基于MADF-DE-GRU的瓦斯浓度动态预测  

Dynamic prediction of gas concentration based on MADF-DE-GRU

作　　者：赵瑞晓 符学龙[1] 刘欢 ZHAO Ruixiao;FU Xuelong;LIU Huan(Intelligent Manufacturing Training Center,Jiangsu Polytechnic of Finance and Economics,Huaian 223003,Jiangsu,China)

机构地区：[1]江苏财经职业技术学院智能制造实训中心,江苏淮安223003

出　　处：《智能计算机与应用》2024年第8期115-120,共6页Intelligent Computer and Applications

基　　金：江苏省职业教育“双师型”教师团队培育项目。

摘　　要：为保障煤矿回采工作面安全生产,提高瓦斯浓度预测精度,本文提出一种基于门控循环单元(GRU)神经网络的瓦斯浓度动态预测模型。首先,用差分法改进的移动平均滤波器(MAF)去除原始数据中的噪声和趋势;其次,利用差分进化算法(DE)对GRU神经网络的隐藏层数、隐藏层神经元数量、时间步长和迭代次数等超参数进行寻优;最后,通过优化后的超参数搭建瓦斯浓度预测模型(MADF-DE-GRU)。通过仿真实验,并与多种预测模型进行对比,结果表明MADF-DE-GRU模型具有较高的预测精度,预测效果较好,能够应用到实际生产中。In order to ensure the safe production of coal mine back face and improve the accuracy of gas concentration prediction,a dynamic prediction model of gas concentration based on gated recurrent unit(GRU)neural network is proposed.Firstly,a moving average filter(MAF)improved by difference method is used to remove the noise and trend in the original data;then the hyperparameters such as the number of hidden layers,the number of neurons in the hidden layer,the time step and the number of iterations of the GRU neural network are used to find the optimal ones by using the differential evolution(DE)algorithm;and finally,a gas concentration prediction model(MADF-DE-GRU)is constructed by optimized hyperparameters,and the experimental results comparing with many kinds of prediction models,the MADF-DE-GRU model has higher prediction accuracy and better prediction effect,and can be applied to practical production.

关 键 词：瓦斯浓度 移动平均滤波器 门控循环单元 差分进化算法 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]