矿井突水水源识别的RS-LSSVM模型

为了对矿井突水水源进行准确、高效的判别,综合考虑水化学特征,选取Ca~(2+),Mg~(2+),K~++Na~+,HCO-3,SO2-4,Cl~-和总硬度7个指标的质量浓度(mg/L)作为矿井突水水源的最初判别指标。利用粗糙集(RS)理论的属性约简来筛选水化学特征指标,用以作为水源识别的核心判别指标,建立基于RS的矿井突水水源识别的最小二乘支持向量机(LSSVM)模型。选用约简处理后的13组煤矿数据对模型进行训练,再用训练好的模型对另外12组突水数据进行水源判别,并与未进行属性约简的LSSVM模型及Fisher判别分析法、随机森林方法进行对比。结果表明,利用属性约简方法可以很好地排除原始数据中的冗余信息干扰,因而能有效判别矿井突水水源,使矿井突水水源模型的误判率降低至0;而且指标约简过程可以降低LSSVM运算的复杂度,也能够提高判别效率。

RS-LSSVM model for identifying and determinating the mining water inrush origin

The given paper is aimed to identify and discriminate the mining water bursting inrush source accurately and efficiently.For the research purpose,we have combined two algorithms of rough set and the least square support vector machine so as to build up a mine water inrush identification method.In so doing,we have first of all adopted the continuous attributes discretization method based on the rough set theory to discrete the continuous data from the original samples of the mining water inrush.Then,we have obtained the information of the water inrush from the coal mine to create a decision-making table after discretization based on the RS theory,which can help to simplify and delete some redundant information to contribute to the final decision making and improve the quality of the samples chosen under the premise to keep the classification ability and data integrity unchanged.Afterwards,it would be possible to optimize the parameters of LSSVM by choosing the appropriate kernel function and by K-fold cross validation with the help of the influential factors of the water inrush from the coal seam floor after the RS attribute reduction as the input vectors and sample training due to the LSSVM reduction.And,finally,the model can be used to identify the mine water inrush source.Hence,a comparison of the prediction results can be done with the direct LSSVM,the Fisher discrimination analysis method,with the random forest method being exploited.The experimental results show that it is a nice way to use the method of attribute reduction to eliminate the interference of the redundant information of the original data,and to discriminate the source of mine water inrush promptly and efficiently.In such a case,the false rate of mine water inrush source model should be made to be equal to 0.And,simultaneously,it is also possible to reduce the LSSVM computational complexity by improving the index reduction process and the discrimination efficiency.