CAJ | 학술논문

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。
준학예측지철차참개알과정중적지표침강이성위성시지하공정풍험공제적중점화난점。침대전통시간서렬예측모형예측시적단일선성화홀략시공인소영향적정태국한성，제출료대외부수입적비선성자회귀신경망락(NARXNN)시간서렬예측모형。해모형본신구유연지단원화반궤결구，차통과인입시공영향인자작위외부수입적일부분，가이비선성동태지고필지철차참시공과정。운용 NARXNN 시간서렬예측모형대북경지철륙호선북해북참개알과정적지표침강진행예측，결과표명：(1)여전통적 ARMA 시간서렬예측모형상비，NARXNN 시간서렬예측모형괄응성경호、준학성경고；(2)통과인입시공영향인자，NARXNN시간서렬예측모형능구준학예측침강력시곡선돌변점처적변화추세；(3)가이통과인입다조시공영향인자혹우화시공영향인자적취치방법래진일보제고NARXNN시간서렬예측모형적예측정도。
How to precisely predict the ground subsidence induced by metro station construction becomes the key problem in urban underground engineering. A NARXNN time series prediction model is proposed due to the single linearity of traditional time series prediction model and its static limitation caused by ignorance of construction factor. In order to take the process of metro station construction into consideration nonlinearly and dynamically, construction impact factors,as a part of external inputs,are applied in this model that itself has delay unit and feedback architecture. Based on the NARXNN time series prediction model,the prediction results of ground subsidence induced by the Beihaibei station of Beijing metro line 6 construction show that:(1) Compared with traditional ARMA time series prediction model,NARXNN time series prediction model has a better adaptability and precision. (2) The NARXNN time series prediction model has a precise trend forecast at breakpoints of the settlement-time curve. (3) Multiple construction impact factors or subdividing each group of construction impact factors can be used to improve forecasting precision in using NARXNN time series prediction model.