基于Adam算法BP神经网络的矿井环境瓦斯浓度预测模型

On Mine Environmental Gas Concentration Prediction Model Based on Adam Algorithm BP Neural Network

为了解决现有矿井环境瓦斯浓度预测方法无法处理大数据量、适应性差、误差较大、易陷入局部最优等问题,提出一种基于Adam算法的改进型BP(Back Propagation)神经网络模型,模型适用于矿井多环境参数下,对某区域内环境瓦斯浓度进行预测.对监测监控系统采集到的真实数据进行归一化处理并形成数据集,通过将Adam算法与BP网络模型进行有效结合形成新的网络模型.运用训练集对模型进行训练及调优后,迭代次数在1 200次后损失率趋于平稳,验证集预测的结果整体平均误差率为1.258%,结果表明:该优化模型提高了网络训练速度,且避免了传统BP模型容易陷入局部最小的缺点,同时降低了预测的相对误差.

In order to solve the problems of existing mine environment gas concentration prediction methods that cannothandle large data volume,poor adaptability,large errors,and easy to fall into local optimality,an improved BP(Back Propagation)neural network model based on Adam algorithm is proposed.The model is suitable for multiple environments in mines.Under the parameters,it predicts the environmental gas concentration in a certain area.The real data collected by the monitoring system is normalized and formed into a data set,and a new network model is formed by effectively combining the Adam algorithm with the BP network model.After training and tuning the model with the training set,the loss rate stabilized after 1200 iterations,and the overall average error rate of the verification set prediction is 1.258%.The results show that the optimized model has improved the network training speed and avoided the shortcoming that the traditional BP model is easy to fall into local minimum,and meantime it reduces the relative error of prediction.