摘要
采用标准神经网络(BP)构建的大坝安全监控预报模型难以满足工程精度要求。利用滤波理论(Unscented Kalman filter,UKF),引入自适应因子,并对神经网络进行赋权训练,提出了一种自适应非线性训练神经网络的混凝土坝安全监控模型。计算实例表明,该模型计算结果相对误差较小,满足工程精度要求,提高了神经网络的学习质量、收敛效率和泛化能力,减小了神经网络学习中的陷入局限极小值的可能性。该模型可推广应用于面板堆石坝、高边坡等结构安全监控分析中。
Because the prediction model of dam safety monitoring established on standard BP neural network is difficult to meet the requirements on engineering accuracy, a prediction model for concrete dam safety monitoring based on adaptive Unscented Kalman Filter and BP neural network adaptive （AUKF-BP） is proposed after using Unscented Kalman Filter （UKF） filtering theory and introducing adaptive factor into BP neural network for training connecting weight variables. The case calculation shows that the calculation relative error of improved model is smaller, and the results meet engineering accuracy requirements. The model can also improve the learning quality, convergence efficiency and generalization ability of neural network, and meanwhile, reduce the possibility of convergence to local least value. This prediction model can also be applied into the analyses of safety monitoring of face rockfill dams and high slopes.
出处
《水力发电》
北大核心
2013年第4期83-86,共4页
Water Power
基金
国家自然科学基金重点项目(51139001)
国家自然科学基金资助项目(51079046
50909041
50809025
50879024)
关键词
神经网络
UKF滤波
混凝土坝
安全监控
模型
BP neural network
UKF filter
, concrete dam
safety monitoring
model
