摘要
渗流监测是掌握大坝安全性态的重要手段,针对土石坝渗流压力存在滞后于库水位的特点,引入具有延时输入特性的带外源输入的非线性自回归神经网络(Nonlinear Auto-regressive with Exogenous inputs neural network,NARX)实现土石坝渗压的有效预测。以某一水库大坝为例,将历史某时段的库水位和降雨等影响因子作为输入序列,渗压测值作为输出序列,分别建立NARX网络多因子和单因子模型进行拟合训练和多步预测,并将预测结果与传统回归模型和传统BP神经网络进行对比。研究结果表明,在RMSE、MAE、MAPE 3种精度指标下,NARX模型均优于2种传统模型。其中,在单因子条件下,NARX仍具有良好的表现。NARX的延迟输入特性可在一定程度上模拟坝体水流渗透的滞后性,对于土石坝的渗压预测具有良好的应用效果。
Seepage monitoring is an important means to master the safety state of a dam.As the seepage pressure of earth-rockfill dams lags behind the reservoir water level,a nonlinear auto-regressive neural network with exogenous inputs(NARX),a network characterized by delayed input,is introduced to effectively predict the seepage pressure of the dams.In the case of a reservoir dam,with the influencing factors such as the reservoir water level and rainfall in a certain period of history as the input sequence and the measured value of seepage pressure as the output sequence,the multi-factor and single-factor models of the NARX network are built separately for fitting training and multi-step prediction.Then,the prediction results are compared with those of the traditional regression model and the traditional BP neural network.The results reveal that the NARX model outperforms the two traditional models under the three accuracy indexes of RMSE,MAE,and MAPE.Moreover,the NARX model still has good performance under the condition of a single factor.The delayed input characteristic of the NARX network can simulate the hysteresis of dam seepage to a certain extent,and the network has a good application effect for seepage pressure prediction of earth-rockfill dams.
作者
赵普
谷艳昌
吴云星
ZHAO Pu;GU Yanchang;WU Yunxing(Dam Safety Management Department,Nanjing Hydraulic Research Institute,Nanjing 210029,China;Dam Safety Management Center of the Ministry of Water Resources,Nanjing 210029,China)
出处
《人民珠江》
2022年第7期126-130,共5页
Pearl River
基金
国家自然科学基金(51979175)。
关键词
NARX
渗压预测
滞后性
NARX
seepage pressure prediction
hysteresis
