论文部分内容阅读
准确预测地铁车站开挖过程中的地表沉降已成为城市地下工程风险控制的重点和难点。针对传统时间序列预测模型预测时的单一线性和忽略施工因素影响的静态局限性,提出了带外部输入的非线性自回归神经网络(NARXNN)时间序列预测模型。该模型本身具有延迟单元和反馈结构,且通过引入施工影响因子作为外部输入的一部分,可以非线性动态地考虑地铁车站施工过程。运用NARXNN时间序列预测模型对北京地铁六号线北海北站开挖过程的地表沉降进行预测,结果表明:(1)与传统的ARMA时间序列预测模型相比,NARXNN时间序列预测模型适应性更好、准确性更高;(2)通过引入施工影响因子,NARXNN时间序列预测模型能够准确预测沉降历时曲线突变点处的变化趋势;(3)可以通过引入多组施工影响因子或优化施工影响因子的取值方法来进一步提高NARXNN时间序列预测模型的预测精度。
Accurately forecasting the surface subsidence in the subway station excavation has become the key and difficult point in the risk control of urban underground engineering. Aiming at the single linearity of traditional time series forecasting model and ignoring the static limitations of construction factors, a nonlinear time series forecasting model with NARXNN input is proposed. The model itself has a delay element and a feedback structure. By introducing the construction impact factor as part of the external input, the subway station construction process can be nonlinearly and dynamically considered. The NARXNN time series forecasting model is used to predict the ground subsidence during the excavation process of Beihai North Station of Beijing Subway Line 6. The results show that (1) NARXNN time series forecasting model is more adaptive than the traditional ARMA time series forecasting model (2) Through the introduction of construction impact factors, NARXNN time series prediction model can accurately predict the trend of change at the sudden change point of the settlement duration curve; (3) By introducing multiple sets of construction impact factors or optimizing construction impact factors Value method to further improve the prediction accuracy of NARXNN time series prediction model.