摘要
为了提高可燃气体异常检测的预测精度,针对回声状态网络中权值的随机初始化会造成实际值和预测值之间误差较大的问题,提出一种改进回声状态网络的权值初始化方法.首先利用Xavier初始化方法使得各层激活值的方差、状态梯度的方差分别在传播过程中保持一致,从而确保网络中信息更好地流动,然后采用改进回声状态网络对时间序列数据进行学习,建立预测模型,最后分别在三类不同的数据集上对模型性能进行仿真测试.结果表明,与原回声状态网络相比,改进后模型的均方误差、归一化均方根误差和平均绝对百分比误差明显降低,可以对可燃气体信息进行更准确的预测,根据预测值和实际值的残差实现对可燃气体的异常检测,具有更好的应用价值.
In order to improve the prediction accuracy of combustible gas anomaly detection,aiming at the problem that the random initialization of weight in echo state network will cause large error between actual value and predicted value,an improved weight initialization method of echo state network is proposed.Firstly,Xavier initialization method is used to make the variance of activation value and state gradient of each layer consistent in the propagation process,so as to ensure the better flow of information in the network.Then,the improved echo state network is used to learn the time series data and establish the prediction model.Finally,the performance of the model is simulated and tested on three different data sets.The results show that compared with the original echo state network,the Mean Square Error,Normalized Root Mean Square Error and Mean Absolute Percentage Error of the improved model are significantly reduced,which can predict the combustible gas information more accurately,and realize the anomaly detection of combustible gas according to the residual between the predicted value and the actual value,which has better application value.
作者
赵月爱
郗林栋
ZHAO Yueai;XI Lindong(Department of Computer Scienc,Taiyuan Normal University,Jinzhong 030619,China;Department of Mathematics,Taiyuan Normal University,Jinzhong 030619,China)
出处
《太原师范学院学报（自然科学版）》
2022年第1期26-33,共8页
Journal of Taiyuan Normal University:Natural Science Edition
基金
国家社科基金项目(20BJL080)
山西省重点研发计划项目(201803D121088)。
关键词
可燃气体异常检测
回声状态网络
权值初始化
combustible gas anomaly detection
echo state network
weight initialization
