摘要
针对大坝变形时间序列的非线性及形变值累计特性,引入NARX神经网络进行分析并实现变形预测。首先,NARX神经网络通过非线性自回归网络与外源输入相结合,较好地解决了传统BP神经网络存在的收敛速度慢和易陷入局部极值等问题;其次,建立基于NARX神经网络的大坝变形预测模型,对原始数据预处理后采用周期为输入序列、变形值为输出序列训练模型;最后,以官地水电站大坝监测序列为例验证NARX神经网络模型预测性能。结果表明,在MSE、MAPE及RMSE三项精度指标测算中,BP神经网络分别为5.10 mm^(2)、30%、3.31 mm,而NARX神经网络分别为0.78 mm^(2)、12%、2.21 mm,均小于BP神经网络的,说明了NARX神经网络模型具有更高的预测精度。此外,NARX神经网络预测模型收敛时间为0.36 s,收敛速度较BP神经网络有较大提升。
Aiming at the nonlinear and cumulative characteristics of dam deformation time series,NARX neural network was introduced to analyze and predict the deformation. First of all,NARX neural network combined the nonlinear autoregressive network with the external input,which solved the issues of the traditional BP neural network,such as slow convergence speed and falling into the local extreme value.Secondly,a dam deformation prediction model based on NARX neural network was established,which used the period as the input sequence and the deformation value as the output sequence training model after pretreatment. Finally,the Guandi Hydropower Station was taken as the training model and the dam monitoring sequence was taken as an example to verify the prediction performance of NARX neural network model. The results show that the BP neural network is 5.10 mm^(2),30% and 3.31 mm respectively in the calculation of MSE,MAPE and RMSE precision indexes,while the NARX neural network is 0.78 mm^(2),12% and 2.21 mm respectively,which are smaller than that of the BP neural network,indicating that NARX neural network model has higher prediction accuracy. In addition,the convergence time of the NARX neural network prediction model is 0.36 seconds,which is a significant improvement over the BP neural network.
作者
范哲南
刘小生
FAN Zhenan;LIU Xiaosheng(School ofArchitecture and Surveying Engineering,Jiangxi University of Science and Technology,Ganzhou 341000,China)
出处
《人民黄河》
CAS
北大核心
2022年第2期125-128,共4页
Yellow River
基金
国家自然科学基金资助项目(41561091)。
关键词
大坝变形
变形预测
神经网络
NARX
评价模型指标
dam deformation
deformation prediction
neural network
NARX
evaluation model index
