基于STL-EEMD-GA-SVR的采煤工作面瓦斯涌出量预测方法及应用

瓦斯涌出量准确预测可为矿井通风及瓦斯灾害防治措施提供重要依据。为提高采煤工作面瓦斯涌出量预测精度,根据陕西黄陵某矿采煤工作面绝对瓦斯涌出量监测数据,应用基于局部加权回归的周期趋势分解(Seasonal-Trend decomposition procedure based on Loess, STL),将监测数据分解成趋势项、周期项和不规则波动项;利用集成经验模态分解(Ensemble Empirical Mode Decomposition,EEMD),将不规则波动项分解得到不同特征尺度的IMFs(Intrinsic Mode Functions, IMFs)分量以及残差余量;通过遗传算法(Genetic Algorithms, GA)参数寻优后的支持向量回归机(Support Vector Regression, SVR),对各项分解数据进行预测;叠加各分量模型预测结果,得到最终瓦斯涌出量预测结果。结果表明：在预测集为247、147和70组3种情景下,对比分析了STL-EEMD-GA-SVR模型(简称SEGS)、EEMD-GA-SVR模型、GA-SVR模型和高斯过程回归(Gauss...

Prediction method and application of gas emission from mining workface based on STL-EEMD-GA-SVR

Accurate prediction of gas emission can provide important basis for mine ventilation and the prevention and measures of gas disasters. In order to improve the prediction accuracy of gas emission in the mining workface, the monitoring data of gas emission were decomposed into the trend term, periodic term and irregular fluctuation term by the Seasonal-Trend decomposition procedure based on Loess (STL) based on the monitoring data of gas emission from the mining workface of Huangling Mine in Shaanxi. Besides, the irregular fluctuation term was further broken down into the Intrinsic Mode Functions (IMFs) components with different characteristics and the residual margins by the Ensemble Empirical Mode Decomposition (EEMD). Then, each decomposed data was predicted by the Support Vector Regression (SVR) through parameter optimization by Genetic Algorithms (GA). Moreover, the prediction result of each component model was superposed to obtain the final prediction result of gas emission. In addition, the evaluation indicators for precision of STL-EEMD-GA-SVR model (hereinafter referred to as SEGS), EEMD-GA-SVR model, GA-SVR model and Gaussian Process Regression (GPR) model were analyzed comparatively in the 3 scenarios with 247, 147 and 70 groups of prediction set. According to the results, SEGS model is the best, of which the fitting degree R2 was 0.81, 0.92 and 0.99 respectively, and the average relative error at the peak point was 3.15%, 2.33% and 1.04% respectively. In general, the constructed SEGS model could accurately predict the gas emission of mining workface.