摘要
针对大坝变形监测中存在的大量小样本时间序列所具有的强非线性特性,引入组合建模的思想,综合应用ARIMA时间序列模型和BP神经网络模型实现了小样本大坝变形监测数据序列的分析,即先利用ARIMA时间序列模型对大坝变形监测数据进行拟合和预测,然后依据时间序列残差建立BP神经网络模型对残差进行预测,最后将两者结合以获得大坝变形的预测。实例分析表明,ARIMA-BP组合模型较单一模型的预测精度高,预测值更接近实测值。
The dam deformation monitoring data exists in large number of small sample time series and it has strong nonlinear character. So, a combination model was introduced based on the ARIMA time series model and BP neural net- work. The paper studied the application of two above models to analyze the small sample of dam deformation monitoring data. Firstly, the ARIMA model is used to fit and forecast dam deformation monitoring data. And then BP neural net- work is built with the ARIMA residuals for prediction of the residuals of dam deformation. Finally, combination of two models is applied to predict dam deformation. Compared with the single model, the example results show that the ARI- MA-BP model has obvious advantages in prediction accuracy and the predictive values are closer to the measured values.
出处
《水电能源科学》
北大核心
2015年第6期72-75,共4页
Water Resources and Power