电子世界
電子世界
전자세계
ELECTRONICS WORLD
2014年
11期
76-77,78
,共3页
冲击地压%混沌时间序列%多源%支持向量机%特征聚类
遲擊地壓%混沌時間序列%多源%支持嚮量機%特徵聚類
충격지압%혼돈시간서렬%다원%지지향량궤%특정취류
Rockburst%chaos time series%multi-source%ensemble learning%feature clustering
考虑到单一变量描述系统信息的局限性，利用多源混沌监测时间序列来评价和预测冲击危险性，提高冲击地压预测精度。由于冲击地压监测时间序列具有混沌特性，经重构得到的状态向量能更好反映开采过程中煤岩体结构变化，本文以此作为预测模型的输入变量。考虑到状态向量的特征间存在冗余性和互补性，提出了基于特征聚类的随机子空间集成学习算法，并详细说明了该算法中特征聚类、特征子集的形成、个体学习器的训练等环节的实现。最后以25110工作面地压监控数据为例进行了验证。结果表明相对于使用单一监测变量作为输入的预测方法，本文方法精度明显提高。
攷慮到單一變量描述繫統信息的跼限性，利用多源混沌鑑測時間序列來評價和預測遲擊危險性，提高遲擊地壓預測精度。由于遲擊地壓鑑測時間序列具有混沌特性，經重構得到的狀態嚮量能更好反映開採過程中煤巖體結構變化，本文以此作為預測模型的輸入變量。攷慮到狀態嚮量的特徵間存在冗餘性和互補性，提齣瞭基于特徵聚類的隨機子空間集成學習算法，併詳細說明瞭該算法中特徵聚類、特徵子集的形成、箇體學習器的訓練等環節的實現。最後以25110工作麵地壓鑑控數據為例進行瞭驗證。結果錶明相對于使用單一鑑測變量作為輸入的預測方法，本文方法精度明顯提高。
고필도단일변량묘술계통신식적국한성，이용다원혼돈감측시간서렬래평개화예측충격위험성，제고충격지압예측정도。유우충격지압감측시간서렬구유혼돈특성，경중구득도적상태향량능경호반영개채과정중매암체결구변화，본문이차작위예측모형적수입변량。고필도상태향량적특정간존재용여성화호보성，제출료기우특정취류적수궤자공간집성학습산법，병상세설명료해산법중특정취류、특정자집적형성、개체학습기적훈련등배절적실현。최후이25110공작면지압감공수거위례진행료험증。결과표명상대우사용단일감측변량작위수입적예측방법，본문방법정도명현제고。
Al owing for the limitations of single variable to describe system,rockburst risk is predicted on multi-source monitoring chaos time-series to improve the prediction precision.Because the time-series to monitor rockburst have chaotic characteristics and their reconstruction state vectors can bet er reflect the structure change of coal and rock in mining process,the paper uses the state vectors as input variables of prediction model and gives the principle to identify rockburst risk.Considering the redundancy and complementarity among the features of the state vector,the random subspace ensemble learning algorithm on feature clustering is presented,and the algorithm’s details including feature clustering,feature subspace structuring and individual learners training are explained.Final y,the proposed method is applied to 25110 mining face and the results show that our method can significantly improve prediction precision,relative to the method of taking multi-source monitoring variables directly as prediction model inputs.
