基于EMD-LSSVM的瓦斯浓度动态预测模型  

New gas concentration dynamic prediction model based on the EMD-LSSVM

作　　者：魏林[1] 白天亮[2] 付华[3] 尹玉萍[3] 

机构地区：[1]辽宁工程技术大学基础教学部,辽宁葫芦岛125105 [2]辽宁工程技术大学电子与信息工程学院,辽宁葫芦岛125105 [3]辽宁工程技术大学电气与控制工程学院,辽宁葫芦岛125105

出　　处：《安全与环境学报》2016年第2期119-123,共5页Journal of Safety and Environment

基　　金：国家自然科学基金项目(51274118);辽宁省大学生创新训练项目(201310147028)

摘　　要：为提高回采工作面瓦斯体积分数预测时效性,建立了EMDLSSVM的瓦斯体积分数动态预测模型;为能够快速有效地反映瓦斯体积分数当前状态,避免早期历史数据对模型预测的影响,采用复合窗口技术对瓦斯体积分数时间序列进行动态更新;为提高算法预测精度,先采用经验模态分解算法(EMD)对更新后的窗口数据进行分解得到高频项、低频项和趋势项,考虑到瓦斯体积分数变化受到诸多因素干扰导致预测难度较大,但由同类因素影响的瓦斯体积分数变化特征具有较高的相似性,利用聚类方法将瓦斯体积分数监测数据划分成性质相似的若干个模式类别,以减少各种随机因素对预测结果的影响,再利用最小二乘支持向量机(LSSVM)对高、低频项进行加权预测,用自回归(AR)模型对趋势项进行预测,最后进行组合预测。实例对比分析表明,该预测模型能够有效地预测瓦斯体积分数的变化趋势,减少了预测时间,预测精度也满足矿山安全工程实际要求。To improve the prediction efficiency of the accurate and reliable gas concentration,this paper is aimed at introducing a composite window EMD-LSSVM dynamic prediction model by analyzing the gas monitoring data in the heading faces. To achieve the purpose,we have adopted a genetic algorithm（ GA） to analyze the optimal window parameters through training the sampling data. In so doing,we have first of all used the empirical mode decomposition（ EMD） algorithm to build up a high frequency model,a low frequency model and a developmental trend model. And,secondly,we have succeeded in using the clustering algorithm to reduce the impact of the various random factors to some gas concentration monitoring data so as to be divided into a number of similar nature pattern categories for the gas concentration is to be subject to the interference of many factors,in spite of the fact that the gas concentration change is very complicated and difficult to get the gas concentration efficient prediction. In addition,the gas concentration change characteristics have to be subject to the interference of the similar factors,too. What is more,it would be possible for us to use the least squares to support the vector machine（ LSSVM） to predict the high and low frequency models with the necessary weights,especially with those on the time decay that can be better in accord with the current gas concentration situation,with the autoregressive model being oriented to forecast the trend model. And,last of all,the three forecast results are supposed to be joined together so as to gain the final prediction results. By comparing the results of our simulation with the experimental ones,it can be concluded that the new method can effectively forecast the gas concentration changing trend and reduce the time of prediction,with the prediction precision being able to meet the requirements of the mining safety engineering actuality.

关 键 词：安全工程 复合窗口 经验模态分解 最小二乘支持向量机 时间序列 

分 类 号：X936[环境科学与工程—安全科学]