灰色-进化神经网络模型在深埋隧道围岩变形预测中的应用

由于深埋隧道围岩的变形受到构造、应力场、地下水、开挖方式等复杂因素的综合影响，具有位移时间序列的单调增长的特殊性和非线性，运用响应成分模型将隧道围岩位移量分解成具有确定性的趋势项和具有不确定性的随机项。建立灰色–进化神经网络模型对趋势项和随机项进行预测，既结合GM（1,1）模型较好预测序列增长趋势的特点，又结合神经网络利用自适应实现网络总体误差最小的特点，进而解决了单一利用GM（1,1）模型时预测值的随机偏离量较大的问题，保证了预测的精度。将该模型应用于基于实测位移资料的堡镇隧道围岩水平收敛位移短期预测，较好地揭示了隧道围岩收敛位移演化的规律，为合理选取二次衬砌时机提供了参考。

Application of Grey-ENN model to prediction of wall-rock deformation in deep buried tunnels

With the comprehensive effect of structure, stress distribution, groundwater, opening mode, etc., displacement time series of the deep buried tunnels’ wall-rock display monotone increasing characteristics and nonlinearity. Based on the responding composition model, the displacement of tunnels’ wall-rock is decomposed into trend item and random item. The trend and random items of displacement time series are predicted by Grey-ENN model, which combines the advantages of GM(1,1) and ANN model. GM(1,1) can forecast the trend of increasing sequence, and ANN model can reduce error by self-adapting of network. Thus the correctness of prediction is assured due to avoiding the problem that forecast result is far away from real one by using GM(1,1) only. The Grey-ENN model is applied to predict the short-term convergence displacement of wall-rock in Baozhen Tunnel based on the in-situ displacement data. The reasonable convergence rule of tunnels wall-rock is achieved, which provides eference for choosing opportunity to do the second support.