洞室围岩位移长期预报混沌-神经网络模型

The Chaotic-Neural Network Model on Long-Term Prediction of Cavern Rock Mass Displacement

为实现开挖结束后大型地下洞室围岩位移的长期预报,及时评价围岩长期稳定,结合位移混沌力学参数优化BP神经网络结构,建立混沌-动态时间延滞神经网络长期预报模型。将嵌入维数m作为神经网络的输入层个数,增加神经网络预报反馈模式,动态生成预报训练样本,选取较大的时间延迟τ,预测步数为h,使相点间的时间延迟为hτ,通过有限预测步数,实现位移长期预报。实例表明,模型计算速度快,计算稳定性好。当预测步数h≤5,预测次数不大于10次时,预报精度在10%以内,预报结果实时有效,实现了大型地下洞室位移的长期预报,为大型地下洞室围岩稳定性评价提供了快速有效的新思路。

To realize long-term prediction of surrounding rock displacement of large underground grave after its construction,and analyze long-term stability of surrounding rock in-time,the structure of BP neural network is optimized based on chaotic-dynamic parameters of displacement,and the Chaos-dynamic time delay BP neural network model is built,then the long-term forecasting are enabled.Embedding dimension m is set as the number of input layer,and a predicting feedback mode of neural network is added,and prediction training samples are generated dynamically.The bigger delay time τ is selected,and the forecasting step is h,then the time delay between adjacent phase points is hτ,so the long-time prediction is achieved by limited number of forecasting steps.The instances show that computational stability of built prediction model is preferable with faster calculating speed,and prediction precision is all within 10% when predicting step h is no more than 5,and the number of forecast is no more than 10.Therefore,the forecasting results are of real time and effective,and the built model can provide a fast and available new thought for evaluating stability of surrounding rock mass of large underground cavern.