摘要
针对基于最小二乘支持向量机(LS-SVM)的瓦斯浓度预测方法进行参数优化时存在的易陷入局部最优解、搜索效率较低、易产生早熟收敛等问题,提出了一种基于改进蚁群算法-最小二乘支持向量机(ACO-LSSVM)的瓦斯浓度预测模型。首先,对采集的大量煤矿综采工作面瓦斯数据进行k-means聚类分析,以降低数据维数;然后,采用改进蚁群算法对LS-SVM的惩罚参数和核函数参数进行寻优,再代入LS-SVM模型中进行回归预测。仿真结果表明,当瓦斯体积分数绝对误差阈值分别为0.03%,0.04%,0.05%时,基于ACO-LS-SVM的瓦斯浓度预测模型的预测准确度都在95%左右,比SVM模型和LS-SVM模型表现更好。
In view of problems of gas concentration prediction method based on least squares support vector machine(LS-SVM)such as easy to fall into local optimal solution,low search efficiency and easy to occur premature convergence during parameter optimization process,a gas concentration prediction model based on ACO-LS-SVM was proposed.Firstly,k-means clustering analysis is performed on collected large amount of gas data on fully mechanized coal mining face to reduce dimension.Then,improved ant colony algorithm is used to optimize penalty parameters and kernel function parameters of LS-SVM,and the optimized parameters are substituted into the LS-SVM model for regression prediction.The simulation results show that when absolute error threshold of gas concentration is 0.03%,0.04%,0.05%,the prediction accuracy of the gas concentration prediction model based on ACO-LS-SVM is about 95%,which is better than SVM model and LS-SVM model.
作者
李欢
贾佳
杨秀宇
宋春儒
LI Huan;JIA Jia;YANG Xiuyu;SONG Chunru(School of Information and Control Engineering,China University of Mining and Technology, Xuzhou 221008,China;Wangjialing branch,Shanxi China Coal Huajin Energy Co.,Ltd., Yuncheng 043000,China;Tongmei Guodian Tongxin Coal Mine,Datong 037000,China)
出处
《工矿自动化》
北大核心
2018年第12期48-53,共6页
Journal Of Mine Automation
基金
国家重点研发计划项目(2016YFC0801808)