基于RF-GWO-LSSVM的煤矿地表下沉系数预测

Prediction of surface subsidence coefficientin mining zrea based on RF-GWO-LSSVM

为提高煤矿开采地表下沉系数预测精度,将随机森林(RF)、灰狼算法(GWO)和最小二乘支持向量机(LSSVM)模型相结合,建立RF-GWO-LSSVM模型。利用RF算法计算每个特征的重要性,通过特征选择筛选出重要性高的特征作为特征子集。LSSVM模型在处理小样本,非线性数据方面具有很大的优势,但LSSVM模型泛化能力非常容易受到内部参数的影响,采用GWO算法寻求最优的惩罚因子c和核函数参数σ。将优化后的模型对地表下沉系数进行预测,并与GWO-LSSVM模型、PSO-LSSVM模型精度对比。结果表明:RF-GWO-LSSVM模型预测精度最高,预测结果决定系数为0.996,可为预测地表下沉系数研究提供一定的参考价值。

In order to improve the prediction accuracy of surface subsidence coefficient in coal mining,the random forest(RF)-grey wolf optimization(GWO)-least squares support vector machine(LSSVM)model was established by combining random forest,gray wolf algorithm and least squares support vector machine model.The importance of each feature was calculated by the RF algorithm,and the features with high importance were selected as feature subsets through feature selection.The LSSVM model has great advantages in dealing with small samples and nonlinear data,but the generalization ability of the LSSVM model is very easily affected by internal parameters.The GWO algorithm was applied to find the optimal penalty factor and kernel function parameters.The optimized model was introduced to predict the surface subsidence coefficient,and the accuracy was compared with the GWO-LSSVM model and the PSO-LSSVM model.The results showed that the prediction accuracy of the RF-GWO-LSSVM model was the highest,and the determination coefficient of the prediction results was 0.996,which could provide a certain reference value for the prediction of the surface subsidence coefficient.