基于动态神经网络的瓦斯浓度实时预测方法  

Forecasting Coalmine Gas Concentration Based on Dynamic Neural Network

作　　者：张昭昭[1] 乔俊飞[2] 余文 

机构地区：[1]辽宁工程技术大学电子与信息工程学院,葫芦岛125105 [2]北京工业大学电子信息与控制工程学院,北京100124 [3]墨西哥国立理工大学高级研究中心(CINVESTAV-IPN),mexicodf墨西哥07360

出　　处：《控制工程》2016年第4期478-483,共6页Control Engineering of China

基　　金：国家自然科学基金(61440059);辽宁省教育厅科学研究一般项目(2013129);国家留学基金委资助

摘　　要：针对矿井中瓦斯浓度变化的时变性、非线性等特点,提出了一种动态神经网络瓦斯浓度实时预测模型。该模型利用历史数据建立初步预测模型,通过实时采集的瓦斯浓度数据进行预测,并用新数据及时调整预测模型的学习参数和结构参数,使得预测模型能够根据瓦斯浓度的动力学特性及时更新。用矿井实测瓦斯浓度数据进行试验,结果表明该模型较其他静态预测模型的预测精度有明显的提高。A dynamic neural network was presented to forecasting coalmine gas concentration in the view of the coal mine characteristics such as time-varying and nonlinear etc. The preliminary forecast model was established by using historical data. The prediction result can be achieved by the real-time acquisition gas concentration data, and the new data was also used to adjust the learning parameters and structure parameters of the predictive model, this method helps to update the parameters of the prediction model according to the dynamic characteristics of the real gas concentration in time. Experiment with the actual mine gas concentration data and the results shows that this model has higher prediction accuracy compared with other static model.

关 键 词：动态神经网络 瓦斯浓度预测 时变系统 实时 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]