基于贝叶斯估计法的矿区地表沉降预测组合模型

A combined model for prediction of mine surface subsidence based on Bayesian estimation method

针对单一模型在矿山地表沉降预测中的缺陷,提出了一种基于贝叶斯估计法的沉降预测组合模型。以甘肃金昌西二采区为例,以Sentinel-1A卫星SAR影像为数据源,分别用差分整合移动平均自回归(ARIMA)模型和径向基函数神经网络模型进行单一模型预测,另外基于残差修正法和贝叶斯估计法进行组合模型预测,结果表明:ARIMA模型、径向基函数神经网络模型、残差修正法组合模型、贝叶斯估计法组合模型的平均绝对误差分别为14.067、9.427、7.415、3.326 mm,平均相对误差分别为11.40%、7.39%、6.14%、2.65%,均方根误差分别为15.878、12.097、11.335、6.254 mm;组合模型预测精度较单一模型有明显的提升,其中基于贝叶斯估计法的组合模型性能最佳,可用于中长期的矿山地表沉降预测。

To remedy the defects in prediction of mine surface subsidence using a single model this paper presented a combined prediction model based on Bayesian estimation method.Taking the mine area No.2 in west Jinchang Mine as an example,with Sentinel-1 A satellite SAR images as data source,the single model-based prediction of mine surface subsidence was performed using the autoregressive integrated moving average(ARIMA) model and the radial basis neural network model separately.Furthermore,the prediction with combined model was carried out based on residual correction method and Bayesian estimation method.The results show that the average absolute errors of prediction using ARIMA model,RADIAL basis neural network model,residual correction method combination model and Bayesian estimation method-based model were 14.067 mm,9.427 mm,7.415 mm and 3.326 mm respectively,while the average relative errors were 11.40%,7.39%,6.14% and 2.65% respectively,while the root mean square errors were 15.878,12.097,11.335 and 6.254 mm respectively.Compared with the single model,the prediction accuracy using the combined model can be significantly improved.Particularly the combined model based on Bayesian estimation method presented the best performance,which is proposed to be used for the long-term prediction of mine surface subsidence.