal Mines
基金
天地科技股份有限公司科技创新创业资金资助项目(2018-TD-MS058)。
关键词
不安全行为
时间序列
循环神经网络
长短期记忆网络
机器学习
unsafe behavior
time series
recurrent neural network
long and short term memory
machine learning
