基于D-ELM的矿井气体FTIR光谱定量分析  

Application of D-ELM in quantitative analysis of FTIR spectrum of mine gas

作　　者：沈永[1] 郭天太[1] 孔明[1] 赵军[1] 沈海栋 Shen Yong Guo Tiantai Kong Ming Zhao Jun Shen Haidong(Department of Metrological Technology and Engineering, China Jiliang University, Hangzhou 310018, China)

机构地区：[1]中国计量学院计量测试工程学院,浙江省杭州市310018

出　　处：《应用光学》2016年第5期725-729,共5页Journal of Applied Optics

基　　金：国家自然科学基金(51476154)

摘　　要：为了对D-ELM的矿井气体定量分析效果进行比较和分析,建立了3种矿井气体定量分析模型,分别是SVM模型、极限学习机(ELM)模型和动态极限学习机(D-ELM)模型。由于每个模型每次预测结果在一定范围内变化,所以取每个模型运行10次预测结果平均均方根误差(ARMSE)、平均相关系数(AR)、平均模型运行时间(AT)来评价各模型对气体定量分析的性能。PSO-SVM模型的结果为:[0.054 2,0.998,200.38];ELM模型的结果为:[1.042 1,0.989 4,0.26];D-ELM模型结果为:[0.043 8,1,2.01]。综合预测精度和预测速度表明,DELM模型要优于另外2种模型。Three quantitative analysis models for mine gas were established, the support vector machine （SVM） model, the extreme learning machine （ELM） model and the dynamic extreme learning machine （D-ELM） model, to analysis and compare with the result of D-ELM. Since the results of each model were changed in a certain range, taking the average root mean square error （ARMSE）, the average correlation coefficient （AR） and the average model running time （AT）, all of them were the average values of 10 results of each model, as the standards to e- valuate the performance of the model in the quantitative analysis of the mine gas. The results of particle swarm optimization（PSO）-SVM model, ELM model and D-ELM model were as follow： [0.054 2,0. 998,200.38], [1. 042 1,0.989 4,0.26-], [0.043 8,1,2.01]. Considering the predicted accuracy and predicted speed, D-ELM is better than the other two models.

关 键 词：D-ELM 定量分析 红外光谱 PSO-SVM 矿井气体 

分 类 号：TD71[矿业工程—矿井通风与安全]